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

claude-kickoff

Skill7 starsChanged 6 months agoMITView on GitHub

Claude Code-specific complement to kickoff — Bash(goc:*) permission grant, /plugin install cadence, CLAUDE.md merge prompts. AUTO-INVOKE after Skill(kickoff) completes on a Claude Code repo, or on "finish kickoff for Claude".

  • Reads credentials

The file

We split this skill into the rules it gives an agent

10 rules. About 0 are common, 5 are distinctive, and 5 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈5

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 5

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Run this skill after the generic kickoff returns.

    line 25
  2. Distinctive:

    Read the on-disk signals that determine which stages still have work to do:

    line 34
  3. Very specific:

    Read the project's .claude/settings.json and ~/.claude/settings.json with the Read tool.

    line 44
  4. Distinctive:

    If the user prefers to vendor skills into source control (CI without plugin support, repos that fork or template GoC), add:

    line 77
  5. Distinctive:

    If the expected import is missing or stale, run:

    line 104
  6. Very specific:

    Use goc install --briefing-target <target> instead if Stage 1 has not scaffolded the repo yet.

    line 110
  7. Very specific:

    If the briefing target is CLAUDE.local.md and the file does not yet exist (the generic kickoff already created it as the briefing home, so …

    line 115
  8. Very specific:

    If Stage 0 detected Bash(goc:*) already in either settings.json (or an interactive grant during the generic kickoff's goc install already …

    line 123
  9. Very specific:

    If .game-of-cards/config.yaml records autonomy: as loop, cron or action, first show the user that mode's entry from § Reference: autonomy …

    line 151
  10. Distinctive:

    When the GoC plugin is installed, three hooks fire automatically:

    line 202

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “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 after Skill(kickoff) completes on a Claude Code repo, or when the user says "finish kickoff for Claude", "set up Claude Code permissions", "/plugin update for goc", or initiates Claude-specific GoC setup that the generic kickoff intentionally skipped.

Finish kickoff on Claude Code

The generic kickoff skill is host-agnostic: it introduces GoC, runs the persona dialog, asks where the briefing should live (AGENTS.md, CLAUDE.md, or CLAUDE.local.md), and runs goc install --briefing-target <choice> to scaffold .game-of-cards/. This complement handles everything Claude Code-specific that the generic kickoff intentionally leaves alone:

  • the Bash(goc:*) permission grant for future sessions,
  • the /plugin install (and /plugin marketplace update) cadence,
  • the minimal CLAUDE.md @<chosen-file> pointer when the briefing lives in AGENTS.md or CLAUDE.local.md,
  • writing GoC-managed entries into .claude/settings.json.

Run this skill after the generic kickoff returns. Re-running is safe: every stage detects existing on-disk state before asking.

Updating the plugin? Run /plugin marketplace update zauberzeug/game-of-cards before /plugin install — Claude Code reuses its local marketplace clone and does not refresh it automatically. Skipping this step silently installs the old bytes.

Stage 0 — state detection sweep

Read the on-disk signals that determine which stages still have work to do:

grep -l '<!-- BEGIN GOC' AGENTS.md CLAUDE.md CLAUDE.local.md 2>/dev/null

The first matching file (or the only matching file) is the briefing target — the home the generic kickoff chose. Hold that path; Stage 2 needs it.

Read the project's .claude/settings.json and ~/.claude/settings.json with the Read tool. Note whether "Bash(goc:*)" appears in either permissions.allow array. Note whether .claude/hooks/ already contains the GoC-managed hook scripts (the plugin path supplies them; if they're absent the user is on the vendored path).

Hold all of these flags in mind through the rest of the flow. The generic kickoff has already run, so .game-of-cards/deck/ exists, goc is on PATH, and the briefing lives in exactly one of the three candidates — this skill does not re-check those.


Stage 1 — /plugin install cadence note

Skip this stage if Claude Code's plugin runtime already exposes the GoC skill set in this session (the user reached this skill via Skill(claude-kickoff), which means the plugin is registered).

Otherwise, deliver this note to the user:

The Game of Cards Claude Code plugin ships skills, runtime hooks, and the bundled goc CLI. To install (one-time per machine):

/plugin marketplace add zauberzeug/game-of-cards
/plugin install game-of-cards@game-of-cards

When updating later, run /plugin marketplace update zauberzeug/game-of-cards before /plugin install — Claude Code reuses its local marketplace clone and does not refresh it automatically.

If the user prefers to vendor skills into source control (CI without plugin support, repos that fork or template GoC), add:

Alternatively, install via pipx install game-of-cards and run goc install --local-skills to vendor skills, hooks, and settings into .claude/. The plugin-bundled goc refuses --local-skills; the pipx CLI is the only path that writes a vendored .claude/skills/ tree.


