Fable Soul skill
A judgment layer for agents: rules for verifying results, checking before saying done and answering briefly, from real failures.
- 108 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- The sync script writes rules into global instruction files and can touch settings in the home directory.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/fable-soulDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The sync script writes rules into global instruction files and can touch settings in the home directory.
This is third-party code. Review the repository files before installing.
What it does
Fable Soul is a rule set drawn from real AI agent failures, packaged as a skill. The rules cover typical misses: saying done without running tests, patching the symptom instead of the cause, claiming faster without a measurement, answering a one-line question with a long report. The core file `references/soul.md` is about 250 lines. The skill has three modes: Embody loads the rules, Maintain syncs them and detects drift, Transfer turns experience into new skills with templates. The `sync_soul.py` script places the rules in global instruction files, `check_update.py` compares them with the repository, and `examples/hooks.json` adds an optional Stop hook for Claude Code.
Who it is for. Developers tired of unverified "done" answers who want a shared judgment style for Claude Code and Codex.
Good fit when
- The agent reports a fix without running a test, and it keeps happening.
- A small question gets a five-section report.
- The rules are wanted in every session, not only when the skill is invoked.
- A team collects its own rules from real agent failures and wants a verification loop.
Not a fit when
- You need knowledge of a particular framework or API: the rules teach working manner, not a domain.
- The project already has a long `CLAUDE.md` or `AGENTS.md` and another ~250 lines in every session is unwelcome.
- The agent runs without write access to the home directory.
Example request
Load Fable Soul and, before replying done, run the check and show its output.Limitations
The always-on mode (option B) writes a compact version of the rules into global instruction files on your machine. If a `CLAUDE.md` or `AGENTS.md` already exists, the script skips that file and the body of `references/soul.md` is pasted in by hand. Rules grow from your cases: adding new ones without reproducing a failure is not advised. If `soul.md` grows past ~250 lines the README advises trimming first. The skill targets Claude Code and Codex; other environments read the rules file manually.
How to disable. Delete the `fable-soul` folder from `~/.claude/skills/` or `~/.codex/skills/`. For always-on mode remove the rules block `sync_soul.py` wrote into the global instruction files.
FAQ
How does option A differ from B?
Option A copies the folder and the rules load when the skill is invoked. Option B runs `sync_soul.py` and the rules land in global instructions for every session.
How do I update?
Run `python scripts/check_update.py`: it compares local files with the repository. On first use the skill asks about a weekly check.
What does the sync verify?
`sync_soul.py --check` shows whether all install locations match and refuses to sync when the structure diverges.
Is there a translated README?
Yes, in Traditional Chinese: `README.zh-TW.md`.
Are there dependencies?
No. It is plain Markdown plus two small Python scripts with no third-party packages.
Discussion
No questions or reviews yet.
Related
Skillsfable-skills (DizzyMii)
fable-skills
Six discipline skills for Claude Code: answer first, edits only as asked and a claim of success only with proof.
A Claude Code plugin with four working-method skills: classify the task, act in steps, verify the result and generate domain adapters
A meta-skill that folds rules scattered across AGENTS.md, CLAUDE.md and Cursor files into one project skill with task routes.
An execution-discipline skill for Claude: staged planning, delegation to agents, failable checks and a double-check before delivery
A skill that makes the lead Claude model a foreman: it plans work, hands tasks to worker agents and verifies the result itself.
A Claude Code plugin that injects Anthropic guide rules into sessions, with the fable command expanding a vague ask
README in short
The README compares common agent mistakes with what a senior engineer does, then covers the folder contents, the rules and a failure-capture loop: record the excuse verbatim, reproduce, edit minimally.