claude-skill-proxy-sync
Skill1 starsChanged 4 months agoMITView on GitHub
Synchronize canonical repository-local Claude skills from .claude/skills into portable thin Codex proxies under .codex/skills. Use after adding, renaming, moving, or removing a local skill, or after changing its YAML trigger description or agents/openai.yaml interface metadata. Do not use for ordinary SKILL.md body edits because Codex reads the canonical Claude file at runtime.
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
- 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 skill into the rules it gives an agent
11 rules. About 0 are common, 8 are distinctive, and 3 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈8
Distinctive · estimate
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 8
Keep one canonical skill implementation for Claude and Codex without installing anything into a user's global skill directories.
- Distinctive: line 15
Every proxy resolves the repository root at runtime and reads the canonical Claude skill through a repository-relative path.
- Very specific: line 17
Never read from or write to %USERPROFILE%\.claude\skills, %USERPROFILE%\.codex\skills, ~/.claude/skills, ~/.codex/skills, or …
- Distinctive: line 20
Never embed the current machine's absolute repository path in a proxy.
- Distinctive: line 25
Run the sync after:
- Distinctive: line 32
Do not sync after an ordinary body, script, reference, or asset edit.
- Distinctive: line 38
Read AGENTS.md, CLAUDE.md, and repository rules before changing files.
- Distinctive: line 53
If a canonical skill was deliberately removed, review the stale list and prune only generated proxies:
- Very specific: line 60
Run the script once more without --apply.
- Distinctive: line 62
Run the bundled tests and skill validation.
- Very specific: line 64
Use --name <skill-name> to inspect or update one skill.
≈ “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
Claude Skill Proxy Sync
Keep one canonical skill implementation for Claude and Codex without installing anything into a user's global skill directories.
Contract
.claude/skills/<name>/SKILL.mdis the canonical implementation..codex/skills/<name>/SKILL.mdis a generated thin proxy.- Every proxy resolves the repository root at runtime and reads the canonical Claude skill through a repository-relative path.
- Never read from or write to
%USERPROFILE%\.claude\skills,%USERPROFILE%\.codex\skills,~/.claude/skills,~/.codex/skills, or$CODEX_HOME/skills. - Never embed the current machine's absolute repository path in a proxy.
- Preserve
csk-adopt-plan-scaffold; it is a required bundled skill.
Decide Whether Sync Is Needed
Run the sync after:
- adding, renaming, moving, or deleting a directory in
.claude/skills; - changing
nameordescriptionin a canonical skill's YAML frontmatter; - changing
display_name,short_description, ordefault_promptin a canonical skill'sagents/openai.yaml.
Do not sync after an ordinary body, script, reference, or asset edit. The proxy loads those canonical repo-local files at runtime, so those changes are live.
Workflow
- Resolve the current Git repository root. Stop if
.claude/skillsis missing. - Read
AGENTS.md,CLAUDE.md, and repository rules before changing files. Inspect the plan without writing:
python .claude/skills/claude-skill-proxy-sync/scripts/sync_claude_skill_proxies.py --repo .- Review every
create,update,stale, orconflictresult. Never overwrite an unmanaged Codex skill. Apply safe creates and updates:
python .claude/skills/claude-skill-proxy-sync/scripts/sync_claude_skill_proxies.py --repo . --applyIf a canonical skill was deliberately removed, review the stale list and prune only generated proxies:
python .claude/skills/claude-skill-proxy-sync/scripts/sync_claude_skill_proxies.py --repo . --apply --prune- Run the script once more without
--apply; it must reportokfor every canonical skill and no stale proxies. - Run the bundled tests and skill validation.
Use --name <skill-name> to inspect or update one skill. Use --json for a
machine-readable plan. Read references/proxy-plan.md
before changing the proxy schema or generator behavior.
Safety
--pruneis valid only with--applyand removes only directories carrying this generator's marker.- A target without the marker is a conflict and remains untouched.
- Generated paths must remain inside
.codex/skillsunder the resolved repo. AGENTS.mdandCLAUDE.mdare maintained source files, not generated output. Change them only when the user authorized that exact change.
Completion Evidence
Report the canonical skill count, created/updated/pruned proxies, conflicts, test results, and confirmation that no generated file contains an absolute or global user path.
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.

