菜单
Use when the user wants to review their recent Claude Code sessions for patterns — analyzes the last N sessions (default 5) in the current project, dispatching parallel reviewers per session, then synthesizing cross-session findings
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Use when analyzing a large corpus of text, code, or data that exceeds a single agent's effective context - orchestrates parallel Worker subagents, Critic review subagents, and a final Summarizer subagent with task tracking and failure recovery
Use after initial design context is gathered, before brainstorming - resolves contradictions in requirements, disambiguates terminology, clarifies scope boundaries, and verifies assumptions to prevent building the wrong solution
Use when creating or developing anything, before writing code or implementation plans - refines rough ideas into fully-formed designs through structured Socratic questioning, alternative exploration, and incremental validation
Use when executing implementation plans with independent tasks in the current session - dispatches fresh subagent for each task, reviews once per phase, loads phases just-in-time to minimize context usage
Use when implementation is complete, all tests pass, and you need to decide how to integrate the work - guides completion of development work by presenting structured options for merge, PR, or cleanup
Use when completing tasks, implementing major features, or before merging to verify work meets requirements - dispatches code-reviewer subagent, handles retries and timeouts, manages review-fix loop until zero issues
| name | review-recent-sessions |
| description | Use when the user wants to review their recent Claude Code sessions for patterns — analyzes the last N sessions (default 5) in the current project, dispatching parallel reviewers per session, then synthesizing cross-session findings |
Review multiple recent sessions from the current project directory to identify cross-session patterns.
Do not use nested subagents. This workflow may dispatch first-level reviewer and synthesis agents. Those agents must read the provided files directly and must not dispatch additional subagents.
ed3d-extending-claude plugin must be installed.ed3d-session-reflection plugin must be installed (provides the conversation-reviewer agent and reduce-transcript.py script).The user may invoke this as:
/review-recent-sessions — review last 5 sessions/review-recent-sessions 10 — review last 10 sessionsUse the current session's transcript path to determine the project directory. The transcript path looks like:
~/.claude/projects/-Users-ed-Development-.../SESSION_ID.jsonl
The directory containing it is the project's session directory.
If you cannot determine the project directory, ask the user.
Find the most recent JSONL files in the project directory, sorted by modification time, limited to the requested count (default 5).
ls -t "<project_session_dir>"/*.jsonl | head -<count>
Exclude the current session's transcript (the user doesn't want to review the review session itself).
If fewer than 2 sessions are found, tell the user there aren't enough sessions to do a cross-session review and suggest using /review-session instead.
Create a working directory:
mkdir -p /tmp/session-review-batch
For each session, run the reduction script:
python3 "${CLAUDE_PLUGIN_ROOT}/scripts/reduce-transcript.py" "<session.jsonl>" "/tmp/session-review-batch/reduced-<N>.txt"
This can be done in a single bash command with a loop.
For each reduced transcript, dispatch a conversation-reviewer agent in the background:
Transcript path: /tmp/session-review-batch/reduced-N.txt Write your findings to: /tmp/session-review-batch/findings-N.md
Read the transcript, analyze it, and write your findings following your output format. Do not dispatch or invoke any subagents.
Dispatch ALL reviewers in a single message to maximize parallelism. Tell the user you've dispatched N reviewers and are waiting for results.
Once all reviewers complete, dispatch a general-purpose Sonnet agent to synthesize:
ed3d-basic-agents:sonnet-general-purpose Synthesize session reviews You are synthesizing findings from multiple Claude Code session reviews into a cross-session analysis.Read all findings files in /tmp/session-review-batch/findings-*.md
Produce a synthesis that identifies:
Recurring patterns — issues that appear across multiple sessions. These are the highest-value findings because they represent systematic problems.
Progression — is the user getting better or worse at prompting over time? Is the agent handling certain tasks better or worse?
Highest-impact recommendations — across all sessions, which recommendations would have the biggest effect? Prioritize:
Session-specific highlights — any single-session finding that's particularly noteworthy even if it didn't recur.
Write your synthesis to /tmp/session-review-batch/synthesis.md
Format as Markdown. Be specific — reference which sessions showed which patterns. Be concise — this is a summary, not a repetition of individual findings. Do not dispatch or invoke any subagents.
Read /tmp/session-review-batch/synthesis.md and present the full synthesis to the user.
If any individual session findings are particularly interesting, mention that the user can find per-session details in /tmp/session-review-batch/findings-N.md.