Home / All Skills / Workflow & Core Tools / retro
English (EN)简体中文 (ZH)日本語 (JA)
Workflow & Core Tools · mattpocock/skills

retro

Conduct a retrospective on a coding session.

L2 Local · Restricted workspaceV1 Commit-PinnedWrite FileDual-pass Audit
Author mattpocockGitHub mattpocock/skills ↗Stars 257659License MIT（GitHub API spdx MIT；仓库根 LICENSE: Copyright (c) 2026 Matt Pocock）Commit 3cca18b368
Control levels describe skill capabilities and suggested control measures; they do not represent quality or safety certifications, nor do they dictate usability. Usage decisions rest entirely with users and administrators. Different agent hosts apply varying constraints; do not assume hosts provide automatic protection. Third-party skill copyrights belong to their original authors. Use at your own risk.

1How It Works & Mechanism

The retrospective target is the coding agent's runtime environment rather than the code: the objective is to propose improvements to the agent environment to optimize future executions.

skills/in-progress/retro/SKILL.md
The user has asked for a **retrospective**. You are suggesting improvements to the coding agent's **environment** to improve future runs.
Completely different from code-review (reviewing diffs): retro treats the session as an observable sample and outputs environment improvement candidates.

Delegates style standards first: the first step invokes writing-for-agents (the repository's model-invoked style guide skill) as the writing style for improvement recommendations.

skills/in-progress/retro/SKILL.md
Call the Skill tool with `writing-for-agents` for the writing style guide.
writing-for-agents is located in skills/productivity/ and is model-invoked (no policy block in openai.yaml); it is compliant for retro (user-invoked) to call it.

Evidence source = main session source/local session logs: reads user-specified session records (may search session logs locally); defaults to the current session if unspecified.

skills/in-progress/retro/SKILL.md
Read the primary sources for the session the user specifies. This may mean searching through session logs on this machine. If the user doesn't specify a session, default to the current one.
This is the skill's sole data ingestion action: local reading with no network access.

Scans seven categories of improvement signals: navigation / automated checks / coding standards / global AGENTS.md / tool economy / no-ops / information access, each with a 'when to use' criterion.

skills/in-progress/retro/SKILL.md
**Navigation**: how easy was it for the agent to find the right files? Are there hidden dependencies between files? Would a **navigation pointer** make it easier? _Use when_ the session took a long time to find a piece of information.
The other six categories follow the same pattern: Automated checks (_Use when_ the agent made a mistake that could have been caught by an automated check), Coding standards (when missed by reviewer agents), Global AGENTS.md (when the file grows too large), Tool economy (during expensive tool invocations), No-ops (when steering files are bloated yet ineffective), Information access (when critical information is inaccessible).

Outputs are presented to the user ranked by severity, accompanied by an 'implementation vs. review' responsibility model: the implementation agent has the highest context pressure, while the review agent has the lowest → coding standards should be enforced on the review side rather than the implementation side.

skills/in-progress/retro/SKILL.md
The review agent has the least context pressure - it receives a diff, so no exploration needed. It often does not need to write code or debug. This means that the review agent should be responsible for imposing coding standards, not the implementation agent.
File responsibility layering (Reference > Files): CLAUDE.md/AGENTS.md serve only as navigation pointers, CODING_STANDARDS.md is read during reviews (navigation pointers added if >1000 lines), docs serve as reference files, and skills are used for documentation or user commands brought into context.

2Core Capabilities

01Conduct a retrospective on a specified or current coding session by reading local session logs for evidence
02Scan for seven categories of environment improvement candidates (navigation / automated checks / coding standards / AGENTS.md / tool economy / no-ops / information access)
03Output improvement recommendations ranked by severity
04Reuse the writing-for-agents style guide to standardize recommendation text
05Provide a repository file hierarchy responsibility model (respective roles of CLAUDE.md, CODING_STANDARDS, docs, and skills)

4Recommended Governance & Evidence

Suggested Controls for Safe Operation
For Agent Runtime
Restrict working directory
For Administrators
Retain operation logs
Evidence & Triggers (File Paths & Excerpts)
  • 本地文件: Operate on local project files or workspace artifacts
Audit Notes: Only reads local sessions, files, and history, with no outbound network traffic, no credential access, and no file-write operations (recommendations are presented to the user as a list for manual application). Note: session logs may contain secrets or PII, which should be redacted when regurgitated (the skill does not explicitly require this); behavior may vary due to its STUB/experimental status and dependence on the host's log format.

5Second-Pass Verification

  • [ok] 七类改进候选 + 每类 Use-when 判据 — SKILL.md Step 3 七类全列（Navigation/Automated checks/Coding standards/Global AGENTS.md/Tool economy/No-ops/Information access），各带 _Use when_ 条件。
  • [ok] 读取会话日志（本机） — 'This may mean searching through session logs on this machine. If the user doesn't specify a session, default to the current one.' 原文在。
  • [ok] 调 writing-for-agents 文体指南 — 'Call the Skill tool with `writing-for-agents` for the writing style guide.' 原文在；writing-for-agents 为 model-invoked，调用合规。
  • [ok] 按严重度呈现给用户 — 'Present these candidates to the user, in order of severity.' 原文在。
  • [ok] 无写文件/无网络/无凭证 — 全文 token 扫描零命中写文件类与网络类指令；输出是给用户的候选清单。
  • [discrepancy] README STUB 标注 vs SKILL.md 完整步骤（找茬点） — in-progress/README.md 标 retro 为 'STUB: design notes only, not functional yet'，但同 pin 的 SKILL.md 已是完整可执行步骤（Step 1-4 + Reference 模型）。判断为 README 滞后于 SKILL.md 或作者仍视为设计稿；已在 traits/risks 如实标注，不替作者定性。
  • [ok] 元数据 + beta 状态 — GitHub API MIT/257659/2026-09-04；本地 HEAD == 3cca18b；in-progress README 列 beta。

6Verdict & Scenarios

  • Unique retrospective focus: reviews the agent's runtime environment rather than code, complementing code-review.
  • Seven actionable signals with 'Use-when' criteria, transforming vague retrospectives into a structured scan.
  • Clear responsibility model: context pressure determines where coding standards should be enforced (review side), and file hierarchy concepts (keeping CLAUDE.md minimal) are directly reusable.
  • Restrained output: only provides a ranked list of candidates without modifying the user's steering files directly.
    • Best for: Ideal for developers regularly working with agents (e.g., weekly or after major features): those who want to systematically refine steering files, coding standards, automated checks, and tool configurations, turning friction from sessions into an actionable environment improvement list.
    Not for: Not suitable for code-level review (choose code-review instead); not suitable for environments where the host does not keep readable session logs; not suitable for users expecting the skill to directly edit AGENTS.md or CODING_STANDARDS.md (it only outputs a recommendation list); not suitable for production-critical workflows due to its STUB status.
    Install Agent-Ready
    ① Verified Mirror 2026-09-09
    Option A · Direct DownloadDownload retro.tar.gz
    sha256: d1a796d39173a1c6…
    Option B · Copy Guide for Agent
    install.json
    Agents automatically download and verify mirrors from this JSON guide.
    ② Upstream GitHub
    GitHub Repo ↗
    Pinned to commit 3cca18b368.