Find the semantic decisions in a codebase that could become typed Jev questions.
altslate-labs.github.io/jevable-code
jevable-code packages the find-jevable-code agent skill. Point it at a
repository and it finds jevable code — existing semantic judgments (routing,
triage, relevance, quality, escalation, fuzzy matching) that could be expressed
as a typed Jev Choice, Score, or Noul question. It produces ranked,
source-grounded findings with concrete question contracts, batching and
dependency analysis, fallback behavior, and a validation plan — or a defensible
finding that nothing qualifies.
It defaults to read-only review: no application code is run, nothing is sent to Jev, and no paid inference is made just to discover candidates.
| Primitive | Model judgment | Typical target |
|---|---|---|
| 🔴 Choice | select one supplied option | dispatch, intent, categorization |
| 🟠 Score | locate content on ordered levels | severity, quality, relevance |
| 🟡 Noul | evaluate one yes/no proposition | flags, matching, escalation signals |
The skill runs a seven-step read-only pipeline. Discovery searches in both directions (from model/heuristic producers and from the consumers of their results); suitability gates reject deterministic logic; each survivor gets a concrete decision contract; composition analysis groups compatible questions and keeps ranking and joint constraints in application code. The three bundled references are loaded on demand.
This one repo works as a skill in both Claude Code and Codex — they read the
same skills/find-jevable-code/SKILL.md.
Claude Code — copy (or symlink) the skill into your skills directory:
git clone https://github.com/AltSlate-Labs/jevable-code.git
cp -r jevable-code/skills/find-jevable-code ~/.claude/skills/Or install the whole repo as a plugin (.claude-plugin/plugin.json is included).
Codex — install the plugin; .codex-plugin/plugin.json declares the
./skills/ directory and agents/openai.yaml supplies the display name and icon:
codex skill install https://github.com/AltSlate-Labs/jevable-code.gitFrom a session with the skill available:
Use find-jevable-code to audit this repository for decisions suited to Jev Choice, Score, or Noul, and rank the opportunities with code evidence and concrete question contracts.
See examples/ for a worked end-to-end run: a toy keyword-based
support router (support_router.py)
and the audit report the skill produces for it.
jevable-code/
├── skills/find-jevable-code/ # the skill (portable across both platforms)
│ ├── SKILL.md # instructions — name + description frontmatter
│ ├── references/ # jev-contract · discovery · report-template
│ ├── assets/ # icon.svg · architecture.svg
│ └── agents/openai.yaml # Codex/OpenAI interface metadata
├── examples/ # worked input + expected audit report
├── test/validate_skill.py # package validator (run in CI)
├── .claude-plugin/plugin.json # Claude Code plugin manifest
└── .codex-plugin/plugin.json # Codex plugin manifest
Validate the package (frontmatter, links, manifests, grounded examples):
python3 test/validate_skill.pyThe same check runs on every push via GitHub Actions.
MIT © AltSlate Labs. "Jev" and the Choice/Score/Noul primitives are referenced as an external decision API; this repository only contains the audit skill and does not require a Jev API key.