一键导入
canary
Check system health and pressure signals. Shows current pressure score, component breakdown, and recommendations.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
菜单
Check system health and pressure signals. Shows current pressure score, component breakdown, and recommendations.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Source and independently verify candidates for a role from a job description URL. Runs the source-candidates workflow (Webset creation → Exa verification of every claim → validated/rejected CSVs). Use whenever the user wants to find candidates, source people for a role, run a recruiting search from a job posting, or invokes /source-candidates.
Verify a webset item's enrichments against public data using independent Exa searches. Returns structured verdict with confidence score, evidence, and discrepancies. Use when a channel event or workflow step requires enrichment verification before proceeding.
Produce uniform, verified background reviews / biographies / profiles for a SHORTLIST of entities (people or companies) by dispatching one research subagent per entity in parallel, then synthesizing and persisting the results. Use when the user asks for "bios", "background reviews", "profiles", "tell me about each of these", or a comparison across several candidates — especially after a [[websets-secondary-review]] produced a shortlist. For a single entity use [[deep-research-item]] instead; this is the parallel, many-entities harness.
Run comprehensive research on an entity — news, funding, tech stack, buyer mapping, angle building. Writes findings to the webset store and optionally to Airtable. Use after verification passes or when a channel event requires deep research on a specific entity.
View, add, modify, or remove per-source workflow configurations. These configs control how channel events are routed to workflow chains. Manages data/workflow-configs.json.
Build Claude Code Channels — MCP servers that push events into a live Claude Code session so Claude can react to things happening outside the terminal. Use this skill whenever the user wants to: build a custom channel server, push webhooks/CI alerts/monitoring events into Claude Code, build a two-way chat bridge (Telegram, Discord, or custom), relay tool-use permission prompts remotely, understand the channel notification protocol, declare the claude/channel or claude/channel/permission capability, wire up sender gating/allowlists, or package a channel as a plugin. Trigger on any mention of Channels, channel MCP, push events to Claude Code, Claude Code from phone, or remote control Claude Code.
| name | canary |
| description | Check system health and pressure signals. Shows current pressure score, component breakdown, and recommendations. |
| argument-hint | ["none"] |
| user-invocable | true |
| allowed-tools | Bash(cat *), Read |
Read the current canary signals and system phase, then produce a health report.
!cat .claude/state/canary_signals.json 2>/dev/null || echo '{"pressure": 0, "edit_count": 0, "files_touched": [], "turn_count": 0, "warnings": []}'
!cat .claude/state/system_phase.json 2>/dev/null || echo '{"phase": "exploration"}'
## Canary Health Report
### Pressure: [X%] [GREEN|YELLOW|ORANGE|RED]
### Components
- Edit velocity: [X%] ([N] edits)
- Scope breadth: [X%] ([N] files)
- Turn count: [X%] ([N] turns)
- Elapsed time: [X%]
### Current Phase: [phase]
### Warnings
[List any active warnings]
### Recommendation
[Based on pressure level and phase:
- GREEN: Continue current work
- YELLOW: Monitor scope, avoid new explorations
- ORANGE: Consider converging. Use /helm to evaluate phase transition.
- RED: Transition to completion. Use /close-out to identify cuts.]