jf-agentic-coding-starter-kit
AGENTS.md1 starsChanged 4 months agoMITView on GitHub
This starter kit supports both Codex and Claude. CLAUDE.md and the repository rules define the shared workflow; these instructions add the Codex entry point. 1. Inspect the skills and tools currently available. 2. Prefer an applicable skill when one exists. Use deterministic tools for mechanical checks and LLM judgment for semantic, contextual, or non-deterministic decisions. 3. Read CLAUDE.md and relevant .claude/rules/. 4. Run $csk-start before selecting work. It validates the tracked first-use decision in .csk/project-state.json, routes adopted repositories to $csk-adopt-plan-scaffold,…
Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.
Of the AGENTS.md files we read, the 32,671 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.
In this repository18 other files it gives its agents.
- 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-helpSkill
- csk-refineSkill
- csk-startSkill
- finish-branchSkill
- pull-main-ffSkill
- review-loopSkill
The file
We split this AGENTS.md into the rules it gives an agent
11 rules. About 0 are common, 5 are distinctive, and 6 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.
6
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 11
Prefer an applicable skill when one exists.
- Very specific: line 14
Read CLAUDE.md and relevant .claude/rules/.
- Very specific: line 15
Run $csk-start before selecting work.
- Very specific: line 18
Follow .claude/rules/workflow-state.md as the routing and transition SSOT.
- Distinctive: line 25
Never install, copy, or synchronize kit skills into a user's global Claude or Codex directories.
- Very specific: line 27
Run $claude-skill-proxy-sync after adding, renaming, moving, or removing a canonical skill, or after changing its frontmatter trigger/UI …
- Very specific: line 41
Use source-of-truth files instead of duplicating state.
- Distinctive: line 44
Create task rows only for work that survives the current session, is blocked or deferred, or is a verified interruption point.
- Very specific: line 46
After verified work on a non-default branch, use $finish-branch.
- Distinctive: line 48
Read files before modifying them, preserve unrelated user changes, and verify edits and tests before claiming completion.
- Distinctive: line 51
Keep all generated proxy paths repository-relative and reject global fallbacks.
≈ “Common” is an estimate: at least 1 in 5 of the 41,581 AGENTS.md 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
Codex Repository Instructions
<!-- CODEX_CLAUDE_COMPATIBILITY -->This starter kit supports both Codex and Claude. CLAUDE.md and the repository
rules define the shared workflow; these instructions add the Codex entry point.
Session Start
- Inspect the skills and tools currently available.
- Prefer an applicable skill when one exists. Use deterministic tools for mechanical checks and LLM judgment for semantic, contextual, or non-deterministic decisions.
- Read
CLAUDE.mdand relevant.claude/rules/. - Run
$csk-startbefore selecting work. It validates the tracked first-use decision in.csk/project-state.json, routes adopted repositories to$csk-adopt-plan-scaffold, and then readstasks/INDEX.mdplus live Git state. - Follow
.claude/rules/workflow-state.mdas the routing and transition SSOT.
Claude And Codex Skills
- Canonical skills live only in
.claude/skills/. - Ready-to-use Codex proxies live only in
.codex/skills/and load their canonical Claude skill at runtime. - Never install, copy, or synchronize kit skills into a user's global Claude or Codex directories.
- Run
$claude-skill-proxy-syncafter adding, renaming, moving, or removing a canonical skill, or after changing its frontmatter trigger/UI metadata. - Ordinary canonical skill body and resource changes do not require a sync.
/csk-adopt-plan-scaffold/$csk-adopt-plan-scaffoldis required and must remain bundled./finish-branch/$finish-branch,/audit-plan-loop/$audit-plan-loop, and/review-loop/$review-loopare required and must remain bundled./create-worker-worktree/$create-worker-worktreeand/pull-main-ff/$pull-main-ffown parallel-session worktrees and safe default-branch updates;docs/git-basics.mdis the beginner reference for both.
Work Rules
- Codex may perform productive repository work; it is not restricted to review.
- Use source-of-truth files instead of duplicating state.
features/INDEX.mdand feature specs own feature state;tasks/INDEX.mdowns immediate open or blocked durable continuation work;.csk/project-state.jsonowns onboarding state. - Create task rows only for work that survives the current session, is blocked or deferred, or is a verified interruption point.
- After verified work on a non-default branch, use
$finish-branch; Git and external mutations still require explicit authority. - Read files before modifying them, preserve unrelated user changes, and verify edits and tests before claiming completion.
- Change
CLAUDE.mdonly with concrete user authorization for that change. - Keep all generated proxy paths repository-relative and reject global fallbacks.
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.

