deck
Skill7 starsChanged 6 months agoMITView on GitHub
Front door for the deck — the operating substrate for ALL persistent work. AUTO-INVOKE when the user references the deck/methodology/workflow, or at session start as a reminder that persistent work flows through Skill(create-card) → Skill(advance-card) → Skill(finish-card).
7 other projects use a version of this.
Adopted by
Most recently taken up by
- serhiileniv/skillstheir own version, first seen Nov 8, 2026
- serhiileniv/ccstheir own version, first seen Nov 8, 2026
- roaldnefs/skillstheir own version, first seen Sep 29, 2026
- JustOneMoreTry/NoMoreHigstheir own version, first seen Aug 4, 2026
- iuiaoin/agent-skillstheir own version, first seen Jun 8, 2026
- foogunlana/skillstheir own version, first seen May 23, 2026
Other versions7 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 7
- iuiaoin/agent-skills★ 16same name
- drobotmax/claude-skill-deck★ 1same name
- foogunlana/skills★ 1same name
- roaldnefs/skills★ 1same name
- JustOneMoreTry/NoMoreHigs★ 0same name
- serhiileniv/ccs★ 0same name
- serhiileniv/skills★ 0same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
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
- finish-cardSkill
- kickoffSkill
- next-cardSkill
- 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. 0 are common, 1 is distinctive, and 6 are very specific to this project.
0
Common
Most of the 7 edited versions we read keep these.
1
Distinctive
This author's own choices. Worth a look.
6
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 39
Autonomous mode: No human is steering. Skill(pull-card) runs on a loop or schedule, draining the human_gate: none queue.
- Distinctive: line 55
Session mode (silent runtime): When the user asks for persistent work — "rename the button to Export", "add a CSV export", "fix the auth …
- Very specific: line 59
Check the deck — does a card already cover this?
- Very specific: line 60
If not, file a card: goc new <kebab-title> and edit the body.
- Very specific: line 70
Autonomous mode: Before recommending or claiming new work, agents check goc --status active and treat listed cards as already-claimed soft …
- Very specific: line 77
Andon-cord path: When a human asks "what's up?" / "where do you need me?", surface parked cards (oldest-first, with ## Decision required …
- Very specific: line 97
Run goc --help for the full verb list.
“Common” is measured: kept by at least 6 in 10 of the 7 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.
Files
When to invoke
Invoke when the user references the deck, the methodology, or the workflow — or at session start as a reminder that every user request producing persistent work must flow through Skill(create-card) → Skill(advance-card) → Skill(finish-card). XP-style story cards on a kanban board, designed for AI-agent collaborators.
The Deck
deck/ is the project's work-tracking surface. Each subdirectory is one
card: a unit of work — bug, story, epic, idea, derivation gap,
doc-drift catch — with frontmatter-driven status on a kanban board.
The deck is the runtime of the human's intent, not a ceremony they opt into: rigor (machine-checkable DoDs, audit-trail commits, supersession chains) is separated from visibility (zero by default). The user sees the code, not the bookkeeping, unless they ask.
Background and rationale live in reference.md — a sibling file
in this skill's directory. Read the named section only when the
situation actually calls for it:
| Situation | reference.md section |
| Why XP/Scrum/Kanban discipline is load-bearing for agent swarms | Heritage & philosophy |
| Why card operations stay invisible; full operating-mode prose | Game of Cards as the runtime |
| Why the methodology is named Game of Cards | The worldview |
| New evidence after a card closed | Closure is not frozenness |
How status / waiting_on / human_gate compose | Status, overlay, and gate are orthogonal axes |
Pre-2026-05-01 placeholder DoDs; v1 find-todo/work-todo names | Migration legacy |
Three operating modes
- Session mode. A human is talking. Claude runs the card pipeline silently while answering; card operations are NEVER announced unless the user explicitly asks to see the deck.
- Autonomous mode. No human is steering.
Skill(pull-card)runs on a loop or schedule, draining thehuman_gate: nonequeue;Skill(audit-deck)keeps the queue fed on a slower cadence. - Andon-cord mode. A human is unblocking the line:
pull-cardparked a card behind a raised gate;Skill(scan-deck)surfaces parked cards,Skill(decide-card)records the answer and lowers the gate, the next pull resumes the work.
Multiple sessions work cards in parallel. status: active is the
soft lock, but by default a claim stays in your clone, so two clones
can both claim one card; workflow.claim_push: true pushes the claim
and refuses a later conflicting one (reference.md § Game of Cards
as the runtime).
What this looks like in practice
Session mode (silent runtime). When the user asks for persistent work — "rename the button to Export", "add a CSV export", "fix the auth bug" — run the GoC pipeline silently:
- Check the deck — does a card already cover this? (
goc --tag <topic>) - If not, file a card:
goc new <kebab-title>and edit the body. - Claim it:
goc status <title> active. - Implement.
- Close:
goc done <title>, then commit the work and closure.
Card operations are NEVER announced — the user sees the code, not the bookkeeping. No-card exceptions (zero work, no card): exploration ("explain X", "why is Y this way?"), one-shot tooling ("git status", "rebase this"), course-corrections inside an active card.
