MCP Server, CLI & API Reference
Connect any MCP-compatible agent — or use the CLI directly from your terminal. One platform, two interfaces, 60+ tools across every domain in the workspace.
快速开始
托管 MCP 通过 Streamable HTTP 使用 OAuth。请从 Siftable URL 开始;无需复制和粘贴令牌。
1. 添加托管 MCP URL
将 https://siftable.io/api/v1/mcp 添加为远程 MCP 服务器或连接器。请勿使用已弃用的 /sse 端点。
2. 完成 OAuth
在您的客户端中选择 Connect、Authenticate 或 Login,并在 Siftable 授权页面上批准访问。请勿在此配置中添加 Authorization 标头。
{ "mcpServers": { "siftable": { "type": "http", "url": "https://siftable.io/api/v1/mcp" } } }
3. 验证连接
调用 context_current,然后列出项目或任务。当您需要其他操作时,请使用 find_capability。
# Agent calls: project_list() task_list(status: "in_progress") calendar_list_events(startDate: "2026-02-17", endDate: "2026-02-23") # Agent now has your full operational context.
CLI Installation
Siftable CLI 为您提供完整的终端命令界面以及交互式 Copilot。它使用与托管 MCP 相同的记录,并具备 CLI 特定的身份验证和本地功能。二进制文件将安装为 sift(带有 siftable 和 exf 兼容别名)。
1. Install
npm install -g @siftable/cli # npm bun install -g @siftable/cli # bun pnpm add -g @siftable/cli # pnpm # or run without installing npx @siftable/cli <command>
2. Authenticate
The CLI uses device flow authentication — it opens your browser, you approve, and a token is stored locally.
$ sift auth login
Your verification code: ZFMV-SBGJ
Opening browser...
If the browser didn't open, visit: https://siftable.io/app/device?code=ZFMV-SBGJ
Waiting for authorization...
Logged in successfully!
3. Run your first command
$ sift projects list NAME STATUS TASKS Siftable active 12 Marketing Site active 4 Mobile App planning 0 $ sift projects context <id> # Full project context: tasks, signals, notes, members.
Every CLI command supports --json for machine-readable output. Pipe it into jq, feed it to scripts, or let agents parse it directly.
认证
托管 MCP 请使用 OAuth,交互式 CLI 会话请使用设备流,非交互式自动化或本地 MCP 客户端请使用个人访问令牌。
设备流(CLI)
运行 sift auth login。CLI 会打开浏览器,你用 Google 登录并批准设备代码。系统会自动生成 PAT,并存储在 ~/.config/siftable/ 中,无需复制粘贴令牌。
个人访问令牌 (自动化和本地 MCP)
对于非交互式自动化或本地 MCP 客户端,请在 设置 → 集成 → 开发者 / AI 客户端 中创建令牌,并在 Authorization 标头中传递。静态令牌无法使用 OAuth 作用域提升。
Authorization: Bearer sift_pat_your_token_here
令牌的作用域仅限于特定领域(仅限任务、仅限日历、完全访问等)。对于 CI 流水线,请将 SIFT_TOKEN 设置为环境变量。
$ sift auth login $ sift auth status $ sift auth logout
令牌作用域和可用的写入操作取决于您的工作区配置和订阅计划。请检查设置 → 集成 → 开发者 / AI 客户端以及定价,以了解您账户当前的限制。
MCP 连接
Siftable 托管 MCP 使用支持 OAuth 的 Streamable HTTP。端点为:
https://siftable.io/api/v1/mcp
请将此 URL 用于支持 OAuth 的远程客户端,包括 ChatGPT、Codex、Cursor、Claude、Grok 以及兼容的 MCP SDK 实现。托管 MCP 设置指南中包含特定客户端的具体步骤。
任务
| MCP Tool | CLI | Access | Description |
|---|---|---|---|
| task_list | sift tasks list |
READ | List tasks with filters: status (inbox, next_action, in_progress, waiting_for, completed, archived), phase, project, limit. |
| task_get | sift tasks get |
READ | Get a single task with full details: description, due date, priority, project, linked code. |
| task_create | sift tasks create |
WRITE | Create a task. Accepts title, description, priority (do_now, schedule, delegate, someday), project, due date. |
| task_update | sift tasks update |
WRITE | Update task fields: title, description, status, priority, due date, project assignment. |
| task_complete | sift tasks complete |
WRITE | Mark a task as completed. |
| task_delete | sift tasks delete |
DELETE | Permanently delete a task. |
$ sift tasks list --status in_progress TITLE STATUS PRIORITY DUE Ship CLI docs in_progress do_now 2026-02-27 Fix device flow auth in_progress do_now - $ sift tasks create --title "Review PR #312" --priority do_now --json {"id":"abc-123","title":"Review PR #312","status":"inbox","priority":"do_now"}
日历
| MCP Tool | CLI | Access | Description |
|---|---|---|---|
| calendar_list_events | sift calendar list |
READ | List events in a date range. Returns title, start/end times, location, description. Supports limit. |
| calendar_create_event | sift calendar create |
WRITE | Create a calendar event. Requires title, startTime, endTime (ISO 8601). Optional: description, location. |
| calendar_update_event | sift calendar update |
WRITE | Update an existing event's title, times, description, or location. |
| calendar_delete_event | sift calendar delete |
DELETE | Remove an event from the calendar. |
$ sift calendar list --start 2026-02-24 --end 2026-02-28 TITLE START END LOCATION Team standup Feb 25 09:00 Feb 25 09:30 Zoom Product review Feb 26 14:00 Feb 26 15:00 Conf Room B $ sift calendar create --title "Ship CLI v0.3" \ --start 2026-02-27T10:00:00Z --end 2026-02-27T10:30:00Z
项目
| MCP Tool | CLI | Access | Description |
|---|---|---|---|
| project_list | sift projects list |
READ | List projects. Filter by status (planning, active, on_hold, blocked, completed). Include archived. |
| project_get_context | sift projects context |
READ | Full project context: tasks, notes, members, signals. The richest single call for understanding a project. |
| project_create | sift projects create |
WRITE | Create a project with name, summary, status, and emoji. |
| project_update | sift projects update |
WRITE | Update project name, summary, status, or emoji. |
| project_archive | sift projects archive |
WRITE | Archive a completed or inactive project. |
知识库
| MCP Tool | CLI | Access | Description |
|---|---|---|---|
| note_search | sift notes search |
READ | Semantic + full-text search across notes. Filter by project. |
| note_list | sift notes list |
READ | List notes. Filter by type (note, concept, meeting, reference, daily, dataset) and project. |
| note_get | sift notes get |
READ | Get full note content by ID. |
| note_create | sift notes create |
WRITE | Create a note with title, markdown content, type, and optional project. |
| note_update | sift notes update |
WRITE | Update note title, content, or type. |
| note_delete | sift notes delete |
DELETE | Delete a note. |
Datasets
Structured datasets support grounded summaries, grouped analysis, ranking, bucketing, time-series work, import/export, schema changes, and record mutations from the same CLI and MCP-backed platform.
| Capability | CLI | Access | Description |
|---|---|---|---|
| Browse and inspect | sift datasets list, get, query, summarize |
READ | List datasets, inspect schema, query records, and get row-count and sample summaries. |
| Grounded analysis | sift datasets analyze, aggregate, compare |
READ | Generate natural-language insights, grouped metrics, and side-by-side segment comparisons. |
| Ranking and bucketing | sift datasets rank, bucket |
READ | Sort or score records, then bucket numeric or date fields into ranges with metrics per bucket. |
| Time series and plots | sift datasets timeseries, plot |
READ | Compute lag, pct-change, rolling windows, drawdown, or normalize plotting payloads from derived results. |
| Import and export | sift datasets import, export |
WRITE | Bring CSVs in, append to existing datasets, or export filtered results back to CSV. |
| Create and materialize | sift datasets create, materialize |
WRITE | Create datasets directly or turn a derived result into a new scratch dataset. |
| Schema and records | sift datasets schema, add, update-record, delete-record |
WRITE | Modify field definitions, add rows, update rows, and delete rows with explicit commands. |
| Derived workflows | sift datasets join, compute |
READ | Join dataset slices and compute derived fields before materializing or plotting the result. |
$ sift datasets list $ sift datasets summarize <dataset-id> $ sift datasets analyze <dataset-id> --focus-fields BMI,Outcome $ sift datasets aggregate <dataset-id> --group-by Outcome \ --metrics '[{"operation":"count","as":"rows"},{"operation":"avg","field":"BMI","as":"avg_bmi"}]' $ sift datasets compare <dataset-id> --segment-field Outcome \ --metrics '[{"operation":"avg","field":"Glucose","as":"avg_glucose"}]' $ sift datasets rank <dataset-id> --sorts '[{"field":"BMI","direction":"desc"}]' --limit 10 $ sift datasets bucket <dataset-id> --field Age --bucket-count 5 \ --metrics '[{"operation":"count","as":"rows"}]'
人脉
| MCP Tool | CLI | Access | Description |
|---|---|---|---|
| people_search | sift people search |
READ | Search contacts by name. Returns relationship type, company, contact info, interaction history. |
| — | sift people list |
READ | List all contacts. CLI-only convenience wrapper. |
| person_create | sift people create |
WRITE | Create a contact with name, relationship, company, and contact details. |
| person_update | sift people update |
WRITE | Update a contact's details, relationship, or company. |
| person_delete | sift people delete |
DELETE | Delete a contact. |
代码记忆
存储并检索代码库的精选知识点。使用 Git、rg 或你的编辑器在本地检视源码;Siftable 不托管源码索引或搜索。
| MCP Tool | CLI | Access | Description |
|---|---|---|---|
| code_memory_store | sift code memory store |
WRITE | Store a fact. Categories: architecture, integration, convention, entrypoint, gotcha, ownership. Optional: file path, repository. |
| code_memory_search | sift code memory search |
READ | Semantic search over stored code facts. Filter by category or repository. |
| code_memory_list | sift code memory list |
READ | List all stored code memories. Filter by repository. |
| code_memory_delete | sift code memory delete |
DELETE | Delete a stored code memory by ID. |
文档
| MCP Tool | CLI | Access | Description |
|---|---|---|---|
| upload_document | sift documents upload |
WRITE | Upload a PDF, Markdown, or text file into Knowledge. Set by file path or inline content. Auto-detects type. |
Vault
Encrypted secret storage. Store API keys, credentials, OAuth tokens, SSH keys, and sensitive notes. Values are encrypted at rest and audit-logged on read.
| MCP Tool | CLI | Access | Description |
|---|---|---|---|
| vault_create | sift vault create |
WRITE | Store a new encrypted secret. Types: env_var, credential, oauth_token, ssh_key, certificate, note. |
| vault_list | sift vault list |
READ | List vault entries (metadata only — never decrypted values). Filter by type or category. |
| vault_search | sift vault search |
READ | Search vault entries by name, slug, or description. Returns metadata only. |
| vault_update | sift vault update |
WRITE | Update vault entry metadata: name, tags, category, description. |
$ sift vault create --name "Stripe API Key" --json # Interactive prompt for sensitive payload values $ sift vault list NAME TYPE CATEGORY CREATED Stripe API Key env_var payments 2026-02-25 GitHub PAT credential devtools 2026-02-20 $ sift vault materialize run --help # Approved plaintext goes only to the exact local destination. $ sift env pull --help # Pull an approved .sift/environment.yaml bundle.
Workflows
Cross-domain recipes that combine multiple tools. These patterns work identically via MCP or CLI.
Link Code to Tasks
Connect implementation work to the task that motivated it. Agents and teammates can trace why code changed.
task_create(title: "Implement device flow auth", priority: "do_now")
# ... implement the feature ...
task_link_code(taskId: "abc-123", repositoryId: "repo-456",
commitSha: "5cf5f22", filePath: "src/services/deviceAuthService.ts",
notes: "Fixed verification_uri to use /app/device")
task_complete(taskId: "abc-123")
$ sift tasks create --title "Implement device flow auth" --priority do_now $ sift code link abc-123 --repo repo-456 \ --commit 5cf5f22 --file src/services/deviceAuthService.ts $ sift tasks complete abc-123
Project Onboarding
New to a project? Pull the full context in three commands.
# 1. Get the big picture $ sift projects context <id> # 2. See what's in flight $ sift tasks list --project <id> --status in_progress # 3. Check code conventions $ sift code memory search "architecture and conventions"
End-of-Day Review
Summarize what happened today. Works great as an agent prompt or a manual check.
# Agent calls: calendar_list_events(startDate: "2026-02-26", endDate: "2026-02-26") task_list(status: "completed", limit: 20) task_list(status: "in_progress") # Agent now has: today's meetings, completed tasks, and remaining work. # It can draft a standup summary, update project status, or flag blockers.
$ sift calendar list --start 2026-02-26 --end 2026-02-26 $ sift tasks list --status completed --limit 20 --json | jq '.tasks[] | .title' $ sift tasks list --status in_progress
Store a Debug Discovery
Found a gotcha? Store it so your future self (or your agent) doesn't rediscover it the hard way.
$ sift code memory store \ --fact "Device flow verification_uri must use /app/device (SPA path), not /device (marketing homepage)" \ --category gotcha \ --file src/services/deviceAuthService.ts
幂等性
写操作(task_create、note_create、calendar_create_event 等)接受可选的 idempotencyKey 参数。如果你使用相同的 key 重试请求,Siftable 返回原始结果而不是创建重复项。
task_create(
title: "Review PR #247",
priority: "do_now",
idempotencyKey: "agent-run-42-task-pr247"
)
# Safe to retry. Same key = same result.
权限
Token 按领域限定范围。仅日历 token 无法读取任务或人脉。将 token 范围限定到你的 agent 所需的最小权限。
可用范围:
tasks:read/tasks:write— Task listing, CRUD, and completioncalendar:read/calendar:write— Event listing and creationprojects:read/projects:write— Project management and contextknowledge:read/knowledge:write— Notes, search, document uploadpeople:read/people:write— Contact search and CRM updateswork:read/work:write— Agent work queue itemsorg:read— Workspace org metadatamcp:*— All MCP operations (recommended for IDE and agent use)
可用的写入操作取决于你的令牌作用域和当前工作区计划。在生产环境中依赖写入操作之前,请检查你签发的令牌和当前的定价页面。
CLI 命令参考
All 169 commands from @siftable/cli@0.5.29, generated from the oclif manifest so this page cannot drift from sift --help. Every command also accepts the global flags (--json, --token, --api-url, --workspace, --no-input).
