agentleFS
zauberzeugzauberzeug/game-of-cards/.claude/skills/retrospective/SKILL.md

retrospective

Skill7 starsChanged 6 months agoMITView on GitHub

Backwards analysis of the last N closed cards — recurring failure modes, generalization candidates, rough velocity feel. Read-only; never files cards. AUTO-INVOKE on "what have we learned", "retro", "any patterns lately", "what keeps going wrong".

73 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use project-stage-detect, 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.
  • CoralGameSkill

    project-stage-detect

    Automatically analyze project state, detect stage, identify gaps, and recommend next steps based on existing artifacts.

    Also in 17 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

  • 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

  • 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

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

The file

We split this skill into the rules it gives an agent

4 rules. 0 are common, 1 is distinctive, and 3 are very specific to this project.

  • 0

    Common

    Most of the 62 edited versions we read keep these.

  • 1

    Distinctive

    This author's own choices. Worth a look.

  • 3

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    If any ! block below shows goc: command not found, Permission for this action has been denied, or no such file or directory: …

    line 15
  2. Very specific:

    Read the closure entries in log.md for each card.

    line 87
  3. Very specific:

    Do NOT call Skill(create-card).

    line 123
  4. Distinctive:

    No benchmark is implied — this is a feel indicator, not a sprint commitment.

    line 159

“Common” is measured: kept by at least 6 in 10 of the 62 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

When to invoke

Invoke when the user says "what have we learned", "review recent work", "any patterns lately", "look back", "retrospective", "retro", "what's our velocity", or "what keeps going wrong". May propose Skill(create-card) invocations for generalization candidates but never files them.

Preflight

If any ! block below shows goc: command not found, Permission for this action has been denied, or no such file or directory: .game-of-cards/deck/, stop and invoke Skill(kickoff) first. Kickoff finds the missing setup step (CLI not installed, deck not scaffolded); a denied goc call goes to Skill(claude-kickoff) on Claude Code. Re-invoke this skill only after kickoff completes.

Context

!b=.claude/skills/_goc-bootstrap.sh; if [ -f $b ]; then sh $b --closed-since 90d --json; else goc --closed-since 90d --json; fi 2>&1 | head -100

Retrospective

Scrum's Sprint Retrospective (Schwaber & Sutherland) applied to the full card history. This skill looks backwards at completed work — something no other skill does. Skill(audit-deck) hunts NEW defects in the codebase; this skill reads the last N log.md closure entries, clusters by tag, and surfaces patterns: which failure modes recur, which closures mention the same root cause, which clusters are candidates for a generalization card.

Read-only — this skill inspects history and proposes actions. It never files cards, flips status, or mutates the deck. Any generalization card it identifies must be filed via a subsequent Skill(create-card).

N = $ARGUMENTS (default 10 if not provided).

Step 1 — Gather recent closures

A closure is any card in a terminal status — done, disproved, or superseded. --status done would hide the last two, which are the very cards Step 3 asks about, so scope to all and narrow on the terminal set. (Draft scaffolds ride along under --status all but carry closed_at: null, so the closure filter drops them.) The engine lists closed cards after the live ones, most recently closed first, so the first N that survive the filter are the last N closures.

# Read the last N closed cards (every terminal status), newest first
goc --status all --json 2>/dev/null | \
  python3 -c "
import json, sys
TERMINAL = {'done', 'disproved', 'superseded'}
cards = json.load(sys.stdin)
closed = [c for c in cards if c.get('closed_at') and c.get('status') in TERMINAL]
n = int('$ARGUMENTS'.strip() or '10')
for c in closed[:n]:
    print(json.dumps({'title': c['title'], 'closed_at': c['closed_at'],
                      'status': c.get('status'),
                      'tags': c.get('tags', []), 'contribution': c.get('contribution'),
                      'summary': (c.get('summary') or '')[:80]}))
" 2>/dev/null || true

For each surfaced card, also read its log.md closure entry:

# For each title from above, show its log.md
# (run per-card; replace <title> with actual value)
cat ".game-of-cards/deck/<title>/log.md" 2>/dev/null | tail -20

Step 2 — Cluster by tag

Group the N cards by their tags. For each cluster with ≥2 cards, report:

[<tag>] N cards closed
  - <title> (<closed_at>): <one-line summary>
  ...

Tags with only 1 card get a single [<tag>] 1 card line, no detail.

Step 3 — Surface failure patterns

Read the closure entries in log.md for each card. Look for:

  • Cards closed with disproved or superseded (the status field on each Step 1 line) — what was wrong?
  • Cards whose log.md mentions rework, revert, or "fixed incidentally".
  • Cards that carried a waiting_on overlay (check log.md for the wait being set/cleared) before closing — a long-running wait often signals an under-explored dependency or coordination gap.
  • Cards whose DoD had many unchecked items at closure (if the log notes this).

For each pattern that appears in ≥2 cards, report it:

Pattern: <description>
  Seen in: <title-1>, <title-2>, ...
  Root cause hypothesis: <one line>

Step 4 — Generalization candidates

A generalization card is warranted when:

  • The same root cause appears in ≥3 closed cards (a recurring defect that was patched each time but never fixed architecturally).
  • A cluster of ≥4 cards under the same tag all touched the same module or abstraction (signals a missing abstraction or over-busy module).
  • Two or more cards were closed with "fixed incidentally" (signals the fix belonged in a shared utility).

For each identified candidate, output a proposal only:

Generalization candidate: <proposed-title>
  Evidence: <title-1>, <title-2>, <title-3>
  Pattern: <one-line description of the recurrence>
  → propose Skill(create-card) "<proposed-title>" to capture the meta-fix

Do NOT call Skill(create-card). The human decides which proposals are worth pursuing.

Step 5 — Velocity feel

Count cards closed in the last 7 days, 14 days, and 30 days:

goc --status all --json 2>/dev/null | \
  python3 -c "
import json, sys
from datetime import date, timedelta
TERMINAL = {'done', 'disproved', 'superseded'}
cards = json.load(sys.stdin)
today = date.today()
counts = {7: 0, 14: 0, 30: 0}
for c in cards:
    if c.get('status') not in TERMINAL:
        continue
    cd = c.get('closed_at', '')[:10]
    if not cd:
        continue
    delta = (today - date.fromisoformat(cd)).days
    for window in counts:
        if delta <= window:
            counts[window] += 1
print(f'7d: {counts[7]}  14d: {counts[14]}  30d: {counts[30]}')
" 2>/dev/null || true

Report as:

Velocity: <N> cards/7d · <N> cards/14d · <N> cards/30d

No benchmark is implied — this is a feel indicator, not a sprint commitment. Trend matters more than absolute number.

Output format

## Recent closures (last N)
[tag clusters]

## Failure patterns
[patterns or "None detected"]

## Generalization candidates
[proposals or "None identified"]

## Velocity
Velocity: N cards/7d · N cards/14d · N cards/30d

Total output ≤ 60 lines. Full log analysis stays in the skill's working context; only the surfaced patterns and proposals go to chat.

Cross-references

  • Skill(create-card) — file a generalization card from a candidate proposal.
  • Skill(audit-deck) — hunt for NEW defects in the codebase (forward-looking).
  • Skill(standup) — daily read of in-flight and impeded work.
  • Skill(scan-deck) — full queue browse including closed cards.

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.