Lean Context

Activate strict context-saving mode mid-conversation. Use when conversations are getting long, before starting large tasks, or when the user says /lean. Audits conversation state and creates a checkpoint summary.

kevinchamplin f1554bb 1.3 KB Updated

File contents

Lean Context Mode

Context efficiency rules are already enforced globally via CLAUDE.md. This skill adds conversation-specific actions when things are getting long.

Actions (do all of these immediately)

  1. Checkpoint summary: Write a 2-3 sentence summary of everything accomplished so far and what remains. This creates a "save point" that survives compaction.

  2. Stale context audit: Identify any large tool outputs earlier in the conversation that are no longer needed (old file reads, long bash outputs, superseded diffs). Note them — they'll be compressed first.

  3. Remaining work estimate: List what's left to do. If multiple independent tasks remain, suggest parallel agent delegation to keep the main context lean.

  4. Strict mode reminder: For the rest of this conversation:

    • Re-read nothing — reference earlier checkpoint summaries instead
    • Delegate ANY research to Explore agents
    • Keep all responses to 1-3 sentences unless showing code

Output

Respond with:

  • The checkpoint summary
  • "Lean mode active" confirmation
  • Any specific suggestions for the current task

kevinchamplin/claude-skills/tree/main/lean-context commit f1554bb62b

Frequently asked questions

npx skillmds@latest add kevinchamplin/lean-context