所有命令也支持全局标志:--json(原始 JSON 输出)、--token / SIFT_TOKEN、--api-url / SIFT_API_URL(默认为 https://siftable.io)、--workspace / SIFT_WORKSPACE_ID 以及 --no-input(禁用提示)。破坏性命令需使用 --confirm 或 -y。
通用
顶层命令与诊断。
5 个命令
sift capabilities显示 Siftable CLI 功能与就绪状态
sift commands显示适合智能体的命令主题与工作流入口
sift doctor诊断本地 Siftable CLI 配置,不打印敏感信息
sift interactive启动 Siftable 终端副驾驶 (sift interactive) —— 进程内 AI 助手,助你管理任务、工作、日历、项目和人脉。
- Flags
--connected-models— List eligible connected models and exit--connection <value>— Select a Model Connection UUID for a gateway invocation--max-output-tokens <value>— Maximum connected-model output tokens (1-32768)--model <value>— Select an eligible connected model for a gateway invocation--prompt <value>— Invoke the selected connected model once and exit--stream— Consume selected connected-model output incrementally
sift mermaid在终端渲染 Mermaid 图表(流程图、时序图、状态图、类图、ER 图、C4、架构图、思维导图)。支持读取 .mmd 文件或 stdin。
- Arguments
file— Path to a .mmd file (omit to read stdin)- Flags
--ascii— Use ASCII glyphs instead of Unicode box drawing--color <none|truecolor>— Color mode--height <value>— Fit into an exact N-row pane (pads/clips)--max-height <value>— Bound the diagram to N rows (no padding)--max-width <value>— Bound the diagram to N columns (no padding)--overflow <allow|clip|error>— What to do when the diagram exceeds the bounds--unicode— Use Unicode box drawing (default)--width <value>— Fit into an exact N-column pane (pads/clips)
智能体
智能体别名。
6 条命令
sift agents create创建智能体别名
- Flags
--alias <value>— Stable alias slug, e.g. codex--capabilities <value>— Capabilities JSON object--hidden— Hide from normal user-visible lists--name <value>— Display name--operator <value>— Linked daemon/operator ID--permissions <value>— Default permissions JSON object--type <value>— Agent type
sift agents disable禁用智能体别名
- Arguments
aliasrequired — Agent alias or ID
sift agents get获取智能体别名
- Arguments
aliasrequired — Agent alias or ID
sift agents list列出智能体别名
- Flags
--include-disabled— Include disabled aliases
sift agents update更新智能体别名
- Arguments
aliasrequired — Agent alias or ID- Flags
--capabilities <value>— Capabilities JSON object--hidden— Hide from normal user-visible lists--name <value>— Display name--operator <value>— Linked daemon/operator ID--permissions <value>— Default permissions JSON object--status <active|disabled>— Alias status--type <value>— Agent type--visible— Show in normal user-visible lists
sift agents work列出分配给智能体别名的工作
- Arguments
aliasrequired — Agent alias or ID- Flags
--limit <value>— Maximum results--status <value>— Work item status
Ai
5 commands
sift ai connectShow the secure browser flow for adding a non-exportable Model Connection
- Flags
--url <value>— Siftable Model Connections settings URL
sift ai invokeInvoke an eligible connected model (requires ai:invoke and ai:connections:use)
- Flags
--connection <value>— Model Connection UUID returned by sift ai list--max-output-tokens <value>— Maximum output tokens (1-32768)--model <value>— Eligible model ID returned by sift ai list--prompt <value>— Prompt text--stream— Consume and print incremental connected-model output
sift ai listList eligible connected models (requires ai:models:read)
sift ai statusShow non-secret Model Connection status (requires ai:connections:use)
- Arguments
connection— Optional Model Connection UUID
sift ai usageShow connected-model usage totals (requires ai:usage:read)
- Flags
--from <value>— ISO-8601 period start--to <value>— ISO-8601 period end
Approvals
2 commands
sift approvals requestRequest a governed action approval; this command cannot approve or consume it
- Flags
--action <value>— Governed action identifier--destination <value>— Destination binding JSON object--expires-in <value>— Approval lifetime in seconds (30-600)--operation <value>— Operation identifier--purpose <value>— Human-readable non-secret purpose--resource-id <value>— Resource identifier--resource-type <value>— Resource type identifier
sift approvals statusInspect a governed approval requested by this CLI identity
- Arguments
idrequired — Approval ID
身份验证
身份验证命令。
3 条命令
sift auth login登录 Siftable
- Flags
--scope <ai:models:read|ai:invoke|ai:usage:read|ai:connections:use|vault:metadata:read|vault:manage|vault:audit:read>— Incremental AI invocation or Vault scope to request (repeatable; management AI scopes and plaintext reveal are unavailable)
sift auth logout移除已存储的身份验证信息
sift auth status查看身份验证状态
Billing
4 commands
sift billing fallback decideAllow, deny, or always allow personal funding for a quoted workspace operation
- Flags
--decision <allow|deny|always_allow>--monthly-cap-micros <value>— Monthly micro-USD cap for always_allow--quote <value>— Server-issued operation quote ID
sift billing fallback policyRead or update the workspace personal-fallback policy
- Flags
--disabled— Disable and revoke personal fallback--enabled— Enable personal fallback
sift billing fallback revokeRevoke one of your active personal-fallback consents
- Arguments
consentIdrequired
sift billing fallback statusShow your active personal-fallback decisions for a workspace
日历
日历事件。
4 条命令
sift calendar create创建日历事件
- Flags
--description <value>— Event description--end <value>— End time (ISO 8601)--location <value>— Event location--start <value>— Start time (ISO 8601)--title <value>— Event title
sift calendar delete删除日历事件
- Arguments
idrequired — Event ID- Flags
-y, --yes— Skip confirmation
sift calendar list列出日历事件
- Flags
--end <value>— End date (ISO 8601)--limit <value>— Maximum number of results--start <value>— Start date (ISO 8601)
sift calendar update更新日历日程
- Arguments
idrequired — Event ID- Flags
--description <value>— Event description--end <value>— End time (ISO 8601)--location <value>— Event location--start <value>— Start time (ISO 8601)--title <value>— Event title
Capabilities
5 commands
sift capabilities createCreate a reviewed server-brokered Vault capability
- Flags
--adapter <value>— Reviewed static adapter ID--expires-in <value>— Lifetime in seconds (300-2592000)--field <value>— Credential payload field--operation <value>— Comma-separated allowlisted operations--provider <value>— Provider ID--purpose <value>— Non-secret human-readable purpose--vault-entry <value>— Vault entry UUID
sift capabilities describeDescribe safe metadata for one Vault capability
- Arguments
idrequired — Capability metadata ID
sift capabilities executeExecute one typed operation through a Vault capability handle
- Flags
--approval <value>— Governed approval UUID when required--handle <value>— Opaque vcap_ capability handle--idempotency-key <value>— Stable 8-128 character key for safe retries--input <value>— Typed operation input JSON object--operation <value>— Allowlisted operation
sift capabilities listList safe metadata for Vault capability handles
sift capabilities revokeRevoke a Vault capability
- Arguments
idrequired — Capability metadata ID
代码
代码工具。
13 条命令
sift code blameGit blame 文件
- Arguments
filerequired — Relative file path- Flags
--root <value>— Repository root path
sift code memory confirm确认代码记忆候选并附上可审计原因
- Arguments
idrequired — Memory ID- Flags
--reason <value>— Audit reason
sift code memory delete删除存储的代码库记忆
- Arguments
idrequired — Memory ID- Flags
-y, --yes— Skip confirmation
sift code memory detect-stale检测可能已过时的代码记忆,并将其标记为 needs_review (从不删除)
- Flags
--limit <value>— Maximum memories to scan--repo <value>— Repository ID
sift code memory edit原位编辑代码记忆;先前内容保留在审计日志中
- Arguments
idrequired — Memory ID- Flags
--confidence <value>— Confidence 0-1--fact <value>— Updated fact--reason <value>— Audit reason
sift code memory events列出代码记忆的谱系事件 (仅追加)
- Arguments
idrequired — Memory ID- Flags
--limit <value>— Maximum number of events
sift code memory get显示一条代码记忆的来源、证据和生命周期
- Arguments
idrequired — Memory ID
sift code memory list列出存储的代码库记忆
- Flags
--include-inactive— Include rejected and superseded lineage memories--lifecycle <unreviewed|confirmed|rejected|superseded|needs_review>— Filter by lifecycle state--limit <value>— Maximum number of results--repo <value>— Repository ID
sift code memory needs-review因可能过时或存在矛盾,将代码记忆标记为 needs_review (从不删除)
- Arguments
idrequired — Memory ID- Flags
--reason <value>— Audit reason--signal <value>— Staleness signal (repeatable)
sift code memory reject拒绝一条代码记忆并提供可审计的原因 (软拒绝,从不删除)
- Arguments
idrequired — Memory ID- Flags
--reason <value>— Audit reason
sift code memory search搜索已存储的代码库信息
- Arguments
queryrequired — Search query- Flags
--category <architecture|integration|convention|entrypoint|gotcha|ownership>— Filter by category--include-inactive— Include rejected and superseded lineage memories--lifecycle <unreviewed|confirmed|rejected|superseded|needs_review>— Filter by lifecycle state--limit <value>— Maximum number of results--repo <value>— Repository ID
sift code memory store存储代码库信息
- Flags
--agent <value>— Authoring agent (marks memory as agent-authored)--category <architecture|integration|convention|entrypoint|gotcha|ownership>— Fact category--commit <value>— Source commit SHA--confidence <value>— Confidence 0-1--evidence-chunk <value>— Historical indexed chunk ID supporting the fact--fact <value>— Fact to store (1-2 sentences)--file <value>— Related file path--path <value>— Additional related path (repeatable)--reason <value>— Capture reason (required with --agent)--repo <value>— Repository ID--work-item <value>— Related agent work item ID
sift code memory supersede用新事实取代一条代码记忆;旧记忆作为谱系保留
- Arguments
idrequired — Memory ID to supersede- Flags
--category <architecture|integration|convention|entrypoint|gotcha|ownership>— Fact category (defaults to previous)--commit <value>— Source commit SHA--confidence <value>— Confidence 0-1--evidence-chunk <value>— Evidence chunk ID (repeatable)--fact <value>— Replacement fact--file <value>— Related file path--path <value>— Additional related path (repeatable)--reason <value>— Audit reason
Codex
Codex 自动化助手。
1 条命令
sift codex daily-review collect收集只读 Siftable 与本地 git 上下文,用于 Codex 每日工作回顾
- Flags
--calendar-days <value>— Calendar lookahead days--limit <value>— Maximum records per source--skip-git— Skip local git summary
上下文
1 个命令
sift context current说明当前的主体、所有者、项目和本地检出上下文
- Flags
--workspace-root <value>— Absolute local checkout root to describe (never sent as a workspace tenant ID)
Crm
6 个命令
sift crm import apply仅应用现有 CRM 导入计划中已批准的操作;可安全重放
- Arguments
planIdrequired — Durable CRM import plan UUID- Flags
--yes— Confirm canonical CRM writes without prompting
sift crm import approve批准现有 CRM 导入计划中的全部或部分待处理操作
- Arguments
planIdrequired — Durable CRM import plan UUID- Flags
--operation-id <value>— Stable operation ID to approve; repeat to approve a subset--reason <value>— Human approval reason recorded on the plan--yes— Confirm the approval without prompting
sift crm import get根据稳定的计划 ID 恢复并检查持久化 CRM 导入计划
- Arguments
planIdrequired — Durable CRM import plan UUID
sift crm import plan从 CSV、TSV、XLS 或 XLSX 文件创建或复用持久化 CRM 导入计划,此过程不写入正式 CRM 记录
- Arguments
filerequired — Local CSV, TSV, XLS, or XLSX source (8 MiB maximum)- Flags
--mapping-mode <manual|workspace_auto>— Manual mapping or approved workspace mapping profile--provider <generic|salesforce>— Source provider mapping
sift crm organizations convert说明 CRM 组织与工作空间之间的管控边界
- Arguments
idrequired — CRM organization ID
sift crm organizations list仅列出 CRM 组织;不包括协作工作空间
- Flags
--limit <value>— Maximum number of CRM results--search <value>— Optional fuzzy search query
数据集
结构化数据集
43 个命令
sift datasets add向数据集添加记录
- Arguments
idrequired — Dataset ID- Flags
--idempotency-key <value>— Caller-owned stable key for replaying this exact dataset mutation--record <value>— Record as JSON object, e.g. '{"name":"Alice","age":"30"}'--records <value>— Multiple records as JSON array
sift datasets aggregate使用分组指标聚合数据集记录(计数、平均值、总和、最小值、最大值、中位数、标准差、百分位数、比例)
- Arguments
idrequired — Dataset ID- Flags
--filters <value>— JSON array of filters--group-by <value>— Comma-separated field names to group by--having <value>— JSON array of having clauses [{metric, operator, value}]--limit <value>— Max rows--metrics <value>— JSON array of metrics [{operation, field, as}]--sorts <value>— JSON array of sorts
sift datasets analyze生成基于事实的数据集自然语言洞察
- Arguments
idrequired — Dataset ID- Flags
--filters <value>— JSON array of filters--focus-fields <value>— Comma-separated field names to focus analysis on--max-insights <value>— Max insights to generate--mode <descriptive|operational>— Analysis mode--signal-limit <value>— Max decision signals to return
sift datasets apply-diff应用已保存的数据集差异计划
- Arguments
planrequired — Path to a local diff plan or persisted diff plan ID- Flags
--yes— Confirm applying the saved diff plan without prompting
sift datasets archive归档数据集,保留物理表
- Arguments
idrequired — Dataset ID- Flags
-y, --yes— Confirm dataset archival without prompting
sift datasets bucket对数值或日期字段分桶并计算聚合指标
- Arguments
idrequired — Dataset ID- Flags
--boundaries <value>— Comma-separated boundary values (omit for auto-bucketing)--bucket-count <value>— Number of auto-buckets (default: 5)--field <value>— Field to bucket--filters <value>— JSON array of filters--metrics <value>— JSON array of metrics
sift datasets cleanup规划或执行临时数据集清理
- Flags
