Repository navigation
feat(engineering): add the retro skill and session scope for skill feedback - #338
Conversation
Codex Review Summary: ✅ CompletedThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
There was a problem hiding this comment.
🟡 Changes recommended
Two active nit findings remain in plugins/engineering/skills/retro/SKILL.md, involving ambiguous grammar and an imprecise section reference that should be corrected.
2 open issues
What changed in this PR
Adds a manual-only engineering:retro skill for reviewing agent sessions and proposing environment improvements, while extending meta:submit-skill-feedback to analyze explicitly selected sessions. The PR also updates related documentation, metadata, versions, and terminology.
Changes:
- Added
engineering:retro, including session-selection boundaries, finding categories, remedy placement, and severity-ordered reporting. - Extended
meta:submit-skill-feedbackwith multi-session selection, per-session source classification, recurrence handling, and recommendations for repository-owned feedback. - Updated plugin READMEs, marketplace documentation, manifests, versions, and terminology.
| File | Description |
|---|---|
| README.md | Lists the new retro skill and documents multi-session feedback capture. |
| plugins/meta/skills/submit-skill-feedback/SKILL.md | Adds session selection, source classification, recurrence handling, and recommendations to use engineering:retro. |
| plugins/meta/README.md | Documents the expanded feedback workflow and session scope. |
| plugins/meta/plugin.json | Bumps the meta plugin version to 1.3.0. |
| plugins/meta/.claude-plugin/plugin.json | Synchronizes the Claude extension version to 1.3.0. |
| plugins/engineering/skills/retro/SKILL.md | Adds the session retrospective workflow, environment-improvement guidance, and reporting rules. |
| plugins/engineering/skills/retro/agents/openai.yaml | Defines Codex display metadata and manual-only invocation policy for the retro skill. |
| plugins/engineering/skills/grill-me/SKILL.md | Replaces the avoided “harness” terminology with “agent.” |
| plugins/engineering/README.md | Documents the retro workflow and adds it to the engineering skill list. |
| plugins/engineering/plugin.json | Bumps the engineering plugin version to 3.3.0 and updates its description. |
| plugins/engineering/.claude-plugin/plugin.json | Synchronizes the Claude extension version and description with the engineering plugin. |
Review effort: Lite
Adapt upstream retro as a manual-only review of agent sessions that reports environment improvements and changes nothing. It reads only the running agent's own session store unless the user names a path, and it hands findings owned by a marketplace skill off to meta:submit-skill-feedback.
709ab29 to
9cbabff
Compare
There was a problem hiding this comment.
🟡 Changes recommended
A critical active issue remains because the guidance does not explicitly prohibit merging sessions across public and private source classifications, risking boundary-crossing evidence.
1 open issue
2 resolved since last review
Review effort: Lite
Add a session-scope argument with the same session-store boundary as engineering:retro, classify each session's source separately, and recommend engineering:retro for items the repository or the agent's environment owns.
9cbabff to
f784229
Compare
There was a problem hiding this comment.
🔵 Needs a closer look
Approval is deferred because the new session-selection and source-boundary workflows cannot be runtime-tested here, leaving safe historical-session selection, redaction, and public/private destination separation unverified.
No open issues 🎉
1 resolved since last review
Review effort: Lite


The upstream sync in #328 deferred
retrountil the maintainer decided the shape of a similar skill for this repository. That decision is made: a session retrospective for the agent's environment goes inengineering, andmeta:submit-skill-feedbackstays the separate workflow for feedback on this marketplace's skills. This change adds the first and connects the two.engineering:retroA manual-only skill adapted from Matt Pocock's
retro. It reviews agent sessions for moments where a session lost time, made a mistake, or lacked a fact. It then reports findings in severity order, each with a proposed change to the agent's environment: a navigation pointer, an automated check, a review rule, a smaller always-loaded instruction file, a cheaper tool call, or new access to information. It changes nothing; the user decides which findings to act on.Differences from upstream:
REVIEW.mdor## Code Review Rulessection replaces upstream'sCODING_STANDARDS.md.AGENTS.mdto navigation pointers. Here an always-loaded file also keeps what the implementing agent must know before it acts and no check or reviewer can enforce.metais installed,retrorecommendsmeta:submit-skill-feedback.plugins/AGENTS.md.meta:submit-skill-feedbackretro. It classifies the source of each session as public or private separately, and merges items across sessions only when their classifications match.engineering:retrowhenengineeringis installed.Both skills are manual-only, so each names the other only as a recommendation for the user to invoke.
engineering:grill-meOne sentence said "harness" where
wayfindersays "agent", and the rootAGENTS.mdlists "harness" as an alias to avoid.Review notes
add-skillcalls for a pressure test of a skill that preserves a safety boundary. It was not run. A failing run would search this machine for session stores, and no isolated context was available that contains that search.Versions
engineering3.2.1 → 3.3.0, for the new skill.meta1.2.2 → 1.3.0, for the new argument.