csk-help
Skill1 starsChanged 4 months agoMITView on GitHub
Context-aware CSK guide for locating the current project phase, reconciling durable tasks and Git state, and recommending exactly one safe next action from the repository workflow. Use when the user asks what to do next, where the project stands, how skills fit together, or which skill should handle a finding.
In this repository18 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- 1-csk-initSkill
- 2-csk-write-specSkill
- 3-csk-architectureSkill
- 4-csk-frontendSkill
- 5-csk-backendSkill
- 6-csk-qaSkill
- 7-csk-deploySkill
- audit-plan-loopSkill
- claude-skill-proxy-syncSkill
- create-worker-worktreeSkill
- csk-adopt-plan-scaffoldSkill
- csk-refineSkill
- csk-startSkill
- finish-branchSkill
- pull-main-ffSkill
- review-loopSkill
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
- Very specific: line 15
Read .csk/project-state.json through /csk-start state validation.
- Very specific: line 16
Read .claude/rules/workflow-state.md.
- Very specific: line 19
Read tasks/INDEX.md, live Git status/branch, the adapter-resolved PRD, master feature, architecture, feature index, and only applicable …
- Very specific: line 21
Read docs/engineering-principles.md and .claude/rules/loop-policy.md when quality, architecture, QA, release, planning, or review is in …
- Distinctive: line 26
Do not contact another repository or rely on chat memory.
- Very specific: line 55
Keep a feature in In Review while remediation is open.
- Distinctive: line 61
Use this compact structure:
- Distinctive: line 104
Never claim that a task, feature transition, commit, push, merge, pull request, tag, release, or deployment happened without direct …
- Distinctive: line 106
Never install repo skills globally.
- Distinctive: line 107
Never bypass onboarding, workflow entry contracts, or explicit external-action approval merely because the next command appears obvious.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 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
Project Help Guide
Answer the user's specific question first, then show where the repository is and recommend exactly one primary next action.
Read first
- Read
.csk/project-state.jsonthrough/csk-startstate validation. - Read
.claude/rules/workflow-state.md; it is the transition and routing SSOT. - Resolve optional local paths through
.claude/rules/adapter-config.mdand repository-rootcsk.config.json. - Read
tasks/INDEX.md, live Git status/branch, the adapter-resolved PRD, master feature, architecture, feature index, and only applicable feature specs. - Read
docs/engineering-principles.mdand.claude/rules/loop-policy.mdwhen quality, architecture, QA, release, planning, or review is in scope. - Inventory code with
rg --filesorgit ls-files; assume no language, framework, package manager, test runner, or deployment target.
Do not contact another repository or rely on chat memory. Repository and Git state win over a task's advisory branch or any prior recommendation.
Determine the next action
Apply the routing precedence and entry contracts from
.claude/rules/workflow-state.md. Stop at the first applicable gate:
- Invalid onboarding state: repair the state; do no productive work.
- Adoption pending or blocked: run
/csk-adopt-plan-scaffold. - Actionable durable work: run
/csk-start, reconcile it, and continue the selected item with its recorded runtime-neutral resume skill. - Dirty or ambiguous Git state: inspect and attribute it before switching scope.
- Incomplete project initialization: run
/1-csk-initwith the user's idea. - Missing master product contract: run
/csk-refine --master. - Unowned requested scope: run
/2-csk-write-spec. - Existing feature: use its authoritative status and the workflow-state rule.
- Verified work on a non-default branch: run
/finish-branchbefore deployment or closure.
For QA or review findings, route the fix to the component owner:
- user-facing surface or interaction:
/4-csk-frontend; - core behavior, service, data, automation, pipeline, or integration:
/5-csk-backend; - architecture mismatch:
/3-csk-architectureor/csk-refinebefore code; - test-only gap:
/6-csk-qa; - release-only gap:
/7-csk-deployafter the feature returns toApproved.
Keep a feature in In Review while remediation is open. Create one durable task
only if the remediation will survive the current session, is deferred, blocked,
or is a verified interruption point.
Response format
Use this compact structure:
Project Status
Onboarding: <mode/status and actual distribution scope>
Product: <initialized/incomplete>
Architecture: <open/chosen/adopted>
Git: <branch and clean/dirty>
Tasks: <open/blocked counts>
Feature: <PROJ-X — authoritative status, when applicable>
Next step: Run /<skill> to <specific action>.
Why: <one or two evidence-based sentences>Mention a blocker or state drift before the recommendation. Do not list several competing next commands unless the user explicitly asks for options.
Common questions
- Why this order: the numbered steps follow the classic engineering cycle —
requirements, then specification, then architecture, then implementation,
then QA against the original acceptance criteria, then release. See
"Why this order?" in
README.mdfor the short rationale. - Available workflow skills: numbered skills 1–7 plus
/csk-start,/csk-adopt-plan-scaffold,/csk-refine,/csk-help,/finish-branch,/create-worker-worktree,/pull-main-ff,/audit-plan-loop,/review-loop, and/claude-skill-proxy-sync. - Git for beginners and parallel agent sessions:
docs/git-basics.mdexplains the needed terms, the three levels of parallel work, and the orchestrator/worker worktree cycle. - Plan/review hardening: explain
/audit-plan-loop,/review-loop, and.claude/rules/loop-policy.md. - Codex compatibility:
.claude/skillsis canonical and repo-local; ready Codex proxies live in.codex/skills;/claude-skill-proxy-syncrefreshes them. - Deterministic versus non-deterministic: deterministic work has repeatable rules and mechanically checkable outputs. Its opposite is non-deterministic; in this workflow that usually means semantic judgment can legitimately depend on context, evidence, and tradeoffs rather than one fixed algorithmic answer.
Safety
- Help and status inspection are read-only.
- Never claim that a task, feature transition, commit, push, merge, pull request, tag, release, or deployment happened without direct evidence.
- Never install repo skills globally.
- Never bypass onboarding, workflow entry contracts, or explicit external-action approval merely because the next command appears obvious.
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.