--dry-run— Return a deterministic cleanup plan without deleting datasets--lifecycle <value>— Lifecycle kind to clean, e.g. scratch, benchmark, research-run--limit <value>— Maximum lifecycle datasets to inspect--now <value>— Deterministic timestamp for tests and scheduled cleanup--older-than <value>— Only include datasets older than this duration, e.g. 12h, 7d--orphaned— Include stale dataset notes that no longer have a backing dataset row--tag <value>— Lifecycle tag to clean, e.g. benchmark-y, --yes— Confirm deletion when applying cleanup with --no-dry-run
sift datasets compare横向对比分类字段各分段的指标
- Arguments
idrequired — Dataset ID- Flags
--filters <value>— JSON array of filters--limit <value>— Max segment values to compare--metrics <value>— JSON array of metrics--segment-field <value>— Categorical field to segment by--segment-values <value>— Comma-separated segment values (auto-discovers if omitted)
sift datasets compute计算数据集或先前结果的派生字段
- Arguments
id— Dataset ID- Flags
--computed-fields <value>— JSON array of computed fields, e.g. '[{"as":"spread","expression":"right.Close-left.Close"}]'--filters <value>— JSON array of filters--limit <value>— Maximum rows--order-by <value>— JSON array of order clauses--select <value>— Comma-separated fields to include--sorts <value>— JSON array of output sorts--source-result <value>— Inline JSON for a prior derived result--source-result-file <value>— Path to a JSON file containing a prior derived result
sift datasets contract显示智能体可读的数据集模式与能力契约
- Arguments
idrequired — Dataset ID- Flags
--resolve <value>— Comma-separated semantic field references to resolve--template <value>— Validate contract against a built-in template
sift datasets create创建数据集
- Flags
--description <value>— Dataset description--fields <value>— Field definitions as JSON array, e.g. '[{"name":"age","type":"number"}]'--idempotency-key <value>— Caller-owned stable key for replaying this exact dataset mutation--lifecycle <value>— Lifecycle kind for generated datasets, e.g. scratch, benchmark, research-run--metadata <value>— Dataset metadata as JSON object--note-id <value>— Link to an existing note--run-id <value>— Lifecycle run identifier--scratch— Shortcut for --lifecycle scratch --tags scratch--tags <value>— Comma-separated lifecycle tags--title <value>— Dataset title--ttl <value>— Lifecycle TTL duration, e.g. 12h, 7d, 30d
sift datasets dedupe按键值查找重复记录,不修改数据
- Arguments
idrequired — Dataset ID- Flags
--key <value>— Field name used to group duplicates--limit <value>— Maximum records to scan in one bounded pass
sift datasets delete永久删除数据集及其物理表
- Arguments
idrequired — Dataset ID- Flags
-y, --yes— Confirm dataset deletion without prompting
sift datasets delete-record删除数据集中的记录
- Arguments
idrequired — Dataset IDrecord-idrequired — Record ID- Flags
--idempotency-key <value>— Caller-owned stable key for replaying this exact dataset mutation-y, --yes— Skip confirmation
sift datasets diff预览 CSV、JSON 或 JSONL 文件中的数据集行更改
- Arguments
idrequired — Dataset ID- Flags
--batch-size <value>— Records per backend batch--from-file <value>— Path to CSV, JSON, or JSONL rows to compare--persist— Persist the diff plan in Siftable for later review/apply--save-plan <value>— Write an applyable diff plan JSON file--template <sources|people|events|claims|evidence_sources|evidence_source_fragments|evidence_claims|evidence_people|evidence_organizations|evidence_places|evidence_artifacts|evidence_events|evidence_relationships|evidence_contradictions>— Built-in template name--upsert-by <value>— Field name used to match existing rows
sift datasets diff-plans list列出已保存的数据集差异计划
- Flags
--dataset-id <value>— Filter by dataset ID--limit <value>— Maximum plans to return--status <draft|validated|applied|rejected|expired>— Filter by plan status
sift datasets diff-plans show查看已保存的数据集差异计划
- Arguments
idrequired — Diff plan ID
sift datasets export导出数据集记录为 CSV、JSON、JSONL 或 Markdown
- Arguments
idrequired — Dataset ID- Flags
--filters <value>— JSON array of filters--format <csv|json|jsonl|markdown>— Export format--limit <value>— Max rows to export-o, --output <value>— Output file path (writes to stdout if omitted)--sorts <value>— JSON array of sorts
sift datasets facets显示数据集字段的分面摘要
- Arguments
idrequired — Dataset ID- Flags
--fields <value>— Comma-separated field names to facet--limit <value>— Maximum values per facet
sift datasets formula-plan计算公式字段并预览数据集更新
- Arguments
idrequired — Dataset ID- Flags
--computed-fields <value>— JSON array of computed fields, e.g. '[{"as":"score","expression":"confidence * reliability"}]'--filters <value>— JSON array of filters for compute source--limit <value>— Maximum rows to compute and plan--order-by <value>— JSON array of order clauses--save-plan <value>— Write an applyable diff plan JSON file--select <value>— Comma-separated fields to include in compute source--sorts <value>— JSON array of output sorts--target-fields <value>— Comma-separated computed field names to write; defaults to every computed field alias--template <sources|people|events|claims>— Built-in template name for validation--upsert-by <value>— Field used to match rows for update
sift datasets get获取数据集详情和结构
- Arguments
idrequired — Dataset ID
sift datasets impact解释数据集公式、图表、视图、质量及物化影响
- Arguments
idrequired — Dataset ID- Flags
--from-plan <value>— Persisted diff plan ID to inspect--operation <value>— Committed dataset operation ID to inspect
sift datasets import将 CSV、JSON 或 JSONL 数据行导入新数据集或现有数据集
- Arguments
filerequired — Path to CSV, JSON, or JSONL file- Flags
--batch-size <value>— Records per backend batch--dataset-id <value>— Import into existing dataset instead of creating a new one--description <value>— Dataset description--dry-run— Validate and plan the import without writing--lifecycle <value>— Lifecycle kind for generated datasets, e.g. scratch, benchmark, research-run--metadata <value>— Dataset metadata as JSON object when creating a new dataset--run-id <value>— Lifecycle run identifier--scratch— Shortcut for --lifecycle scratch --tags scratch--tags <value>— Comma-separated lifecycle tags--template <sources|people|events|claims|evidence_sources|evidence_source_fragments|evidence_claims|evidence_people|evidence_organizations|evidence_places|evidence_artifacts|evidence_events|evidence_relationships|evidence_contradictions>— Built-in template name--title <value>— Dataset title (defaults to filename)--ttl <value>— Lifecycle TTL duration, e.g. 12h, 7d, 30d--upsert-by <value>— Field name used to update matching rows instead of creating duplicates--yes— Confirm mutating imports without prompting
sift datasets inspect-file检查本地 CSV、TSV、XLS 或 XLSX 文件的结构,此操作不会创建或更改数据集
- Arguments
filerequired — Path to a local CSV, TSV, XLS, or XLSX file- Flags
--delimiter <value>— Single-byte delimiter override for CSV/TSV--format <csv|tsv|xls|xlsx>— Format override when the extension is absent or wrong--header-row <value>— Zero-based header row index within the sheet grid--max-bytes <value>— Refuse files larger than this many bytes--max-cols <value>— Maximum columns returned per sheet--max-rows <value>— Maximum data rows returned per sheet--max-sheets <value>— Maximum sheets returned--sheet <value>— Inspect only this sheet; the full sheet inventory is still reported
sift datasets join使用 left.Close 和 right.Close 等别名作用域字段自连接数据集
- Arguments
idrequired — Dataset ID- Flags
--join-keys <value>— JSON array of join keys, e.g. '[{"leftField":"Date","rightField":"Date"}]'--join-type <inner|left|right>— Join type--left-alias <value>— Left alias--left-filters <value>— JSON array of left-side filters--limit <value>— Maximum joined rows--right-alias <value>— Right alias--right-filters <value>— JSON array of right-side filters--select <value>— Comma-separated alias-scoped fields to return--sorts <value>— JSON array of sorts
sift datasets list列出数据集
- Flags
--limit <value>— Maximum number of results
sift datasets lookup通过键/值精确匹配查找数据集记录
- Arguments
idrequired — Dataset ID- Flags
--key <value>— Field name to match--limit <value>— Maximum matching records--value <value>— Exact value to match
sift datasets materialize将派生结果物化到新的临时数据集中
- Flags
--description <value>— Dataset description--idempotency-key <value>— Caller-owned stable key for replaying this exact dataset mutation--source-result <value>— Inline JSON for a derived result--source-result-file <value>— Path to a JSON file containing a derived result--title <value>— Title of the new dataset
sift datasets pivot从分组数据集指标创建透视摘要
- Arguments
idrequired — Dataset ID- Flags
--cols <value>— Column field--filters <value>— JSON array of filters--limit <value>— Maximum grouped cells to request--metrics <value>— JSON metrics array; defaults to count--rows <value>— Row field
sift datasets plot验证并规范化来自派生结果的轻量级绘图负载
- Flags
--chart-type <line|bar|scatter>— Chart type--series-field <value>— Optional series field--source-result <value>— Inline JSON for a derived result--source-result-file <value>— Path to a JSON file containing a derived result--x-field <value>— X-axis field--y-fields <value>— Comma-separated Y-axis fields
sift datasets profile显示数据集的有界概况信息
- Arguments
idrequired — Dataset ID- Flags
--sample-limit <value>— Number of sample rows to include
sift datasets quality通过结构化的缺失值指标和重复值观察,检查数据集质量
- Arguments
idrequired — Dataset ID- Flags
--fields <value>— Comma-separated field names to check--repeat-threshold <value>— Minimum occurrences for a repeated value to be reported
sift datasets query查询数据集记录
- Arguments
idrequired — Dataset ID- Flags
--cursor <value>— Pagination cursor from previous query--filters <value>— Filter conditions as JSON array, e.g. '[{"field":"status","value":"active"}]'--include-deleted— Include soft-deleted records--limit <value>— Maximum number of records--sorts <value>— Sort spec as JSON array, e.g. '[{"field":"name","direction":"asc"}]'
sift datasets rank按排序规则或加权数值公式对数据集记录排名
- Arguments
idrequired — Dataset ID- Flags
--filters <value>— JSON array of filters--formula <value>— JSON formula object {weights: [{field, weight}]}--limit <value>— Max rows--sorts <value>— JSON array of sorts
sift datasets reconcile按键值对比两个数据集,不修改原始数据
- Arguments
leftrequired — Left dataset IDrightrequired — Right dataset ID- Flags
--left-key <value>— Left dataset key field--limit <value>— Maximum rows to scan from each dataset--right-key <value>— Right dataset key field; defaults to --left-key
sift datasets schema修改数据集模式(添加、更新或删除字段)
- Arguments
idrequired — Dataset ID- Flags
--field <value>— Field definition as JSON, e.g. '{"name":"email","type":"text"}'--field-id <value>— Field ID (required for update/delete)--idempotency-key <value>— Caller-owned stable key for replaying this exact dataset mutation--operation <add_field|update_field|delete_field>— Schema operation
sift datasets search在选定的文本字段中搜索数据集记录
- Arguments
idrequired — Dataset IDqueryrequired — Search text- Flags
--fields <value>— Comma-separated fields to search; defaults to profile columns--filters <value>— JSON array of base filters applied to every field search--limit <value>— Maximum merged records--per-field-limit <value>— Maximum records to request per searched field
sift datasets summarize获取数据集摘要(行数、字段、示例行)
- Arguments
idrequired — Dataset ID
sift datasets templates list列出内置数据集模板
sift datasets templates show查看内置数据集模板模式
- Arguments
templaterequired — Template name
sift datasets timeseries分析数据集时间序列(支持滞后、涨跌幅、滚动窗口、回撤、波动率及相关性)
- Arguments
idrequired — Dataset ID- Flags
--date-field <value>— Date field name--filters <value>— JSON array of filters--limit <value>— Maximum output rows--metrics <value>— JSON array of metric definitions--order-direction <asc|desc>— Time ordering--pivot— Emit explicit pivoted output--segment-field <value>— Optional segment field--segment-values <value>— Comma-separated segment values--transforms <value>— JSON array of transform definitions
sift datasets update-record更新数据集记录
- Arguments
idrequired — Dataset IDrecord-idrequired — Record ID- Flags
--fields <value>— Field updates as JSON object, e.g. '{"status":"done"}'--idempotency-key <value>— Caller-owned stable key for replaying this exact dataset mutation
sift datasets validate使用内置模板验证数据集
- Arguments
idrequired — Dataset ID- Flags
--template <sources|people|events|claims|evidence_sources|evidence_source_fragments|evidence_claims|evidence_people|evidence_organizations|evidence_places|evidence_artifacts|evidence_events|evidence_relationships|evidence_contradictions>— Built-in template name
文档
文档上传。
1 条命令
sift documents upload将文档 (PDF、Markdown 或文本) 上传为笔记
- Arguments
filerequired — Path to file- Flags
--project <value>— Project ID--title <value>— Note title (defaults to filename)--type <note|concept|meeting|reference|daily|dataset>— Note type
环境
声明式环境合约与受管控的 Vault 配置包物化。
11 个命令
sift env check验证环境合约,供本地或 CI 使用
- Flags
--checkout-root <value>— Exact Git checkout root containing .sift/environment.yaml--ci— Fail non-zero when contract, example, Git, or checkout safety findings exist
sift env diff显示不含具体值的环境合约偏差
- Flags
--checkout-root <value>— Exact Git checkout root containing .sift/environment.yaml
sift env endpoint enroll注册公共端点身份,但不授予 Vault 访问权限
- Flags
