界面已本地化;技能介绍与说明文档按所选语言由 AI 自动翻译,代码、命令、路径和专业术语保留原文,以原作者文档为准。
探索技能
效率与协作 · SKILL.md

recover-context

recover-context

【效率与协作】Extract full context of the last task from the most recent parent session shown in the session lineage. Strategically uses sub-agents to avoid bloating your own context.

独立 GitHub 快照 · 自动分类 · 以原作者文档为准

技能说明

GitHub

name: recover-context description: Extract full context of the last task from the most recent parent session shown in the session lineage. Strategically uses sub-agents to avoid bloating your own context.

recover-context

Use this skill to extract context from a parent session when a session lineage is present (shown in the first user message of this conversation).

Instructions

  1. Identify the most recent parent session from the lineage chain (the last file in the chronological list).

  2. Use sub-agents to explore (to avoid bloating your own context):

    • If you have the Task tool with subagent support, use the session-searcher subagent (subagent_type: session-searcher) to analyze the most recent session
    • If sub-agents are NOT available, use the aichat:session-search skill instead
  3. Extract the following from the most recent session:

    • What was the last task being worked on?
    • What was the current state of that task (completed, in-progress, blocked)?
    • Any pending items or next steps mentioned?
    • Key decisions made or approaches chosen
  4. Also check for associated documents:

    • Issue specs or task descriptions referenced in the session
    • Any markdown files created during that session (check WORKLOG/, issues/, etc.)
    • Code files that were being modified
  5. Report back concisely:

    • State your understanding of the task context
    • List any files you found that are relevant
    • Ask the user how they'd like to proceed

Example Sub-agent Prompt

If using the Task tool with session-searcher subagent:

Analyze the session file at [path from lineage] and extract:
1. The last task being worked on (look at the final 20-30 messages)
2. Current state of that task
3. Any referenced markdown files (issue specs, work logs, etc.)
4. Pending next steps or blockers

Return a concise summary.

Constraints

  • Do NOT read large session files directly into your own context
  • ALWAYS delegate to sub-agents or the session-search skill
  • Keep your summary concise - the user knows what they were working on
  • Focus on the LAST task, not the entire session history
我的收藏