agent-efficiency-retrofitlisted

# Agent Efficiency Retrofit Use this skill when the user asks to analyze recent agent conversations, transcripts, or session history to reduce tool calls, token consumption, repeated errors, retry loops, or setup friction. This is the discoverable entrypoint; do not rely on a playbook file being found by chance. ## Goal Turn observed agent waste into a durable, visible improvement: 1. Sample recent Claude Code and Codex CLI sessions. 2. Identify repeated inefficient patterns with concrete examples. 3. Rank fixes by expected benefit and implementation ease. 4. Present up to 10 options, with a short recommendation. 5. After explicit user approval, implement the smallest durable fix in the right surface: hook, skill, concise agent instruction, script, or project docs. ## Fast Workflow ### 1. Bound the Analysis Prefer a bounded, recent sample over a broad transcript dump: - Claude Code: inspect recent `~/.claude` project/session data, stats cache, and history where available. - Codex CLI: inspect recent `~/.codex/sessions` and `~/.codex/history.jsonl` where available. - If the user requests subagents, split the sample into batches and ask each subagent for repeated patterns, not exhaustive summaries. Use cheap metadata first: command names, output sizes, retries, repeated file reads, repeated status/diff calls, failed commands, sleeps/polling, and hook blocks. Open full transcripts only for representative examples. ### 2. Classify Waste Patterns Look for patte