--dedicated— Endpoint is a dedicated runner--endpoint-class <interactive_desktop|remote_vm|noninteractive_runner|shared_host|ci_runner|production_host|multi_tenant>— Explicit endpoint policy class--kill-switch— Runner has an enforced kill switch--owner <value>— personal or workspace:<uuid>--storage-tier <hardware_host_bound|systemd_managed|convenience_keyring|none>— Actual endpoint key storage tier; convenience_keyring is not hardware-backed
sift env example生成一个确定性的、仅包含名称和占位符的示例
- Flags
--checkout-root <value>— Exact Git checkout root containing .sift/environment.yaml--write— Write the deterministic output to .env.example instead of displaying it
sift env import apply原子化应用一份确切的已审查 dotenv 导入计划
- Flags
--checkout-root <value>— Checkout used only to reject Sift-managed source files--owner <value>— Exact Vault owner: personal or workspace:<tenant-id>--plan-file <value>— Reviewed value-free plan artifact--source <value>— Same explicit dotenv source used to create the plan
sift env import plan为单个明确指定的 dotenv 源创建一份不含值、所有者范围的计划
- Flags
--checkout-root <value>— Checkout used only to reject Sift-managed source files--decision <value>— Explicit per-index decision: INDEX:keep-existing|overwrite|skip|rename=NAME--owner <value>— Exact Vault owner: personal or workspace:<tenant-id>--plan-file <value>— New local value-free plan artifact to create--project <value>— Exact owner-scoped project UUID--source <value>— Explicit dotenv source file
sift env pull将一份确切的、受管控的 Vault 配置包拉取到由 Sift 管理的开发环境文件中
- Flags
--approval-timeout <value>--checkout-root <value>— Checkout containing .sift/environment.yaml--destination <value>— Exact relative destination when the manifest declares more than one bundle--endpoint-class <development|production|ci|shared_host|policy_ineligible>— Explicit endpoint policy class; managed pulls accept development only--owner <value>— Exact Vault owner: personal or workspace:<tenant-id>--purpose <value>--replace— Replace a changed Sift-managed file using its exact observed precondition
sift env push apply原子化应用一份确切的已审查 dotenv 导入计划
- Flags
--checkout-root <value>— Checkout used only to reject Sift-managed source files--owner <value>— Exact Vault owner: personal or workspace:<tenant-id>--plan-file <value>— Reviewed value-free plan artifact--source <value>— Same explicit dotenv source used to create the plan
sift env push plan为单个明确指定的 dotenv 源创建一份不含值、所有者范围的计划
- Flags
--checkout-root <value>— Checkout used only to reject Sift-managed source files--decision <value>— Explicit per-index decision: INDEX:keep-existing|overwrite|skip|rename=NAME--owner <value>— Exact Vault owner: personal or workspace:<tenant-id>--plan-file <value>— New local value-free plan artifact to create--project <value>— Exact owner-scoped project UUID--source <value>— Explicit dotenv source file
sift env rm删除一个指定的、由 Sift 管理的本地环境文件
- Flags
--checkout-root <value>— Exact Git checkout root containing .sift/environment.yaml--destination <value>— Exact relative destination when the manifest declares more than one bundle--yes— Confirm removal of the exact recognized managed file
sift env status报告不含具体值的本地环境状态
- Flags
--checkout-root <value>— Exact Git checkout root containing .sift/environment.yaml
事件
基于时间轴事实的研究事件。
3 条命令
sift events attach-person为现有研究事件关联人脉参与者
- Arguments
eventrequired — Existing temporal fact IDpersonrequired — Person UUID to attach- Flags
--role <value>— Participant role--yes— Confirm participant attachment without prompting
sift events create创建包含参与者的研究事件时间轴事实
- Flags
--body <value>— Event notes/body--confidence <low|medium|high>— Confidence level--entity <value>— Participant/entity as type:uuid or type:uuid:role; repeatable--org <value>— Organization UUID participant; repeatable--person <value>— Person UUID participant; repeatable--precision <millisecond|minute|hour|day|month|year|decade|century|millennium|mega_year|era>— Temporal precision--source <value>— Source entity as type:uuid or type:uuid:role; repeatable--source-label <value>— Source/provenance label--source-note <value>— Source/provenance note--source-url <value>— Source/provenance URL--timestamp <value>— ISO timestamp--title <value>— Event title--visibility <org_public|private|restricted>— Timeline visibility--year <value>— Historical year CE--year-end <value>— Historical end year CE
sift events list列出研究事件时间线事实
- Flags
--cursor <value>— Pagination cursor--end <value>— End boundary--entity <value>— Filter by entity ref type:uuid--limit <value>— Maximum events--order <asc|desc>— Sort order--person <value>— Filter by person UUID--q <value>— Text search query--start <value>— Start boundary
证据
证据图谱配置与证明工作流编排。
11 个命令
sift evidence diff apply应用已审核的证据图谱差异计划
- Arguments
idrequired — Persisted diff plan ID- Flags
--yes— Confirm applying the reviewed diff plan without prompting
sift evidence diff impact解释已持久化差异计划对证据图谱的影响
- Arguments
idrequired — Persisted diff plan ID, or local when using --from-file- Flags
--from-file <value>— Local diff plan JSON file to explain without API access
sift evidence diff list列出已持久化的证据图谱差异计划
- Flags
--dataset-id <value>— Filter by evidence dataset ID--limit <value>— Maximum plans to return--project <value>— Filter locally by Evidence Graph project ID when present on plans--status <draft|validated|applied|rejected|expired>— Filter by plan status
sift evidence diff show显示证据图谱差异计划(含领域感知摘要)
- Arguments
idrequired — Persisted diff plan ID
sift evidence extract创建证据图谱候选提取的智能体任务(不执行应用)
- Flags
--agent <value>— Assigned agent alias--context <value>— Additional input context JSON object--context-file <value>— Additional input context JSON file--dry-run— Preview work item payload without writing--no-apply— Keep extraction in proposed/diff-first mode--pack <company-origin|family-history|investigation|compliance-evidence|account-history|codebase-history>— Evidence workflow pack--project <value>— Project ID--source-dataset <value>— Evidence sources dataset ID--targets <value>— Comma-separated extraction targets--yes— Confirm work item creation without prompting
sift evidence init创建证据图谱项目及基于数据集的工作表
- Arguments
namerequired — Evidence Graph project name- Flags
--dry-run— Preview project/dataset creation without writing--pack <company-origin|family-history|investigation|compliance-evidence|account-history|codebase-history>— Evidence workflow pack--yes— Confirm creation without prompting
sift evidence plan写入可信状态前规划 Evidence Graph 工作流
- Arguments
goalrequired — Evidence Graph goal- Flags
--pack <company-origin|family-history|investigation|compliance-evidence|account-history|codebase-history>— Evidence workflow pack--project <value>— Existing project ID--source-dataset <value>— Existing evidence sources dataset ID
sift evidence project试运行 Evidence Graph 时间线与关系投影
- Flags
--dry-run— Preview projection without writing--from-file <value>— Local diff plan JSON file to project from without API access--from-plan <value>— Persisted diff plan ID to project from--pack <company-origin|family-history|investigation|compliance-evidence|account-history|codebase-history>— Evidence workflow pack--project <value>— Evidence Graph project ID
sift evidence proof report根据数据集证据包生成 Evidence Graph 证明报告
- Flags
--format <json|markdown>— Report format--from-file <value>— Evidence packet JSON file to report on--project <value>— Evidence Graph project ID for report metadata
sift evidence sources import将 Evidence Graph 源账本行导入数据集源表
- Arguments
filerequired — Path to CSV, JSON, or JSONL source ledger rows- Flags
--batch-size <value>— Records per backend batch--dataset-id <value>— Evidence sources dataset ID--dry-run— Validate and plan source import without writing--upsert-by <value>— Field name used to update matching source rows--yes— Confirm mutating imports without prompting
sift evidence verify验证 Evidence Graph 溯源、审核、投影及引用不变性
- Flags
--from-file <value>— Evidence packet JSON file to verify--project <value>— Evidence Graph project ID for report metadata
Grants
4 commands
sift grants adaptersList reviewed local execution adapters and honest containment tiers
sift grants requestRequest a human-approved grant for a pre-registered trusted local runner
- Flags
--adapter <value>--audience <value>--credential-field <value>--cwd <value>--executable <value>— Resolved reviewed executable path--executable-digest <value>--issuer <value>--operation <value>--purpose <value>--runner-fingerprint <value>--runner-public-key <value>— PEM public-key file from the trusted local runner--scope <value>— Provider scope JSON with string values--vault-entry <value>
sift grants runRequest approval, redeem in memory, and run exactly one reviewed child process
- Flags
--adapter <github_gh|github_publish_pr|terraform_apply>--approval-timeout <value>--audience <value>--body-file <value>— PR body file for github_publish_pr--credential-field <value>--cwd <value>--issuer <value>--operation <value>--purpose <value>--scope <value>--vault-entry <value>
sift grants statusInspect safe status for an ephemeral local execution grant
- Arguments
idrequired
图谱
实体图搜索与邻域。
5 个命令
sift graph between解释两个实体间的有界图路径
- Arguments
sourcerequired — Source entity reference as type:uuidtargetrequired — Target entity reference as type:uuid- Flags
--depth <value>— Maximum path depth, backend clamps to 1-5--frontier-limit <value>— Maximum links to inspect per path expansion, backend clamps to 1-1000
sift graph explain解释两个实体间的有界图路径
- Arguments
sourcerequired — Source entity reference as type:uuidtargetrequired — Target entity reference as type:uuid- Flags
--depth <value>— Maximum path depth, backend clamps to 1-5--frontier-limit <value>— Maximum links to inspect per path expansion, backend clamps to 1-1000
sift graph neighbors显示实体的本地图邻居
- Arguments
entityrequired — Entity reference as type:uuid- Flags
--depth <value>— Graph depth, backend clamps to 1-3--limit <value>— Maximum graph items, backend clamps to 1-200
sift graph preview预览单个图谱实体
- Arguments
entityrequired — Entity reference as type:uuid
sift graph search搜索可链接的实体用于图谱操作
- Arguments
queryrequired — Search query- Flags
--limit <value>— Maximum results--types <value>— Comma-separated entity types
笔记
知识库笔记。
7 个命令
sift notes bulk-delete预览或批量删除笔记
- Flags
--archived— Filter by archived state--confirm— Execute deletion instead of preview--ids <value>— Comma-separated note IDs--title-contains <value>— Title substring filter--title-equals <value>— Exact title filter--title-starts-with <value>— Title prefix filter--type <note|concept|meeting|reference|daily|dataset>
sift notes create创建笔记
- Flags
--content <value>— Note content (markdown)--metadata <value>— Note metadata as JSON--metadata-file <value>— Read note metadata JSON from a file--project <value>— Project ID--title <value>— Note title--type <note|concept|meeting|reference|daily|dataset>— Note type
sift notes delete删除笔记
- Arguments
idrequired — Note ID- Flags
-y, --yes— Skip confirmation
sift notes get获取笔记完整内容
- Arguments
idrequired — Note ID
sift notes list列出笔记
- Flags
--archived— Filter by archived state--limit <value>— Maximum number of results--project <value>— Filter by project ID--title-contains <value>— Title substring filter--title-equals <value>— Exact title filter--title-starts-with <value>— Title prefix filter--type <note|concept|meeting|reference|daily|dataset>— Filter by note type
sift notes search搜索笔记
- Arguments
queryrequired — Search query- Flags
--limit <value>— Maximum number of results--project <value>— Filter by project ID
sift notes update更新笔记
- Arguments
idrequired — Note ID- Flags
--content <value>— Note content (markdown)--metadata <value>— Replace note metadata with this JSON object--metadata-file <value>— Read replacement note metadata JSON from a file--title <value>— Note title--type <note|concept|meeting|reference|daily|dataset>— Note type
组织
组织与公司。
5 个命令
sift organizations bulk-delete预览或批量删除组织
- Flags
--confirm— Execute deletion instead of preview--contains <value>— Name substring filter--equals <value>— Exact name filter--ids <value>— Comma-separated organization IDs--relationship <value>— Filter by relationship status--starts-with <value>— Name prefix filter--type <value>— Filter by organization type
sift organizations create创建组织
- Flags
--domain <value>— Domain (e.g. acme.com)--industry <value>— Industry--linkedin-url <value>— LinkedIn page URL--location <value>— Location--name <value>— Organization name--notes <value>— Notes--relationship-status <value>— Relationship status (e.g. prospect, customer, partner, vendor)--type <value>— Organization type (e.g. company, nonprofit, government, school)--website <value>— Website URL
sift organizations delete删除组织
- Arguments
idrequired — Organization ID- Flags
-y, --yes— Skip confirmation
sift organizations search搜索组织
- Arguments
query— Optional fuzzy search query- Flags
--contains <value>— Name substring filter--equals <value>— Exact name filter--limit <value>— Maximum number of results--relationship <value>— Filter by relationship status--starts-with <value>— Name prefix filter--type <value>— Filter by organization type
sift organizations update更新组织
- Arguments
idrequired — Organization ID- Flags
--domain <value>— Domain (e.g. acme.com)--industry <value>— Industry--linkedin-url <value>— LinkedIn page URL--location <value>— Location--name <value>— Organization name--notes <value>— Notes--relationship-status <value>— Relationship status--type <value>— Organization type--website <value>— Website URL