Stage 2 — verify CLAUDE.md loads the briefing

Claude Code only auto-loads CLAUDE.md (and CLAUDE.local.md next to it). When the briefing lives elsewhere, Claude needs a one-line @<file> import in CLAUDE.md to transitively load it.

goc install and goc upgrade own this wiring. Use the briefing target path detected in Stage 0:

  • target = CLAUDE.md — Claude already sees the full briefing inline. Skip this stage; do not write a separate import.
  • target = AGENTS.md — CLAUDE.md should contain @AGENTS.md (or a marker-bounded GoC import block if pre-existing user content had to be preserved).
  • target = CLAUDE.local.md — CLAUDE.md should contain @CLAUDE.local.md (or the same marker-bounded import block form).

If the expected import is missing or stale, run:

goc upgrade --briefing-target <target>

Use goc install --briefing-target <target> instead if Stage 1 has not scaffolded the repo yet. Do not hand-edit the import as the normal path; the CLI handles fresh one-line CLAUDE.md files and preserves existing user CLAUDE.md content with its GoC import marker.

If the briefing target is CLAUDE.local.md and the file does not yet exist (the generic kickoff already created it as the briefing home, so this is rare), no extra stub is needed — the file holds the briefing.


Stage 3 — persist Bash(goc:*) permission

If Stage 0 detected Bash(goc:*) already in either settings.json (or an interactive grant during the generic kickoff's goc install already wrote it), this stage is a no-op.

Otherwise, write the project's .claude/settings.json with:

{
  "permissions": {
    "allow": ["Bash(goc:*)"]
  }
}

Merge with any existing .claude/settings.json content rather than overwriting it. After the file is written, tell the user:

Bash(goc:*) is now permitted for this project's future Claude Code sessions. The current session continues to work without a restart.

This is the LAST mutation claude-kickoff makes. Any future enhancement that needs a context-losing session restart must remain after this point so a restart never destroys work already done.


Stage 4 — confirm ready

If .game-of-cards/config.yaml records autonomy: as loop, cron or action, first show the user that mode's entry from § Reference: autonomy recipes — kickoff recorded the choice, nothing has wired it.

Report to the user:

Claude Code-specific kickoff is complete. All GoC skills are live
through the plugin (or the vendored `.claude/skills/` if you opted in).
What should the first card be?

The deck is now live. Skill(create-card), Skill(scan-deck), and all other GoC skills work immediately — no further kickoff needed.


Reference: autonomy recipes

One entry per mode kickoff Stage 6 records. Each recipe starts one pull-card per session and leaves the repetition to the scheduler, so every card gets a fresh context.

  • loop — in a session you keep open, run /loop 30m /pull-card. It fires only while that session stays open, and a recurring loop expires after 7 days; re-issue it to continue.
  • cron — a crontab entry starts one headless session per tick (cron's PATH is minimal; give the absolute path to claude if it is not found):

    */30 * * * * cd /path/to/repo && claude -p "Run Skill(pull-card) once, then exit." --permission-mode acceptEdits --allowedTools "Bash(goc:*)" "Bash(git:*)" >> /tmp/pull-card.log 2>&1

    A headless session cannot answer a permission prompt, so whatever it is not pre-approved for is denied — add the commands your cards need, such as the test runner.

  • action — a scheduled GitHub Actions workflow runs anthropics/claude-code-action@v1 with the same prompt and a CLAUDE_CODE_OAUTH_TOKEN or ANTHROPIC_API_KEY secret. The runner needs the skills (vendored .claude/skills/, or the plugin through the action's plugin_marketplaces and plugins inputs) and a goc it can run. GoC's own pull-card.yml is a worked example: it skips the agent while no card is pullable and re-dispatches itself while cards remain.

Reference: runtime hooks

When the GoC plugin is installed, three hooks fire automatically:

Hook eventScriptPurpose
SessionStartdeck_session_startPrints active-card reminder at session start; silent when no cards are in-flight.
UserPromptSubmitdeck_prompt_routerDetects work-initiating prompts; injects a deck-first reminder into Claude's view.
Stoppattern_generalization_checkAfter code-mutating turns, asks the agent to self-assess whether the change warrants a generalization card. Off by default — it blocks every code-mutating turn to inject the reminder. Opt-in: set hooks.pattern_generalization_check: true in .game-of-cards/config.yaml.

The hooks are optional. Repos without the plugin still get full GoC functionality through the goc CLI and AGENTS.md guidance. Other agent runtimes (Codex, OpenCode, Cursor) use their own hook systems and do not share this registration.

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.