| name | semantic-discovery-loop |
| description | Run a bounded semantic discovery loop over a source repository, producing source and semantic graphs with evidence-backed reporting. |
Semantic Discovery Loop
Use this skill to inspect an existing codebase without modifying source files.
Primary instructions:
- Read
.agent-loop/LOOP.md.
- Use
.agent-loop/skills/00-loop-conductor.md as the governing procedure.
- Use the remaining
.agent-loop/skills/*.md files as phase-specific instructions.
Allowed writes:
.work/semantic-loop/**
.cache/scripts/**
Forbidden:
- modifying source files
- installing dependencies without explicit approval
- inventing unsupported semantic constructs
- marking the loop complete while any verification score is below 8
Final output:
.work/semantic-loop/reports/application-structure.md
.work/semantic-loop/verification.json