人脉
人脉与联系人。
15 条命令
sift people bulk-delete预览或批量删除人脉
- Flags
--confirm— Execute deletion instead of preview--contains <value>— Name substring filter--equals <value>— Exact name filter--has-no-email— Only contacts without an email--ids <value>— Comma-separated person IDs--relationship <value>— Filter by relationshipToUser--starts-with <value>— Name prefix filter
sift people context add添加一条与所有者相关的个人或工作人脉关系
- Arguments
personIdrequired — Person ID- Flags
--basis <value>— Why this relationship is recorded--confidence <value>— Evidence confidence from 0 to 1--context <personal|work>— Relationship context--primary— Make this the primary relationship in its context--relationship <value>— Relationship type, e.g. friend, client, mentor
sift people context list列出当前所有者与某人的人脉关系背景
- Arguments
personIdrequired — Person ID
sift people context remove删除一条与所有者相关的人脉关系背景
- Arguments
personIdrequired — Person IDrelationshipIdrequired — Context relationship ID- Flags
--dry-run— Show what would be removed without writing-y, --yes— Remove without prompting
sift people create创建人脉
- Flags
--birth-year <value>— Birth year--birthday <value>— Birthday (YYYY-MM-DD)--company <value>— Company name (auto-links to organization if exists)--context <personal|work>— How you know this person; repeat for both contexts--email <value>— Email address--estimated-age <value>— Estimated age--job-title <value>— Job title--linkedin-url <value>— LinkedIn profile URL--location <value>— Location--mbti <value>— MBTI type (e.g. INTJ, ENFP)--name <value>— Full name--notes <value>— Notes about this person--phone <value>— Phone number--relationship <value>— Relationship to user (e.g. friend, colleague, client, mentor)--website <value>— Personal website
sift people delete删除人脉
- Arguments
idrequired — Person ID- Flags
-y, --yes— Skip confirmation
sift people enrich以可安全重放的执行方式,搜集关于某人的公开信息
- Arguments
idrequired — Person ID- Flags
--dataset <value>— Optional dataset ID for a reviewable reconciliation plan--dataset-upsert-by <value>— Preferred dataset identity field--idempotency-key <value>— Stable 8-128 character key for exact retries--mode <standard|ultra>— Research depth
sift people get获取包含特征和关系的个人档案
- Arguments
idrequired — Person ID
sift people graph显示以人为中心的关系图谱
- Arguments
idrequired — Person ID- Flags
--depth <value>— Relationship graph depth--include-inactive— Include inactive relationship edges
sift people kinship说明两人之间的亲缘或关系距离
- Arguments
egoPersonIdrequired — Ego/source person IDtargetPersonIdrequired — Target person ID- Flags
--max-depth <value>— Maximum relationship depth
sift people list列出人脉
- Flags
--category <family|romantic|professional|social>— Relationship category--contains <value>— Name substring filter--context <all|personal|work>— Relationship context--equals <value>— Exact name filter--has-no-email— Only contacts without an email--limit <value>— Maximum number of results--relationship <value>— Filter by relationshipToUser--starts-with <value>— Name prefix filter
sift people relate建立或更新两人之间的关联
- Arguments
personAIdrequired — First person IDpersonBIdrequired — Second person ID- Flags
--dry-run— Preview the relationship payload without writing--notes <value>— Relationship notes--type <value>— Relationship type, e.g. colleague, sibling, spouse, collaborator-y, --yes— Apply without prompting
sift people search搜索人脉
- Arguments
queryrequired — Search query- Flags
--contains <value>— Name substring filter--equals <value>— Exact name filter--has-no-email— Only contacts without an email--limit <value>— Maximum number of results--relationship <value>— Filter by relationshipToUser--starts-with <value>— Name prefix filter
sift people timeline列出与该人脉相关的时间轴记录
- Arguments
idrequired — Person ID- Flags
--limit <value>— Maximum facts to return--order <asc|desc>— Sort order--role <value>— Filter by entity role, comma-separated
sift people update更新人脉
- Arguments
idrequired — Person ID- Flags
--birth-year <value>— Birth year--birthday <value>— Birthday (YYYY-MM-DD)--company <value>— Company name--email <value>— Email address--estimated-age <value>— Estimated age--job-title <value>— Job title--linkedin-url <value>— LinkedIn profile URL--location <value>— Location--mbti <value>— MBTI type (e.g. INTJ, ENFP)--name <value>— Full name--notes <value>— Notes about this person--phone <value>— Phone number--relationship <value>— Relationship to user--website <value>— Personal website
项目
项目管理。
7 个命令
sift projects archive归档项目
- Arguments
idrequired — Project ID- Flags
-y, --yes— Skip confirmation
sift projects context获取项目上下文(任务、笔记、信号)
- Arguments
idrequired — Project ID
sift projects create创建项目
- Flags
--emoji <value>— Single emoji--name <value>— Project name--status <planning|active|on_hold|blocked|completed>— Project status--summary <value>— Project summary
sift projects list列出项目
- Flags
--include-archived— Include archived projects--status <planning|active|on_hold|blocked|completed>— Filter by status
sift projects planning获取项目的规范 CSN 规划快照
- Arguments
idrequired — Project ID
sift projects planning-recompute重新计算项目的规范 CSN 规划快照
- Arguments
idrequired — Project ID
sift projects update更新项目
- Arguments
idrequired — Project ID- Flags
--emoji <value>— Single emoji--name <value>— Project name--status <planning|active|on_hold|blocked|completed>— Project status--summary <value>— Project summary
方案
内置研究工作流方案。
2 条命令
sift recipes list列出内置研究工作流方案
sift recipes show查看内置研究工作流方案
- Arguments
idrequired — Recipe ID
Relationships
79 commands
sift relationships actions dry-runPreview a relationship action on the server without committing it
- Arguments
idrequired — Relationship action IDentity— Relationship entity as opportunity:<uuid> or prospect:<uuid>- Flags
--entity <value>— Target as prospect:uuid or opportunity:uuid--idempotency-key <value>— Stable 8-128 character key for reproducible server-side previews--input <value>— Action input JSON object--input-file <value>— Path to action input JSON
sift relationships actions listList registered relationship actions and their commitment policies
- Flags
--approval-policy <value>— Filter by approval policy--commitment <value>— Filter by commitment class--limit <value>— Maximum results
sift relationships actions runRun any canonical relationship action with replay-safe caller identity
- Arguments
idrequired — Canonical action ID from relationships actions listentity— Relationship entity as opportunity:<uuid> or prospect:<uuid>- Flags
--entity <value>— Target as prospect:uuid or opportunity:uuid--expected-version <value>— Current collaboration version--idempotency-key <value>— Stable 8-128 character key for exact retries--input <value>— Action input JSON object--input-file <value>— Path to an action input JSON file--yes— Confirm action execution
sift relationships actions showShow one registered relationship action
- Arguments
idrequired — Relationship action ID
sift relationships briefCompose an exact, evidence-grounded relationship brief
- Arguments
subjectrequired — Subject as prospect:uuid or opportunity:uuid- Flags
--lens <value>— Relationship lens identifier
sift relationships composer approver-edit创建署名审批人修订,并重新开启辅导审查
- Arguments
idrequired — Composer session UUID- Flags
--body <value>--expected-version <value>--reason <value>--subject <value>--yes— Confirm the attributed edit
sift relationships composer bind-sender将一个指定的、已激活的连接邮箱账户绑定到一份持久化草稿
- Arguments
idrequired — Composer session UUID- Flags
--expected-version <value>— Current optimistic session version--sender <value>— Connected email account UUID
sift relationships composer confirm确认已绑定的纲要,并创建其第一个持久化草稿修订版
- Arguments
idrequired — Composer session UUID- Flags
--expected-version <value>— Current optimistic session version--yes— Confirm brief and create draft
sift relationships composer create创建或恢复一个持久化的“关系纲要到草稿”会话
- Arguments
subjectrequired — Subject as prospect:uuid or opportunity:uuid- Flags
--file <value>— JSON file with source, recommendation, brief, and ranked evidence
sift relationships composer dispatch保留精确审批结果,立即发送或定时发送一条受管控的关系消息
- Arguments
idrequired — Composer session UUID- Flags
--at <value>— ISO-8601 time required for send_later--choice <approve_only|send_now|send_later>— Dispatch outcome--expected-version <value>— Current optimistic session version--idempotency-key <value>— Stable retry key for this exact choice and time--yes— Confirm this governed dispatch choice
sift relationships composer evaluate评估当前持久化草稿,并附加锚定到短语的辅导建议
- Arguments
idrequired — Composer session UUID- Flags
--expected-version <value>— Current optimistic session version
sift relationships composer finding-action解释、预览、接受、拒绝或撤销一条可追溯的辅导建议
- Arguments
idrequired — Composer session UUIDfindingrequired — Coaching finding UUID- Flags
--action <explain|preview|accept|reject|undo>--expected-version <value>— Current optimistic session version--yes— Confirm a mutating coaching decision
sift relationships composer history按记录顺序显示仅可追加的指导和草稿决策历史
- Arguments
idrequired — Composer session UUID
sift relationships composer request-approval为精确的编辑器审核请求或恢复审批;审批仅限浏览器内操作
- Arguments
idrequired — Composer session UUID- Flags
--expected-version <value>— Current optimistic session version--idempotency-key <value>— Stable approval-request retry key--yes— Confirm the approval request
sift relationships composer request-changes将精确的审核返回给作者,不修改消息内容
- Arguments
idrequired — Composer session UUID- Flags
--expected-version <value>--reason <value>--yes— Confirm the review interruption
sift relationships composer resolve-identity为一条已审核消息解决或路由内联的潜在重复项门控
- Arguments
idrequired — Composer session UUID- Flags
--candidate <value>— Candidate person UUID--decision <same_person|different_people|route_to_owner>--expected-version <value>--yes— Confirm the identity decision
sift relationships composer review将通过指导的持久草稿继续至受控审查
- Arguments
idrequired — Composer session UUID- Flags
--approver-user-id <value>— Optional active internal workspace member who will approve the exact message--expected-version <value>— Current optimistic session version--yes— Confirm the transition to review
sift relationships composer save-draft保存关系草稿的新持久修订版
- Arguments
idrequired — Composer session UUID- Flags
--file <value>— JSON file with expectedVersion, subject, and bodyText
sift relationships composer show显示一个持久的关系编辑器会话
- Arguments
idrequired — Composer session UUID
sift relationships composer update在创建草稿前,更新绑定的意图和排序后的证据
- Arguments
idrequired — Composer session UUID- Flags
--file <value>— JSON file with expectedVersion, brief, and evidence
sift relationships contact-plan inspect检查一个规范的个人联系方案,不修改其源生命周期
- Arguments
personrequired — Canonical person UUID- Flags
--from <value>— Inclusive ISO-8601 inspection window start--limit <value>— Maximum projected items--timezone <value>— IANA display timezone--to <value>— Inclusive ISO-8601 inspection window end
sift relationships diagnoses create创建一个不可变的、基于证据的关系诊断快照
- Flags
--file <value>— Diagnosis JSON file--yes— Confirm immutable diagnosis creation
sift relationships diagnoses list列出不可变的关系诊断快照
- Flags
--limit <value>--subject-id <value>--subject-type <prospect|opportunity>
sift relationships email-accounts list列出可用于关系草稿和受控发送的发件人账户
sift relationships generated-assets draft从已批准的外联资产创建私人沟通提案;此操作不会发送
- Arguments
idrequired — Generated outreach/follow-up asset UUID- Flags
--file <value>— Exact senderAccountId, destination, optional edits, and idempotency JSON file--yes— Confirm proposal creation
sift relationships generated-assets generate从已批准的精确行动手册和源资产版本生成一个有依据的资产
- Flags
--file <value>— Generation request JSON file--yes— Confirm immutable asset generation
sift relationships generated-assets list列出生成的关系资产及其清单和最新工件审查
- Flags
--limit <value>--subject-id <value>--subject-type <prospect|opportunity>
sift relationships generated-assets review向生成的工件追加人工审批决策
- Arguments
idrequired — Generated asset UUID- Flags
--decision <approved|rejected|needs_changes>--reason <value>--yes— Confirm append-only review
sift relationships generated-assets review-history列出一个生成资产的仅可追加审查历史
- Arguments
idrequired — Generated asset UUID
sift relationships generated-assets show显示一个生成的资产、其精确清单和最新工件审批
- Arguments
idrequired — Generated asset UUID
sift relationships generated-assets usage-list列出一个生成资产的仅可追加使用记录和成果链接
- Arguments
idrequired — Generated asset UUID
sift relationships generated-assets usage-record向生成的资产追加导出、内部分享或成果链接记录
- Arguments
idrequired — Generated asset UUID- Flags
--idempotency-key <value>— Stable 8-128 character retry key--kind <exported|shared_internally|outcome_linked>--outcome <value>— Stable outcome reference; required for outcome_linked--yes— Confirm append-only usage record
sift relationships handoffs listList immutable relationship outcome receipts in the onboarding queue
- Flags
--limit <value>— Maximum receipts
sift relationships handoffs showShow one immutable relationship outcome handoff receipt
- Arguments
idrequired — Outcome handoff receipt UUID
sift relationships meetings brief create根据已批准的人脉证据,撰写一份引证完备的会议简报
- Flags
--input <value>— Exact meeting brief input JSON--input-file <value>— Path to exact meeting brief input JSON
sift relationships meetings brief show通过重构其确切输入来读取当前的引证会议简报
- Flags
--input <value>— Exact owner, subject, approved asset, and optional meeting JSON--input-file <value>— Path to exact meeting brief input JSON
