agentleFS
andymaiandymai/claude-agent-team/skills/retrospective/SKILL.md

retrospective

Skill4 starsChanged yesterdayMITView on GitHub

Review a coding session and propose changes to the agent's environment (checks, rules files, skills, tools) that make the next session better.

72 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use dev-story, team-level and 10 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • lilangMaxSkill

    dev-story

    Read a story file and implement it. Loads the full context (story, GDD requirement, ADR guidelines, control manifest), routes to the right programmer agent for the system and engine, implements the code and test, and confirms each acceptance criterion. The core implementation skill — run after /story-readiness, before /code-review and /story-done.

    Also in 12 projects that use this

  • CoralGameSkill

    team-level

    Orchestrate level design team: level-designer + narrative-director + world-builder + art-director + systems-designer + qa-tester for complete area/level creation.

    Also in 12 projects that use this

  • CoralGameSkill

    team-narrative

    Orchestrate the narrative team: coordinates narrative-director, writer, world-builder, and level-designer to create cohesive story content, world lore, and narrative-driven level design.

    Also in 12 projects that use this

  • lilangMaxSkill

    create-stories

    Break a single epic into implementable story files. Reads the epic, its GDD, governing ADRs, and control manifest. Each story embeds its GDD requirement TR-ID, ADR guidance, acceptance criteria, story type, and test evidence path. Run after /create-epics for each epic.

    Also in 11 projects that use this

  • +7 more from lilangMax
  • +1 more from CoralGame

The file

We split this skill into the rules it gives an agent

8 rules. 0 are common, 7 are distinctive, and 1 is very specific to this project.

  • 0

    Common

    Most of the 60 edited versions we read keep these.

  • 7

    Distinctive

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    If the line above is empty, review the current conversation.

    line 24
  2. Distinctive:

    Read the project's rules files (CLAUDE.md, AGENTS.md) and find the project's check command and its CI workflow.

    line 28
  3. Distinctive:

    Look for each of these seven types.

    line 34
  4. Distinctive:

    Put each change in the first layer of this list that can hold it:

    line 50
  5. Distinctive:

    A rule for the reviewer: use this layer for a rule that needs judgment across files.

    line 55
  6. Distinctive:

    A skill or a reference file behind a pointer: use this layer for a procedure that only some sessions need.

    line 57
  7. Distinctive:

    A line in an always-loaded rules file: use this layer last. Each line costs context in every session.

    line 59
  8. Very specific:

    When the user selects changes, apply only those changes.

    line 82

“Common” is measured: kept by at least 6 in 10 of the 60 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Retrospective

Find where the session lost time or made a mistake, and propose a change to the environment that prevents each one. The environment is everything that the agent works with: automated checks, rules files, skills, agents, tools, and access to information. You propose changes. You apply a change only after the user selects it.

The session to review, if the user named one: $ARGUMENTS

1. Load the writing rules

Invoke the team:agent-docs skill with the Skill tool. Each change that you propose to a rules file, a skill, or an agent must follow those rules.

2. Read the session

If the line above is empty, review the current conversation. If the user named a different session and you do not have its transcript, ask for the path. Do not guess which transcript the user means.

Read the project's rules files (CLAUDE.md, AGENTS.md) and find the project's check command and its CI workflow. You need them to tell a new check from a check that exists and does not run.

3. Find the candidates

Look for each of these seven types. For each candidate, record the moment in the session that shows the cost: the tool call, the error, or the user's correction. Quote it. A candidate with no such moment is a guess, so drop it.

TypeThe session showsPossible change
NavigationMany searches to find one file or one factA pointer in a rules file to that location
Automated checkA mistake that a linter, a type check, a test, or a hook can catchA new check, or a check that exists and is not wired into CI or a hook
Review ruleThe reviewer did not catch a mistake that needs judgmentA new or clearer rule for the reviewer
Rules fileAn instruction that the agent did not follow, or a rules file so long that instructions get lostMove the instruction to a check or to the reviewer, or delete it
Tool costA tool call that returned much more than the agent neededA narrower command, a script, or a filter
No-opAn instruction that does not change what the agent doesDelete it
Information accessThe agent did not have a fact that it needed (a log, a service, a doc)Read access to that source

4. Select the layer for each change

Put each change in the first layer of this list that can hold it:

  1. An automated check: a lint rule, a type, a test, a pre-commit hook, a CI job. Use this layer for each mechanical rule: a banned call, an import shape, a file location, a naming pattern. A check cannot forget, and it costs no context.
  2. A rule for the reviewer: use this layer for a rule that needs judgment across files. The reviewer reads a diff, so it has the most room for rules.
  3. A skill or a reference file behind a pointer: use this layer for a procedure that only some sessions need.
  4. A line in an always-loaded rules file: use this layer last. Each line costs context in every session.

Example. The agent used sed on a source file, and the rules file bans that.

  • Bad: add "IMPORTANT: never use sed" to the rules file a second time.
  • Good: add a pre-tool hook that rejects a sed -i command on a tracked file, and delete the line from the rules file.

5. Present the candidates

List the candidates with the largest cost first. For each one, give:

  • Observation: the quoted moment from the session, and how many times it occurred.
  • Change: the exact file, and the text or the check to add, change, or delete.
  • Layer: which of the four layers, and why a higher layer cannot hold it.
  • Cost: what the change adds (context, run time, maintenance).

End with the candidates that you dropped and why. Then ask which changes the user wants, and end your turn.

6. Apply what the user selects

When the user selects changes, apply only those changes. Follow the team:agent-docs rules for each rules file, skill, or agent that you change. Run the project's checks after the last change. Then report each file that you changed and the result of the checks. If the user selects nothing, change nothing.

Final self-check

  • Each candidate quotes a moment from the session.
  • Each mechanical rule goes to an automated check, not to prose.
  • No candidate adds a line to an always-loaded file when a lower-cost layer can hold it.
  • Each proposed change names an exact file and exact text.
  • You changed no file before the user selected a change, and you changed only what the user selected.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.