recover-context

pchalasani's avatarfrom pchalasani

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.

⭐932stars🔀70forks📁View on GitHub🕐Updated Jan 4, 2026
[Memory and Context Management]

When & Why to Use This Skill

The recover-context Claude skill is a specialized tool designed to maintain continuity across AI agent sessions by extracting task history from session lineages. It strategically employs sub-agents to analyze parent sessions, allowing the primary agent to retrieve critical task states, decisions, and document references without bloating its own context window. This skill is essential for complex, multi-stage agentic workflows where preserving state and efficiency is paramount.

Use Cases

  • Resuming Interrupted Workflows: Automatically identify the last task worked on and its current status (completed, in-progress, or blocked) to pick up exactly where a previous session left off.
  • Context Window Optimization: Use specialized sub-agents to scan large session logs and history files, preventing the main agent from hitting token limits while still accessing vital historical data.
  • Cross-Session Knowledge Retrieval: Locate and reference specific markdown files, work logs, or code modifications mentioned in parent sessions to ensure project consistency.
  • Strategic Decision Tracking: Extract key decisions and chosen approaches from previous interactions to inform current problem-solving and avoid redundant work.
namerecover-context
descriptionExtract full context of the last task from the most recent parent session

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