fable-gpt
Skill5 starsChanged 2 months agoMITView on GitHub
Use for heavy implementation, debugging, test fixing, or multi-file edits Codex could execute, or orchestrating Claude + Codex.
In this repository57 other files it gives its agents.
- adversarial-reviewSkill
- aesthetic-animaxxingSkill
- ai-search-optimizationSkill
- announceSkill
- audio-trackSkill
- blindspotSkill
- brave-wesSkill
- chrome-wesSkill
- clihubSkill
- dashclaw-governanceSkill
- deskclawSkill
- de-vibeSkill
- dispatch-blocksSkill
- epic-infographicsSkill
- frontend-verifySkill
- gsap-astroSkill
- gsap-coreSkill
- gsap-frameworksSkill
- gsap-nextjsSkill
- gsap-nuxtSkill
- gsap-performanceSkill
- gsap-pluginsSkill
- gsap-react-routerSkill
- gsap-reactSkill
- gsap-scrolltriggerSkill
- gsap-sveltekitSkill
- gsap-tanstack-routerSkill
- gsap-timelineSkill
- gsap-utilsSkill
- gsap-vanillaSkill
- harness-healthSkill
- iphoneSkill
- landing-page-cro-expertSkill
- launchSkill
- launch-videoSkill
- logo-revealSkill
- marketingSkill
- marketing-studioSkill
- meditateSkill
- moleresearchSkill
- og-assetsSkill
- petstoreSkill
- phone-gotchasSkill
- phoneSkill
- preflightSkill
- product-demoSkill
- ship-itSkill
- shipSkill
- show-meSkill
- skillfindSkill
- social-clipSkill
- spec-probeSkill
- teamSkill
- tregSkill
- wes-voiceSkill
- wrapSkill
- xplainerSkill
The file
We split this skill into the rules it gives an agent
1 rule. About 0 are common, 1 is distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 44
Run the "Done when" verification command yourself and read the output.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 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
Fable-GPT: Claude orchestrates, Codex executes
Role split
- Claude (main thread): planning, repo understanding, architecture decisions, task decomposition, final review, small surgical edits.
- Codex (codex plugin): heavy implementation, debugging, test fixing, refactors, multi-file edits. Runs on the ChatGPT subscription — zero Claude tokens for execution. Global default is
gpt-5.5atxhighreasoning (~/.codex/config.toml); do not pass--model/--effortunless deliberately overriding.
How to delegate
| Situation | Route |
| Quick, bounded task | /codex:rescue (foreground) |
| Big, open-ended, or long-running | codex-rescue subagent (already model: sonnet, Bash-only — passes agent-model-guard) or --background; poll /codex:status, fetch /codex:result |
| Continue prior Codex work | say "resume"/"continue" (maps to --resume-last) |
| Routine pre-commit review or second-opinion diagnosis | /codex:review — default here before spending Claude subagent tokens; reserve /code-review or /adversarial-review for high-stakes diffs |
Task prompt shape (keep it focused)
Goal: [one sentence]
Files: [paths]
Constraints: [style, don't touch X]
Done when: [verification command + expected result]One task per delegation. Split unrelated work into separate rescues.
Workspace root gotcha: Codex's sandbox root is the session's cwd, not the task's target repo. If the target repo differs from cwd, the codex-rescue subagent and /codex:rescue will spawn a task that can't write ("patch rejected: writing outside of the project") — Codex burns the whole run discovering this. Invoke the companion runtime directly instead:
node "<plugin>/scripts/codex-companion.mjs" task --prompt-file <spec.md> --cwd "<target-repo>" --write --backgroundLong specs go in a --prompt-file (avoids shell quoting). Job state is keyed to the --cwd value — pass the SAME --cwd to status/result/cancel or the job appears not to exist. Errors print to stderr; don't 2>/dev/null status calls.
After Codex finishes — never accept blindly
git diff— read what actually changed.- Run the "Done when" verification command yourself and read the output.
- Wrong or sloppy → re-delegate with specific corrections (use resume); small issues → fix inline.
When NOT to delegate
- Trivial edits (faster inline than writing the delegation prompt).
- Architecture or product decisions — Claude's job.
- Anything under CLAUDE.md Hard Stops (deploy, migrations, auth/billing config, external comms).
- Tasks needing conversation context Codex doesn't have — inline the needed context into the task prompt or keep it in Claude.
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.