sift relationships meetings evidence capture捕获经同意的会议记录证据或主动录入的手动笔记
- Flags
--idempotency-key <value>— Caller-owned stable key for exact retries--input <value>— Exact meeting evidence input JSON--input-file <value>— Path to exact meeting evidence input JSON--yes— Confirm canonical evidence capture
sift relationships meetings evidence show检查一项权威会议证据及其来源标识
- Arguments
idrequired — Relationship meeting evidence event UUID
sift relationships meetings proposals apply完成最新的前提条件检查后,原子化应用一项已批准的会议提案
- Arguments
idrequired — Approved meeting proposal UUID- Flags
--expected-proposal-digest <value>— Exact reviewed proposal digest--yes— Confirm the persistent internal relationship-state change
sift relationships meetings proposals create为一项已注册的人脉操作,创建一个不可变的、引用证据的提案
- Flags
--idempotency-key <value>— Caller-owned stable key for exact retries--input <value>— Exact meeting proposal input JSON--input-file <value>— Path to exact meeting proposal input JSON--yes— Confirm immutable proposal creation
sift relationships meetings proposals list列出源自会议的人脉提案及其审查周期
- Flags
--status <pending|approved|rejected|committed|failed|stale>— Filter by lifecycle status--subject-id <value>— Filter by subject UUID--subject-kind <prospect|opportunity>— Filter by subject kind
sift relationships meetings proposals review为一项未经更改的会议提案,追加一次可追溯的批准或拒绝记录
- Arguments
idrequired — Meeting proposal UUID- Flags
--input <value>— Exact decision, reason, and expectedProposalDigest JSON--input-file <value>— Path to exact proposal review input JSON--yes— Confirm attributable proposal review
sift relationships meetings proposals show检查一项源自会议的人脉提案、相关证据、变更差异和恢复计划
- Arguments
idrequired — Meeting proposal UUID
sift relationships outcomes describe描述运行时的人脉成果合约及安全边界
sift relationships outcomes inspect在 Answer/C0 模式下检查已声明的人脉成果,不进行持久化
- Flags
--input <value>— Exact InspectRelationshipOutcomeInput JSON--input-file <value>— Path to exact inspection input JSON
sift relationships outcomes interventions显示快照中唯一的、有界且可审查的干预批次
- Arguments
snapshotIdrequired — Relationship-outcome snapshot ID
sift relationships outcomes list使用有界游标分页,列出人脉成果快照
- Flags
--cursor <value>— Opaque cursor from the previous page--limit <value>— Maximum snapshots to return--projection <sales_forecast|onboarding_handoff>— Filter by projection--scope-anchor-id <value>— Filter by declared scope anchor ID--scope-anchor-kind <value>— Filter by declared scope anchor kind
sift relationships outcomes records使用有界游标分页,列出精确的快照记录
- Arguments
snapshotIdrequired — Relationship-outcome snapshot ID- Flags
--cursor <value>— Opaque cursor from the previous page--limit <value>— Maximum records to return--posture <value>— Filter by the declared record posture--risk <value>— Filter by low, moderate, high, or unknown operational risk
sift relationships outcomes review记录一次幂等的快照审查,不执行任何提议的已注册操作
- Arguments
snapshotIdrequired — Relationship-outcome snapshot ID- Flags
--dry-run— Validate authority and selected interventions without recording a review--idempotency-key <value>— Caller-owned stable key for this exact review--input <value>— Exact ReviewRelationshipOutcomeSnapshotInput JSON (idempotency is injected)--input-file <value>— Path to exact snapshot review input JSON--yes— Confirm append-only review recording
sift relationships outcomes run创建或模拟运行一个 C2 人脉成果快照;绝不执行提议的操作
- Flags
--dry-run— Validate and project without persisting a snapshot--idempotency-key <value>— Caller-owned stable key for this exact snapshot request--input <value>— Exact CreateRelationshipOutcomeSnapshotInput JSON (idempotency is injected)--input-file <value>— Path to exact snapshot input JSON--yes— Confirm C2 snapshot persistence
sift relationships outcomes show显示一项不可变的人脉成果快照
- Arguments
snapshotIdrequired — Relationship-outcome snapshot ID
sift relationships playbook-recommendations create记录一项包含至少三个可行步骤的重要建议
- Flags
--file <value>— Recommendation JSON file--yes— Confirm immutable recommendation creation
sift relationships playbook-recommendations list列出已记录的人脉行动手册建议及其备选步骤
- Flags
--limit <value>--subject-id <value>--subject-type <prospect|opportunity>
sift relationships playbooks create从 runtime-contract JSON 创建一个不可变的、未经批准的人脉行动手册版本
- Flags
--file <value>— Playbook version JSON file--yes— Confirm immutable version creation
sift relationships playbooks list列出不可变关系剧本的版本及最新审批决定
- Flags
--limit <value>— Maximum versions--stable-key <value>— Filter by stable playbook key
sift relationships playbooks review为剧本版本追加一条人工审批决定
- Arguments
idrequired — Playbook version UUID- Flags
--decision <approved|rejected|needs_changes>--reason <value>— Attributable review reason--yes— Confirm append-only review
sift relationships playbooks show显示单个不可变关系剧本版本及其最新审批信息
- Arguments
idrequired — Playbook version UUID
sift relationships proposals getInspect one relationship action proposal and its evidence
- Arguments
idrequired — Proposal ID
sift relationships proposals listList pending and historical relationship action proposals
- Flags
--enrollment <value>— Filter by enrollment ID--limit <value>— Maximum results--sequence <value>— Filter by sequence ID--status <value>— Filter by proposal status
sift relationships proposals request-approvalRequest short-lived human approval for a proposal; cannot approve or consume it
- Arguments
idrequired — Proposal ID- Flags
--idempotency-key <value>— Stable 8-128 character key for safe retries--purpose <value>— Non-secret reason shown to the human approver
sift relationships queues listList relationship team queues in the active workspace
sift relationships recommendations executeExecute an accepted, current recommendation through the canonical action runtime
- Arguments
idrequired — Recommendation UUID- Flags
--yes— Confirm canonical action execution
sift relationships recommendations listList immutable relationship recommendation snapshots
- Flags
--lens <value>— Relationship lens identifier--limit <value>— Maximum recommendations--subject <value>— Filter by prospect:uuid or opportunity:uuid
sift relationships recommendations reviewAppend an ordinary review decision to a relationship recommendation
- Arguments
idrequired — Recommendation UUID- Flags
--decision <accepted|dismissed|snoozed|needs_information>— Review decision--reason <value>— Review reason--snoozed-until <value>— Offset-aware ISO date-time, required for snoozed--yes— Confirm the append-only review
sift relationships recommendations showShow a relationship recommendation with its exact citations and gates
- Arguments
idrequired — Recommendation UUID
sift relationships sequences createCreate a supervised relationship sequence from a JSON definition
- Flags
--definition <value>— Sequence definition JSON object--definition-file <value>— Path to sequence definition JSON--idempotency-key <value>— Stable 8-128 character key for safe retries
sift relationships sequences dry-runPreview a supervised sequence on the server without committing it
- Arguments
idrequired — Sequence ID- Flags
--idempotency-key <value>— Stable 8-128 character key for reproducible server-side previews--input <value>— Dry-run input JSON object--input-file <value>— Path to dry-run input JSON
sift relationships sequences enrollEnroll a relationship target in a supervised sequence
- Arguments
idrequired — Sequence ID- Flags
--idempotency-key <value>— Stable 8-128 character key for safe retries--input <value>— Enrollment input JSON object--input-file <value>— Path to enrollment input JSON
sift relationships sequences getShow one supervised relationship sequence
- Arguments
idrequired — Sequence ID
sift relationships sequences listList supervised relationship sequences
- Flags
--limit <value>— Maximum results--status <value>— Filter by sequence status
sift relationships sequences pausePause a supervised relationship sequence
- Arguments
idrequired — Sequence ID- Flags
--idempotency-key <value>— Stable 8-128 character key for safe retries--reason <value>— Non-secret pause reason
sift relationships sequences resumeResume a supervised relationship sequence
- Arguments
idrequired — Sequence ID- Flags
--idempotency-key <value>— Stable 8-128 character key for safe retries
sift relationships sequences runsList execution evidence for a supervised relationship sequence
- Arguments
idrequired — Sequence ID- Flags
--limit <value>— Maximum results--status <value>— Filter by run status
sift relationships sequences updateUpdate a supervised relationship sequence from a JSON definition
- Arguments
idrequired — Sequence ID- Flags
--definition <value>— Sequence update JSON object--definition-file <value>— Path to sequence update JSON--idempotency-key <value>— Stable 8-128 character key for safe retries
sift relationships showShow the current relationship collaboration snapshot
- Arguments
entityrequired — Relationship entity as opportunity:<uuid> or prospect:<uuid>
sift relationships source-assets create创建未经审批的不可变溯源资产版本
- Flags
--file <value>— Source-asset version JSON file--yes— Confirm immutable version creation
sift relationships source-assets list列出不可变溯源资产版本及最新审批决定
- Flags
--limit <value>— Maximum versions--stable-key <value>— Filter by stable source-asset key
sift relationships source-assets review为源资产版本追加一条人工审批决定
- Arguments
idrequired — Source-asset version UUID- Flags
--decision <approved|rejected|needs_changes>--reason <value>--yes— Confirm append-only review
sift relationships source-assets show显示单个精确的不可变源资产版本及其最新评审信息
- Arguments
idrequired — Source-asset version UUID
研究
研究工作流规划与编排。
4 个命令
sift research init创建研究项目和标准数据集
- Arguments
namerequired — Research project name- Flags
--dry-run— Preview project/dataset creation without writing--template <historical-research>— Research template--yes— Confirm creation without prompting
sift research plan在写入数据前规划确定性的研究工作流
- Arguments
goalrequired — Research goal- Flags
--project <value>— Existing project ID--source-dataset <value>— Existing sources dataset ID
sift research run为研究方案创建确定性的智能体任务
- Arguments
reciperequired — Research run recipe- Flags
--agent <value>— Assigned agent alias--context <value>— Additional input context JSON object--context-file <value>— Additional input context JSON file--dry-run— Preview work item payload without writing--project <value>— Project ID--source-dataset <value>— Source dataset ID--yes— Confirm work item creation without prompting
sift research status查看研究项目上下文和 CLI 就绪状态
- Arguments
project— Project ID
技能
可安装的 Siftable 技能包。
2 条命令
sift skills install将 Siftable 技能包安装至本地技能目录
- Arguments
idrequired — Skillpack ID- Flags
--force— Replace an existing installed skill--target <value>— Installed skills directory-y, --yes— Confirm replacing an existing skill
sift skills list列出可安装的 Siftable 技能包
任务
人工规划任务。
11 条命令
sift tasks bulk-delete预览或批量删除任务
- Flags
--confirm— Execute deletion instead of preview--done— Filter by completed state--ids <value>— Comma-separated task IDs--phase <draft|open|in_flight|review|blocked|done|cancelled>--title-contains <value>— Title substring filter--title-equals <value>— Exact title filter--title-starts-with <value>— Title prefix filter--when <now|today|soon|later>
sift tasks complete标记任务为已完成
- Arguments
idrequired — Task ID
sift tasks coupling-create在同一项目的任务间创建 CSN 耦合边
- Arguments
idrequired — Source task IDtargetrequired — Target task ID- Flags
--note <value>— Optional note--strength <value>— Coupling strength (0-1)--type <info|resource>— Coupling type
sift tasks coupling-delete删除任务的 CSN 耦合边
- Arguments
idrequired — Task IDedgeIdrequired — Coupling edge ID- Flags
-y, --yes— Skip confirmation
sift tasks coupling-list列出任务的 CSN 耦合边
- Arguments
idrequired — Task ID
sift tasks create创建人工规划任务
- Flags
--acceptance-criteria <value>— Acceptance criteria (semicolon-separated text, e.g. "tests pass; docs updated")--description <value>— Task description--due <value>— Due date (ISO 8601)--effort <trivial|small|medium|large|epic|unknown>— Effort estimate--phase <draft|open|in_flight|review|blocked|done|cancelled>— Lifecycle phase--priority <do_now|schedule|delegate|someday>— Priority level--project <value>— Project ID--scope <value>— Scope boundaries (JSON object with include/exclude arrays)--title <value>— Task title
sift tasks delete删除任务
- Arguments
idrequired — Task ID- Flags
-y, --yes— Skip confirmation
sift tasks get获取人工规划任务详情
- Arguments
idrequired — Task ID
sift tasks list列出人工规划任务
- Flags
--cursor <value>— Continuation cursor from a previous page--effort <trivial|small|medium|large|epic|unknown>— Filter by effort--limit <value>— Maximum number of results--phase <draft|open|in_flight|review|blocked|done|cancelled>— Filter by phase--project <value>— Filter by project ID--status <inbox|next_action|in_progress|waiting_for|completed|archived>— Filter by status--title-contains <value>— Title substring filter--title-equals <value>— Exact title filter--title-starts-with <value>— Title prefix filter
sift tasks planning-update更新任务的 CSN 规划字段
- Arguments
idrequired — Task ID- Flags
--cynefin-confidence <value>— Cynefin confidence (0-1)--cynefin-domain <clear|complicated|complex|chaotic|aporetic>— Cynefin domain--cynefin-rationale <value>— Why this domain fits--cynefin-source <user|assistant|classifier>— Source of the planning classification--duration-model <value>— Duration model JSON, e.g. {"kind":"point","days":2}--reversibility <value>— Reversibility score (0-1)
