next-card
Skill7 starsChanged 6 months agoMITView on GitHub
Pick the highest-leverage open card to work on next. Read-only verdict — does NOT flip status (pull-card claims). AUTO-INVOKE on \"what's next\", \"pick something\", \"what should I do\", or autonomous-loop work. Filters to `human_gate: none` for loop safety.
In this repository19 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- site/llms.txtllms.txt
- advance-cardSkill
- audit-deckSkill
- card-schemaSkill
- claude-kickoffSkill
- create-cardSkill
- decide-cardSkill
- deckSkill
- finish-cardSkill
- kickoffSkill
- pull-cardSkill
- refine-deckSkill
- retrospectiveSkill
- scan-deckSkill
- standupSkill
- tune-cadenceSkill
- upgradeSkill
The file
We split this skill into the rules it gives an agent
7 rules. About 0 are common, 2 are distinctive, and 5 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
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
- Very specific: line 16
If any ! block below shows goc: command not found, Permission for this action has been denied, or no such file or directory: …
- Distinctive: line 38
Before recommending, read the active slice above.
- Very specific: line 83
Every card carries human_gate: none | decision | session:
- Very specific: line 109
If every ready candidate is decision or session, the run halts with a one-line summary listing the parked cards.
- Very specific: line 114
Explicit invocation by title: read the named card via goc show <title> and check the gate.
- Distinctive: line 130
Update an empirical / publication-tier claim.
- Very specific: line 132
If any apply, the gate should be at least decision (likely session).
≈ “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.
Files
When to invoke
Invoke when the user says "what's next", "pick something", "work on the queue", "what should I do", "next item", "drain the deck", or initiates autonomous-loop work. Kanban pull principle — work is taken, not pushed.
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 --status active -v; else goc --status active -v; fi 2>&1 | head -20
!b=.claude/skills/_goc-bootstrap.sh; if [ -f $b ]; then sh $b --ready -v${GOC_WORKER:+ --worker "$GOC_WORKER"}; else goc --ready -v${GOC_WORKER:+ --worker "$GOC_WORKER"}; fi 2>&1 || true
Pick the next card
Kanban's pull principle (Anderson): work is taken when capacity is ready, not pushed by a planner. The runner — human or autonomous loop — pulls one card at a time, top of the value-sorted queue, gated by the autonomy ladder so a cron run cannot accidentally land a research-impacting decision unsupervised. The deck is the queue; this skill is the picker.
Recommend, do not claim. Status does NOT flip here — that is
Skill(advance-card)'s contract. The pull is a two-step gesture (pick,
then claim) so the human or /loop can abort between them without
half-state on disk.
Before recommending, read the active slice above. Active cards are claimed soft locks; avoid recommending the same card or adjacent/conflicting work unless the user explicitly asks to continue that active card.
User argument: $ARGUMENTS — if non-empty (a title or area like
an area tag or path prefix), narrow the queue. If empty, scan the full
ready slice (--ready keeps cards with open advances prereqs —
they are pullable with an "awaiting: <prereqs> (you may start)"
advisory line, since advances is a soft "should precede" not a hard
start-gate; hard waits are expressed via waiting_on /
waiting_until).
Selection criteria
goc lists open cards sorted by value desc
then created asc. The top entry is the auto-pick candidate, subject
to the autonomy gate below.
Contribution ladder
high outranks medium outranks low. Tags refine:
contribution: high— wrong algorithm vs. cited literature, silent state corruption, broken public API, default config that contradicts the science. Doc claims that contradict an authoritative source —tags: [documentation]+contribution: highis the high-impact doc-quality slot. Treat these as load-bearing.contribution: medium— tolerance creep, vacuous assertions, tests that pass for the wrong reason, missing guard rails.contribution: low— README pinned-metric stale, docstring documents removed flag, stale references.
Effort / independence / reversibility
- Effort — files touched, area of the project (library code vs application code, test coverage required). Library default changes carry higher cost.
- Independence — fixes whose effect can be verified without first
fixing another card ship before the dependent ones. Read the
body's "Why it matters" notes and any
advanced_by:chain. - Reversibility — prefer fixes with low blast radius. Local demo edits before library changes.
Autonomy gate (the human_gate field)
Every card carries human_gate: none | decision | session:
none— autonomous-loop-safe; cron may auto-pick.decision— needs ONE human go/no-go before work proceeds. The body MUST already carry the framing in a## Decision requiredsection (perSkill(card-schema)). The human resolves asynchronously.session— needs interactive working session. Research-impacting framework derivations, open architectural choices.
Pickability rule (uniform across cron, /loop, explicit-by-title):
none→ recommend.decision→ DO NOT recommend. End the session cleanly with a one-line summary pointing at the parked card and (if available) its body's## Decision requiredrecommendation line. Do not pause an idle agent. No commits, no half-step writes — control returns to the human at their convenience.session→ refuse to auto-pick under autonomous mode; ask for explicit confirmation under interactive mode that "yes, this is the working session for this topic." On no/silence, end likedecision.
Autonomous-mode rule (/loop, cron, or any non-interactive
invocation): walk down --ready and recommend the top.
If every ready candidate is decision or session, the run
halts with a one-line summary listing the parked cards. Better to
ship nothing than to land a research move without a review
checkpoint.
Explicit invocation by title: read the named card via
goc show <title> and check the gate. Apply
the rules above on the specific card.
Reclassify after reading
The human_gate is a hypothesis from the filing round; the body
often reveals the real shape. Before recommending work, re-check
whether the fix would:
- Close (or reopen) a gap in
docs/framework/*.md. - Pick one of two literature-backed mechanism candidates.
- Set a sign / direction convention.
- Introduce a new named primitive.
- Flip a default whose rationale is documented against a paper or axiom.
- Update an empirical / publication-tier claim.
If any apply, the gate should be at least decision (likely
session). Recommend escalation via Skill(advance-card) to fix
the gate, not working it autonomously. The right move is a 2-line
edit to the frontmatter + (if upgrading to decision) writing the
## Decision required section in the body — then the run ends and
the human comes back.
Output
A single recommendation with:
- Slug + one-line subject of the recommended card.
- 2-line rationale — contribution, why it's the highest-leverage open pick, any advanced_by edges that matter.
- Next-step pointer: typically "Run
Skill(advance-card) <title> activeto claim, then work it." If gateddecisionorsession, point at the parked framing and end.
The output is a recommendation, not a plan. The user (or /loop)
picks whether to actually run Skill(advance-card) next.
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
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.

