AgentSkillsCN

start

通过任务创建深化会话上下文。当严重程度为CRITICAL时，当存在未解决的阻塞问题时，或当用户询问会话状态、交接状态，或下一步该做什么时，可使用此功能。

SKILL.md
--- frontmatter
name: start
description: Deep session context hydration with task creation. Use when severity is CRITICAL, when there are unresolved blockers, or when the user asks about session state, handoff status, or what to work on next.
user-invocable: false
allowed-tools:
  - Bash(git *)
  - Bash(gh *)
  - Bash(date *)
  - Bash(ls *)
  - Read
  - Glob
  - Grep
  - TaskCreate
  - TaskUpdate
  - TaskGet
  - TaskList
model: opus

Handoff Start

<role> Senior engineer. Hospital shift changes. Read the chart, brief the incoming team. Precise, no fluff. </role> <task> Analyze the pre-loaded state below, check for drift, hydrate tasks, and report. Hooks auto-inject a lightweight summary on session start — this provides the thorough version with task hydration and drift detection. </task>

Pre-loaded State

state.json

!cat .handoff/state.json 2>/dev/null || echo "No state found — run /handoff:end first"

CONTEXT.md

!cat .handoff/CONTEXT.md 2>/dev/null || echo "No context found"

Git

Branch: !git branch --show-current 2>/dev/null || echo "unknown" !git status -s 2>/dev/null

Recent commits: !git log -10 --format='%h %s' 2>/dev/null || echo "No commits"

PRs

!gh pr list --limit 5 2>/dev/null || echo "gh not available or no PRs"

Steps

  1. •Check --resume availability: Read _runtime.hostname from state.json. If it matches current hostname (hostname -s) and session_id differs from current session, the previous session may be resumable at protocol level. Note this in output: Previous session resumable: claude --resume <session_id>. This gives full context replay vs. lossy handoff rehydration.
  2. •Analyze pre-loaded state above. Check source field for confidence ("manual-end" > "auto-save" > "init")
  3. •Check drift: verify Structure paths from CONTEXT.md still exist (Glob), report missing/new files
  4. •Check active tasks: TaskList, TaskGet for each — identify handoff/blocker/resume metadata

Hydrate tasks from state.json (idempotent):

  1. •Check existing tasks first — if tasks with metadata handoff: true and matching session already exist for the current session, skip creation (prevents duplicates on repeated invocation)
  2. •Complete previous handoff tasks from prior sessions (TaskUpdate status: "completed")
  3. •Create blocker tasks from blockers array (metadata: blocker: true, handoff: true, session: "${CLAUDE_SESSION_ID}")
  4. •Create resume task from resume object (metadata: resume: true, handoff: true, session: "${CLAUDE_SESSION_ID}")
  5. •Set dependencies: resume blocked by blockers

Output read-back:

code
╔═══════════════════════════════════════════╗
║  HANDOFF RECEIVED                         ║
║  Project: [name] | Stack: [stack]         ║
║  Severity: [emoji]                        ║
╚═══════════════════════════════════════════╝

💡 RESUMABLE: claude --resume <session_id>
   (only if hostname matches — full context replay vs lossy rehydration)

SINCE LAST SESSION
├─ Commits: [N] | PRs: [N] merged
└─ Tasks: [N] hydrated

⚠️  WATCH OUT FOR / 🚫 BLOCKERS / ❌ FAILED
[lists]

▶️  RESUME
[Next action, files, context]
────────────────────────────────────
Ready. What would you like to work on?