How valentynkit/jev-belay works
jev-belay is a Claude Code Stop hook that checks transcript evidence before asking Jev whether an unverified completion claim should be blocked. Its main path is local and fail-open; the README also documents a decision-log viewer and a separate corpus-based evaluation workflow. The graph preserves those research and support capabilities. Some hook internals and optional integrations are unsampled, so uncertain links are omitted rather than inferred.
Main entry points: belay.mjs Stop hook · corpus Stop records · label.mjs Human or proxy labels · measure.mjs Evaluation
Components 15
Stop gate
- Stop hook
belay.mjs - Transcript evidence
belay.mjs - Jev API
- Decision and guard
belay.mjs - Fake Jev
tools/fake-jev.mjs
Decision visibility
- Live view and replay
tui.mjs - Decision log
belay.mjs
Corpus evaluation
- Corpus extraction
tools/extract-corpus.mjs - Stop records
corpus - Human or proxy labels
tools/label.mjs - Evaluation
tools/measure.mjs - Label rubric
corpus/RUBRIC.md
User support
- Explain decision
skills/why/SKILL.md - Install diagnostics
skills/doctor/SKILL.md
Other components
- Claude Code
Connections 20
- Claude Code → Stop hook: invokes stop
README.md - Stop hook → Transcript evidence: reads evidence
belay.mjs - Stop hook → Jev API: asks questions
README.md - Stop hook → Decision and guard: records decision
README.md - Stop hook → Claude Code: blocks or allows
README.md - Stop hook → Fake Jev: can target
README.md - Stop hook → Live view and replay: loads commands
belay.mjs - Decision and guard → Decision log: writes log
README.md - Live view and replay → Decision log: reads decisions
README.md - Explain decision → Decision log: explains last
README.md - Install diagnostics → Stop hook: checks round trip
README.md - Transcript evidence → Corpus extraction: projects turns
tools/extract-corpus.mjs - Corpus extraction → Stop records: writes records
tools/extract-corpus.mjs - Stop records → Human or proxy labels: supplies stops
tools/label.mjs - Label rubric → Human or proxy labels: guides labeling
tools/label.mjs - Human or proxy labels → Stop records: appends labels
tools/label.mjs - Stop records → Evaluation: supplies corpus
tools/measure.mjs - Human or proxy labels → Evaluation: supplies labels
tools/measure.mjs - Evaluation → Jev API: requests answers
tools/measure.mjs - Evaluation → Stop hook: uses ask
tools/measure.mjs