Autonomous mode. Before recommending or claiming new work, agents
check goc --status active and treat listed cards as already-claimed
soft locks. The pull principle is what makes this safe: work isn't
pushed at agents on a timer; agents pull on their own terms, filtered
to gate=none. The human steers by curating WHAT'S in the queue and
at what gate.
Andon-cord path. When a human asks "what's up?" / "where do you
need me?", surface parked cards (oldest-first, with ## Decision
required body section preview). Decision recorded → gate lowered with
goc decide <title> --decision "..." --because "..." → next pull-card
claims and implements per the recorded decision.
Daily CLI verbs
| Verb | What it does |
goc | Show the open queue (value-sorted). |
goc --board | Multi-column kanban view. A ⏳ after an open card's [contribution] marker flags any of three signals: human_gate != none (parked for a human — not pullable), an active impediment overlay (waiting_on / future waiting_until — not pullable), or an advisory derived dependency-block (a non-terminal advanced_by prereq — still pullable, just flagged as "has an open upstream"). Only the first two hide a card from pull-card / next-card / goc --ready; a dependency-block does not. So ⏳ ⇏ unpullable — check the cause. |
goc --closed-since 7d | Recently closed cards (any terminal status). |
goc new <title> | Scaffold a new card under .game-of-cards/deck/<title>/. |
goc status <title> <state> | Flip status (open/active/disproved/superseded). |
goc wait <title> --reason <r> [--until <date>] | Set the impediment overlay for exogenous waits; --clear to drop it. |
goc done <title> | Close + DoD enforcement (no auto-commit). |
goc decide <title> --decision X --because Y | Lower gate from decision/session → none. |
goc validate | Validate every card's frontmatter (pre-commit-friendly). |
Run goc --help for the full verb list. Schema and enum constraints
surface in goc validate error messages. Project-local tag extensions
live in .game-of-cards/canonical-tags.md.
The deck layout
deck/
log.md # deck-level journal (sprint notes, schema bumps)
<title>/ # one dir per card; never moves on state change
README.md # frontmatter + dashboard body — latest knowledge + current state
log.md # append-only journal — history, details, decisions, flow
reproduce.py # OPTIONAL — declared in DoD when present
[other validation scripts]One title, one directory, forever. Status changes mutate frontmatter, not paths.
Two files, two edit disciplines: the README is the dashboard
(rewritten in place as understanding evolves so a cold reader sees
only what is true now); log.md is the append-only journal
(history, details, decisions, and flow preserved verbatim, never
rewritten). See Skill(card-schema)'s "What goes where" subsection
for the routing rule.
Lifecycle
open ──→ active ──→ done (terminal, DoD-100% required)
│
├──→ disproved (terminal; body documents rebuttal)
└──→ superseded (terminal; typed `superseded_by` → successor,
rationale appended to log.md)open is the queue. active claims the work. done, disproved,
and superseded are all terminal — none deletes the directory; the
forensic record stays.
A card that is temporarily not workable does not move to a
separate status; the three-axis model expresses it as either derived
dependency-readiness (an advances prereq is still open and shows as
"awaiting" in the queue, advisory only) or the stored impediment
overlay (waiting_on + optional waiting_until, set via goc wait,
hides the card from the ready queue until cleared or the date
elapses). See Skill(card-schema) "Three-axis 'stuck' model" for the
full contract. Closed cards stay amendable, and the three axes are
orthogonal — reference.md § Closure is not frozenness and § Status,
overlay, and gate are orthogonal axes.
The action skills
One skill per job; compose, don't bundle.
Skill(scan-deck)— browse the deck. Triage default surfaces parked cards (gate ≠ none); also filtered queues, kanban board, JSON dump, and the "decisions to make" Q&A flow. Read-only by default; the Q&A mode callsSkill(decide-card)per answer.Skill(next-card)— auto-pick the highest-leverage opengate=nonecard to work on next. Read-only; does NOT flip status.Skill(create-card)— file a new card with proper frontmatter and a DoD scaffold (reproduce.py is authored by hand for bug-class cards, not scaffolded by the tool).Skill(advance-card)— flip status (open→active, *→open, *→disproved, *→superseded) and manage thewaiting_onimpediment overlay (goc wait). Wrapsgoc statusandgoc wait. Status + overlay only —human_gateisdecide-card's responsibility.Skill(decide-card)— the human's Andon-cord lowering action. Records<decision> + <because>on a parked card and flips gatedecision/session→none. The card comes outopen(one parked whileactiveis released) so the nextpull-cardclaims and implements per the recorded decision.Skill(finish-card)— close a card: tick DoD, append closure log, rungoc done <title>, then run any project-specific post-close or commit handoff defined by the consuming repo's hook.Skill(card-schema)— schema reference (read-only): required fields, enums, canonical tags with predicates, DoD detection, relationship invariants.Skill(refine-deck)— hygiene pass: retag stale cards, prune unverified parks, surface cards citing defunct file:lines, propose new canonical tags.Skill(audit-deck)— discovery: hunt for one previously-undocumented defect / derivation gap / doc drift / missing test. Files viaSkill(create-card).Skill(standup)— daily read: active + impeded cards (those carrying awaiting_onoverlay), closures since yesterday, cards waiting on a human decision gate.Skill(retrospective)— backwards analysis of the last N closed cards: cluster by tag, surface recurring failure modes, propose generalization candidates.
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.