sift tasks update更新人工规划任务
- Arguments
idrequired — Task ID- Flags
--acceptance-criteria <value>— Acceptance criteria (semicolon-separated text, e.g. "tests pass; docs updated")--blocked-reason <value>— Reason task is blocked--description <value>— Task description--due <value>— Due date (ISO 8601)--effort <trivial|small|medium|large|epic|unknown>— Effort estimate--phase <draft|open|in_flight|review|blocked|done|cancelled>— Lifecycle phase--priority <do_now|schedule|delegate|someday>— Priority level--project <value>— Project ID--scope <value>— Scope boundaries (JSON object with include/exclude arrays)--status <inbox|next_action|in_progress|waiting_for|completed|archived>— Task status--title <value>— Task title
时间轴
时间轴事实与叙述。
4 条命令
sift timeline create创建用户编写的时间轴事实
- Flags
--body <value>— Fact body or notes--confidence <low|medium|high>— Confidence level--entity <value>— Participant/entity as type:uuid or type:uuid:role; repeatable--fact-type <value>— Fact type--precision <millisecond|minute|hour|day|month|year|decade|century|millennium|mega_year|era>— Temporal precision--source-label <value>— Source/provenance label--source-note <value>— Source/provenance note--source-url <value>— Source/provenance URL--timestamp <value>— ISO timestamp--title <value>— Fact title--visibility <org_public|private|restricted>— Timeline visibility--year <value>— Historical year CE--year-end <value>— Historical end year CE
sift timeline delete撤回时间轴事实
- Arguments
idrequired — Timeline fact ID- Flags
--yes— Confirm retraction without prompting
sift timeline list使用限定过滤条件列出时间轴事实
- Flags
--cursor <value>— Pagination cursor--end <value>— End boundary, ISO timestamp or supported historical boundary--entity <value>— Entity filter as type:uuid--entity-role <value>— Comma-separated entity roles--fact-types <value>— Comma-separated fact types--limit <value>— Maximum items to return--order <asc|desc>— Sort order--q <value>— Text search query--source-types <value>— Comma-separated source types--start <value>— Start boundary, ISO timestamp or supported historical boundary
sift timeline narrative生成时间轴事实的叙述性摘要或解释
- Flags
--action <summarize|changed_since|led_to|what_next|cross_object>— Narrative action--entity <value>— Entity scope as type:uuid--entity-roles <value>— Comma-separated entity roles--fact-type <value>— Fact type filter--limit <value>— Maximum timeline facts to include--participant <value>— Participant filter as type:uuid--prompt <value>— Question or custom narrative prompt--q <value>— Text query filter--related-entity <value>— Related entity as type:uuid--source-type <value>— Source type filter
保险库
机密保险库。
15 条命令
sift vault auditList Vault audit events (requires vault:audit:read)
- Flags
--limit <value>— Maximum number of results--owner <value>— Exact Vault owner: personal or workspace:<tenant-id>
sift vault create存储新的加密机密
- Flags
--category <value>— Category--description <value>— Description--name <value>— Secret name--owner <value>— Exact Vault owner: personal or workspace:<tenant-id>--payload <value>— JSON payload to encrypt--slug <value>— Machine-friendly identifier--tags <value>— Comma-separated tags--type <env_var|credential|oauth_token|ssh_key|certificate|note>— Entry type--url <value>— Associated URL
sift vault leases clean清理单个本地租约构件,并证实其状态为已确认、失败或未知
- Arguments
idrequired- Flags
--checkout-root <value>— Git checkout used to prove runtime separation--owner <value>— Exact Vault owner: personal or workspace:<tenant-id>
sift vault leases list列出无值环境租约及其清理元数据
- Flags
--owner <value>— Exact Vault owner: personal or workspace:<tenant-id>
sift vault leases revoke撤销租约,移除存在的本地构件,并证实清理操作
- Arguments
idrequired- Flags
--checkout-root <value>— Git checkout used to prove runtime separation--owner <value>— Exact Vault owner: personal or workspace:<tenant-id>
sift vault leases status检查单个无值环境租约的生命周期
- Arguments
idrequired- Flags
--owner <value>— Exact Vault owner: personal or workspace:<tenant-id>
sift vault list列出保险库条目(仅元数据)
- Flags
--category <value>— Filter by category--limit <value>— Maximum number of results--owner <value>— Exact Vault owner: personal or workspace:<tenant-id>--type <env_var|credential|oauth_token|ssh_key|certificate|note>— Filter by entry type
sift vault materialize environment将指定的受管控 Vault 数据包拉取到 Sift 管理的开发环境文件中
- Flags
--approval-timeout <value>--checkout-root <value>— Checkout containing .sift/environment.yaml--destination <value>— Exact relative destination when the manifest declares more than one bundle--endpoint-class <development|production|ci|shared_host|policy_ineligible>— Explicit endpoint policy class; managed pulls accept development only--owner <value>— Exact Vault owner: personal or workspace:<tenant-id>--purpose <value>--replace— Replace a changed Sift-managed file using its exact observed precondition
sift vault materialize requestRequest human approval for one destination-bound Vault materialization
- Flags
--destination <value>--entry <value>--expected-digest <value>--field <value>--materializer-digest <value>--mode <0400|0600>--nonce <value>--overwrite--owner <value>— Exact Vault owner: personal or workspace:<tenant-id>--purpose <value>--runner-fingerprint <value>--runner-public-key <value>--tracked-exception--workspace-root <value>— Absolute local workspace root containing the destination
sift vault materialize revoke撤销一个发往目标的待处理 Vault 实体化
- Arguments
idrequired — Materialization ID- Flags
--legacy-context— Use the Personal/v0 identity of a request created with the deprecated path-shaped --workspace alias--owner <value>— Exact Vault owner: personal or workspace:<tenant-id>
sift vault materialize runRequest approval, wait, and materialize one Vault field at the exact approved path
- Flags
--approval-timeout <value>--destination <value>— Destination path; relative paths resolve inside --workspace-root before approval--entry <value>--field <value>--mode <0400|0600>--overwrite--owner <value>— Exact Vault owner: personal or workspace:<tenant-id>--purpose <value>--tracked-exception--workspace-root <value>— Absolute local workspace root containing the destination
sift vault materialize statusInspect safe status for a destination-bound Vault materialization
- Arguments
idrequired- Flags
--legacy-context— Use the Personal/v0 identity of a request created with the deprecated path-shaped --workspace alias--owner <value>— Exact Vault owner: personal or workspace:<tenant-id>
sift vault search搜索保险库条目
- Arguments
queryrequired — Search query- Flags
--limit <value>— Maximum number of results--owner <value>— Exact Vault owner: personal or workspace:<tenant-id>
sift vault update更新保险库条目元数据
- Arguments
idrequired — Vault entry ID- Flags
--category <value>— Category--description <value>— Description--name <value>— Entry name--owner <value>— Exact Vault owner: personal or workspace:<tenant-id>--tags <value>— Comma-separated tags--url <value>— Associated URL
sift vault use lease在 Git 工作树外创建一个指定的、短暂的环境包构件
- Flags
--approval-timeout <value>--checkout-root <value>— Checkout containing .sift/environment.yaml--consumer <value>— Bound consumer identifier; not a command allowlist--consumer-pid <value>— Optional already-running consumer process lifetime--destination <value>— Manifest destination used only to select one exact bundle--endpoint-class <development|production|ci|shared_host>--holder <value>— Bound holder identifier; defaults to the local runner fingerprint--owner <value>— Exact Vault owner: personal or workspace:<tenant-id>--purpose <value>--ttl <value>
工作
可执行代理工作队列。
25 条命令
sift work block将工作项标记为已阻塞
- Arguments
idrequired — Work item ID- Flags
--artifacts <value>— Artifact refs JSON array--claim-token <value>— Claim token returned by work claim (required for lease-owned transitions)--lease <value>— Lease seconds--memory-decision <value>— Durable code-memory decision JSON (required on agent complete/review): {"decision":"store"|"skip","rationale":"...","fact"?,"category"?,"repositoryId"?,"filePath"?,"paths"?,"evidenceChunkIds"?,"sourceCommit"?,"confidence"?}--owner <value>— Claim owner identity (required for lease-owned transitions)--reason <value>— Block or failure reason--summary <value>— Result summary--verification-results <value>— Verification evidence JSON array: [{"command","exitCode","output"?}]
sift work cancel取消工作项
- Arguments
idrequired — Work item ID- Flags
--artifacts <value>— Artifact refs JSON array--claim-token <value>— Claim token returned by work claim (required for lease-owned transitions)--lease <value>— Lease seconds--memory-decision <value>— Durable code-memory decision JSON (required on agent complete/review): {"decision":"store"|"skip","rationale":"...","fact"?,"category"?,"repositoryId"?,"filePath"?,"paths"?,"evidenceChunkIds"?,"sourceCommit"?,"confidence"?}--owner <value>— Claim owner identity (required for lease-owned transitions)--reason <value>— Block or failure reason--summary <value>— Result summary--verification-results <value>— Verification evidence JSON array: [{"command","exitCode","output"?}]
sift work claim认领下一个可用的可执行智能体工作项
- Arguments
id— Optional specific work item ID- Flags
--agent <value>— Agent alias to claim for--lease <value>— Lease seconds--owner <value>— Claim owner identity
sift work complete批准并完成可执行智能体工作项
- Arguments
idrequired — Work item ID- Flags
--artifacts <value>— Artifact refs JSON array--claim-token <value>— Claim token returned by work claim (required for lease-owned transitions)--lease <value>— Lease seconds--memory-decision <value>— Durable code-memory decision JSON (required on agent complete/review): {"decision":"store"|"skip","rationale":"...","fact"?,"category"?,"repositoryId"?,"filePath"?,"paths"?,"evidenceChunkIds"?,"sourceCommit"?,"confidence"?}--owner <value>— Claim owner identity (required for lease-owned transitions)--reason <value>— Block or failure reason--summary <value>— Result summary--verification-results <value>— Verification evidence JSON array: [{"command","exitCode","output"?}]
sift work contract check确定性地验证工作合约 V1 文件或已获取的工作项
- Arguments
id— Work item ID to fetch and validate- Flags
--file <value>— Path to a Work Contract V1 JSON file
sift work create创建可执行智能体工作项
- Flags
--acceptance-criteria <value>— Acceptance criteria JSON array or semicolon-separated text--agent <value>— Assigned agent alias--allow-deploy— Allow deployment in the authored contract--allow-migrate— Allow migrations in the authored contract--allow-network— Allow network access in the authored contract--allow-vault— Allow audited Vault access in the authored contract--allowed-actions <value>— Allowed actions JSON object--capture-applicability <conversational_capture_required|non_conversational|unclassified>— Prospective capture applicability; this does not determine study eligibility--capture-receipt <value>— Finalized capture receipt UUID to bind atomically; repeat for multi-surface handoffs--context <value>— Input context JSON object--contract <value>— Complete Work Contract V1 JSON--contract-file <value>— Path to a complete Work Contract V1 JSON file--depends-on <value>— Dependency JSON array: [{"workItemId":"<uuid>","requiredGate"?:"done"|"commands_passed"}]--dry-run— Preview the normalized payload and contract without creating work--exclude-scope <value>— Excluded workspace-relative write scope; repeat for multiple paths--profile <coding|read-only|read_only>— Author a complete Work Contract V1 using the coding or read-only profile--project <value>— Linked project ID--prompt <value>— Agent prompt or instructions--rank <value>— Queue rank--scope <value>— Included workspace-relative write scope; repeat for multiple paths--task <value>— Parent human planning task ID--title <value>— Executable work item title--verify <value>— Verification commands separated by semicolons--write-scope <value>— Write scope JSON object
sift work dependencies getGet authoritative dependencies and claimability for a work item
- Arguments
idrequired — Work item UUID
sift work dependencies setAtomically replace the authoritative dependencies for a work item
- Arguments
idrequired — Work item UUID- Flags
--depends-on <value>— Dependency JSON array; pass [] to clear dependencies
sift work dependency-policy getGet a project default work-dependency gate
- Flags
--project <value>— Project UUID
sift work dependency-policy setSet a project default work-dependency gate
- Flags
--gate <done|commands_passed>— Default gate for dependencies that omit requiredGate--project <value>— Project UUID
sift work editC2: Revise queued, unclaimed work while preserving its UUID and dependency edges. Do not use after claim or execution; wait for a queued unclaimed state or create follow-up work. Example: raise queue rank or refine acceptance criteria while the item is queued and unclaimed. Counterexample: do not change status, dependencies, or an active worker lease with this capability. Execution-contract edits supersede the active verification plan and reset verification evidence; queue-rank-only edits do not. The same Idempotency-Key and payload replays the original receipt; a different payload with that key is rejected.
- Arguments
idrequired — Work item ID- Flags
--acceptance-criteria <value>— Replacement acceptance criteria JSON array--allowed-actions <value>— Replacement allowed actions JSON object--context <value>— Replacement input context JSON object--contract <value>— Replacement Work Contract V1 JSON--contract-file <value>— Path to replacement Work Contract V1 JSON--dry-run— Show normalized diff and proof consequences without mutating--expected-revision <value>— Observed work item revision--idempotency-key <value>— Stable retry key for this exact edit--patch <value>— Editable fields as one JSON object--patch-file <value>— Path to an editable-fields JSON object--prompt <value>— Replacement prompt; use --patch for null--rank <value>— Replacement queue rank--reason <value>— Auditable reason for the revision--title <value>— Replacement title--verify <value>— Replacement verification commands as JSON array or semicolon-separated text--write-scope <value>— Replacement write scope JSON object
sift work fail将工作项标记为失败
- Arguments
idrequired — Work item ID- Flags
--artifacts <value>— Artifact refs JSON array--claim-token <value>— Claim token returned by work claim (required for lease-owned transitions)--lease <value>— Lease seconds--memory-decision <value>— Durable code-memory decision JSON (required on agent complete/review): {"decision":"store"|"skip","rationale":"...","fact"?,"category"?,"repositoryId"?,"filePath"?,"paths"?,"evidenceChunkIds"?,"sourceCommit"?,"confidence"?}--owner <value>— Claim owner identity (required for lease-owned transitions)--reason <value>— Block or failure reason--summary <value>— Result summary--verification-results <value>— Verification evidence JSON array: [{"command","exitCode","output"?}]
sift work get获取可执行代理的工作项详情
- Arguments
idrequired — Work item ID
sift work heartbeat延长工作项租约
- Arguments
idrequired — Work item ID- Flags
--artifacts <value>— Artifact refs JSON array--claim-token <value>— Claim token returned by work claim (required for lease-owned transitions)--lease <value>— Lease seconds--memory-decision <value>— Durable code-memory decision JSON (required on agent complete/review): {"decision":"store"|"skip","rationale":"...","fact"?,"category"?,"repositoryId"?,"filePath"?,"paths"?,"evidenceChunkIds"?,"sourceCommit"?,"confidence"?}--owner <value>— Claim owner identity (required for lease-owned transitions)--reason <value>— Block or failure reason--summary <value>— Result summary--verification-results <value>— Verification evidence JSON array: [{"command","exitCode","output"?}]
sift work list列出可执行代理的工作项
- Flags
--agent <value>— Filter by assigned agent alias--limit <value>— Maximum results--project <value>— Filter by project ID--status <value>— Filter by status--task <value>— Filter by parent human planning task ID
sift work release将已认领的工作项释放回队列
- Arguments
idrequired — Work item ID- Flags
--artifacts <value>— Artifact refs JSON array--claim-token <value>— Claim token returned by work claim (required for lease-owned transitions)--lease <value>— Lease seconds--memory-decision <value>— Durable code-memory decision JSON (required on agent complete/review): {"decision":"store"|"skip","rationale":"...","fact"?,"category"?,"repositoryId"?,"filePath"?,"paths"?,"evidenceChunkIds"?,"sourceCommit"?,"confidence"?}--owner <value>— Claim owner identity (required for lease-owned transitions)--reason <value>— Block or failure reason--summary <value>— Result summary--verification-results <value>— Verification evidence JSON array: [{"command","exitCode","output"?}]
sift work requeueReturn blocked work to the queue for a fresh claim
- Arguments
idrequired — Work item ID- Flags
--artifacts <value>— Artifact refs JSON array--claim-token <value>— Claim token returned by work claim (required for lease-owned transitions)--lease <value>— Lease seconds--memory-decision <value>— Durable code-memory decision JSON (required on agent complete/review): {"decision":"store"|"skip","rationale":"...","fact"?,"category"?,"repositoryId"?,"filePath"?,"paths"?,"evidenceChunkIds"?,"sourceCommit"?,"confidence"?}--owner <value>— Claim owner identity (required for lease-owned transitions)--reason <value>— Block or failure reason--summary <value>— Result summary--verification-results <value>— Verification evidence JSON array: [{"command","exitCode","output"?}]
sift work review将可执行代理工作标记为待人工审核
- Arguments
idrequired — Work item ID- Flags
--artifacts <value>— Artifact refs JSON array--claim-token <value>— Claim token returned by work claim (required for lease-owned transitions)--lease <value>— Lease seconds--memory-decision <value>— Durable code-memory decision JSON (required on agent complete/review): {"decision":"store"|"skip","rationale":"...","fact"?,"category"?,"repositoryId"?,"filePath"?,"paths"?,"evidenceChunkIds"?,"sourceCommit"?,"confidence"?}--owner <value>— Claim owner identity (required for lease-owned transitions)--reason <value>— Block or failure reason--summary <value>— Result summary--verification-results <value>— Verification evidence JSON array: [{"command","exitCode","output"?}]
sift work revisionsList immutable, attributable work-item revision receipts and field diffs
- Arguments
idrequired — Work item ID- Flags
--limit <value>— Maximum revisions to return
sift work start将工作项标记为运行中
- Arguments
idrequired — Work item ID- Flags
--artifacts <value>— Artifact refs JSON array--claim-token <value>— Claim token returned by work claim (required for lease-owned transitions)--lease <value>— Lease seconds--memory-decision <value>— Durable code-memory decision JSON (required on agent complete/review): {"decision":"store"|"skip","rationale":"...","fact"?,"category"?,"repositoryId"?,"filePath"?,"paths"?,"evidenceChunkIds"?,"sourceCommit"?,"confidence"?}--owner <value>— Claim owner identity (required for lease-owned transitions)--reason <value>— Block or failure reason--summary <value>— Result summary--verification-results <value>— Verification evidence JSON array: [{"command","exitCode","output"?}]
sift work verification evidenceSubmit externally executed evidence for an exact plan version and step ID
- Arguments
idrequired — Work item ID- Flags
--artifacts <value>— Artifact reference JSON array for larger logs--attempt <value>— Caller-stable attempt identity--claim-owner <value>— Active lease owner for workspace-service evidence--claim-token <value>— Opaque active lease token for workspace-service evidence--environment <value>— Execution environment label--exit-code <value>— Process exit code when applicable--outcome <passed|failed|error>— Attempt outcome--output <value>— Bounded output excerpt; secrets are redacted by the API--plan-version <value>— Active verification plan version--provenance <value>— Evidence provenance JSON object--ran-at <value>— RFC3339 execution timestamp--step <value>— Stable verification step UUID
sift work verification historyList immutable verification-plan history and coverage
- Arguments
idrequired — Work item ID
sift work verification planShow the active versioned verification plan and coverage
- Arguments
idrequired — Work item ID
sift work verification reviseCreate an audited active verification-plan revision
- Arguments
idrequired — Work item ID- Flags
--expected-version <value>— Observed active plan version--provenance <value>— Revision provenance JSON object--reason <value>— Audited revision reason--steps <value>— Verification step JSON array--steps-file <value>— Path to a verification step JSON array--yes— Confirm activation without prompting
sift work verifyDeprecated compatibility command. Hosted LLM verification is retired; use verification-command evidence.
- Arguments
idrequired — Work item ID- Flags
--history— List historical verifier runs recorded before retirement--model <value>— Deprecated; ignored because hosted verification is retired--reps <value>— Deprecated; ignored because hosted verification is retired
Worker
本地可执行任务运行器。
2 个命令
sift worker dispatch为两个 AWS agent worker 构建一个零支出、接口无关的执行计划
- Flags
--max-runtime-minutes <value>— Per-worker runtime ceiling--source-sha <value>— Reviewed 40-character source Git SHA; defaults to a clean current HEAD--surface <cli|t3|codex|claude-code|mcp>— Calling control surface; does not change dispatch identity--work-item <value>— Queued Sift work-item UUID; repeat exactly twice
sift worker run认领可执行任务,运行本地 Worker 命令,并报告待审核产物
- Flags
--agent <value>— Agent alias to claim work for--command <value>— Local command to run for the claimed work item--cwd <value>— Fallback working directory for the local command--lease <value>— Lease seconds--owner <value>— Worker owner fingerprint
工作区
3 个命令
sift workspaces current显示当前选择的个人或协作工作区
sift workspaces list列出认证用户可用的协作工作区
sift workspaces use选择一个协作工作区用于浏览,或选择“个人”以清除设置
- Arguments
idrequired — Workspace ID, or "personal"
交互式 Copilot
sift interactive 命令可启动终端助手,支持对话、运行工具、编辑代码、开启并行智能体分支、规划工作及渲染图表 — 全在终端内完成,直接操作你的文件系统和 Siftable 工作图谱。
$ sift interactive
运行要求与启动
- 需要安装 Bun。
sift interactive会重新调用 Bun;若未安装,将输出curl -fsSL https://bun.sh/install | bash。 - 需要身份验证 — 支持
--token、SIFT_TOKEN或sift auth login。 - 大脑在进程内运行 — 无需启动独立的守护进程。
- 写入范围限定在工作区根目录 — 即启动目录向上包含
.git的最近父目录。这是/status报告的边界,也是原生写入路径强制执行的范围。
界面
The composer is a full readline-capable text area. Enter submits; Shift+Enter / Ctrl+J insert a newline; submitting while the agent is busy queues your message. !command runs via bash -lc into the transcript (a bare !cd <path> changes the session directory; output clipped to ~4000 chars). Large or structured pastes become a collapsed chip; pasted images are validated, normalized, and attached. ? on an empty composer shows hotkeys.
斜杠命令
输入 / 开启命令菜单。隐藏命令仍可输入使用,但不会在菜单中显示。
| Command | Group | Description |
|---|---|---|
/help, /hotkeys, /status | Session | Command list, keyboard shortcuts, and current model/scope/queue status |
/cwd [path] | Session | Show or change the working directory (recomputes the workspace root) |
/copy [last|all|explorer] | Session | Copy the latest reply, whole transcript, or latest explorer report to the clipboard |
/clear, /quit | Session | Reset the transcript; exit |
/threads [clear], /compact | Session | Manage the persisted thread; force a context compaction (requires context compaction enabled) |
/model [id] [effort] | Model | Open the model picker or select a model and reasoning effort directly |
/codex [login|on|use|off|logout|status] | Model | Control the Codex (ChatGPT) engine; default subcommand is status |
/key <provider> <key>, /key vault <provider> | Model | Store a provider API key, or hydrate it from Siftable Vault |
/login | Model | Siftable device-code login from inside the TUI |
/explorer | Model | Configure the repo Explorer (context-gathering backend) |
/skills [name] | Skills | List discovered skills, or print one skill's body |
/branches | Branches | Open the parallel-agent branches hub |
/spawn <title> [--rw <globs>|--rw-any|--ro] | Branches | Start a child agent branch in its own git worktree with an access mode |
/merge, /rebase, /sendback, /reject | Branches | Land, replay, resume, or reject a child branch |
/work | Work | Open the work-queue hub (board of agents and items by status) |
/plan [objective | work [--apply] [--after SRC:DST] [--limit N] | view] | Work | Plan from an objective, or compute a precedence DAG over the agent work queue as a Mermaid graph |
/handoff <title> [--agent ..] [--files ..] [--acceptance ..] [--verify ..] | Work | Create a Siftable work item from the current context |
/proof <claim>, /remember <fact> --category <..> | Work | Gather code/test evidence; store durable code memory |
/crew [list|show|new|run], /collab | Crews | Manage and run multi-agent crews; show in-process collaboration sessions |
/mermaid [request|file.mmd|source], /view | Diagrams | Render Mermaid (NL request, file, or source) in the terminal; open the pannable viewer |
/theme, /sounds [on|off] | Appearance | Open the appearance picker; toggle UI sounds |
模型与引擎
The model picker (/model) is two-stage: choose a model, then a reasoning effort (saved to ~/.siftable/prefs.json). The catalog includes GPT-5.6 Sol, Terra, and Luna (ChatGPT plan via the Codex engine — /codex login / /codex on select Sol by default), Claude Opus 4.8 (OpenRouter or direct Anthropic with ANTHROPIC_API_KEY), Claude Sonnet 4.6 / Haiku 4.5, Gemini 3.x Flash / Flash-Lite, GPT-5.4 mini / nano, and Morph v3 Large (apply-only). Codex drives the OpenAI codex app-server sidecar (ChatGPT device-code login). All other providers route through the bundled OpenFunction agent via <PROVIDER>_API_KEY. /key vault <provider> hydrates a key from Siftable Vault behind an approval prompt; the secret is never printed or written to disk.
Repo Explorer
/explorer configures repository context gathering before a turn. Modes: auto, off, deterministic, scout, fanout, warpgrep. Pick a scout model and budget (cheap/normal/deep). warpgrep needs MORPH_API_KEY (auto-hydrated from Vault when available). Settings persist in prefs.json.
技能
系统按优先级(项目 > 用户 > 内置)从 <root>/{.sift,.claude,.codex,.agents}/skills、~/.claude|.codex|.agents/skills、~/.config/sift/skills 及内置包中检索 SKILL.md 技能。智能体可通过工具调用技能,系统提示词中最多展示约 50 个。
快捷键
覆盖层会优先捕获键盘输入,因此按键绑定与模式相关。
- Enter submit (queues if busy) · Shift+Enter/Ctrl+J newline · Tab complete a lone
/foo· ↑/↓ prompt history - Esc abort turn → clear draft → deselect · Ctrl+C abort → clear → deselect → quit (never copies) · Ctrl+D quit on empty draft
- Cmd/Super+A select all · Cmd+C/Ctrl+Shift+C copy selection or latest reply · Ctrl+O Explorer diagnostics · ? hotkeys
- Approval gate: y/Enter allow once · a always · b bypass-all · n/Esc deny
原生加速
性能关键路径(上下文压缩、长线程记忆、文件系统扫描、合并编排、图像处理)通过 Bun FFI 加载的原生 Zig 模块运行,且均配有同步的 TypeScript 回退方案。本软件包内置 macOS (Apple Silicon) 和 Linux (x64) 的预构建库;其他平台将使用回退方案。设置 SIFT_NO_NATIVE=1 可强制使用回退方案,设置 SIFT_CONTEXT_COMPACTION=1 可启用实时上下文 Token 计数器以及线程持久化与恢复。
外观与声音
/theme 提供 10 种配色方案(默认为 "sieve" — 炭黑底色上的暖琥珀色),保存于 ~/.siftable/appearance.json。/sounds 用于切换 UI 音效(默认关闭),保存于 ~/.siftable/sounds.json,并可通过 SIFT_SOUNDS 覆盖。
The copilot is read-only by default; write/edit tools are scoped to the workspace root and gated by a four-way approval prompt (allow once / always / bypass-all / deny). With nothing listening, requests deny. Auto-approve environment overrides are always scrubbed at launch.
完整参考,包含所有斜杠命令、快捷键绑定、环境变量和配置文件:GitHub 上的 docs/interactive.md。