| name | skill-miner |
| description | Mine loop governance DB, sessions, and memory for reusable skill patterns. Scores findings with nomic-embed-text and reports to the orchestrator via the agent inbox. |
| version | 1.0.0 |
| author | Hermes Cortex |
| license | MIT |
| platforms | ["linux"] |
| metadata | {"hermes":{"tags":["mining","skills","patterns","loop-governance","moses"],"related_skills":["soul-refinement","save-lesson","hermes-agent-skill-authoring"]}} |
Skill Miner
What This Is
A script that mines available local data sources on an agent machine for reusable patterns — high-scoring TDD cycles from loop governance, session history patterns, agent memory workflows, and custom skills not yet in the hermes-cortex repo.
Scores findings with nomic-embed-text (Ollama) and sends high-confidence results to the orchestrator via the agent inbox (inbox_orchestrator) for review and upstreaming.
Location
| File | Path |
|---|
| Bash wrapper | ~/.hermes-cortex/scripts/skill-miner-wrapper (deployed) |
| Python script | ~/hermes-cortex/core/governance/skill_miner.py |
| Config | ~/.hermes-cortex/state/skill-miner.json |
Data Sources
- Loop governance DB — high-scoring TDD cycles (
composite >= 0.7) with
clean feedback_accept notes. These prove a reusable pattern was found.
- Session history — recurring tool sequences and user corrections.
- Agent memory — MEMORY.md / USER.md entries that describe workflows.
- Custom skills —
~/.hermes/skills/ skills not present in the
hermes-cortex repo (candidates for upstreaming).
Scoring
Each finding is embedded with nomic-embed-text and compared against the
embedding of "reusable Hermes agent skill pattern". Findings above a
similarity threshold are reported; low-confidence findings are dropped
to keep the report signal-dense.
Output
The script sends Moses an inbox message with subject
📬 SKILL MINER: N candidates containing:
- Skill name (or pattern name)
- Category suggestion
- Evidence (cycle id / session id / memory entry)
- Score
- Suggested trigger/description
Moses reviews the candidates, upstreams the good ones to skills/, and
updates skill-decisions.json with the disposition.
Running Manually
python3 ~/hermes-cortex/core/governance/skill_miner.py --send
python3 ~/hermes-cortex/core/governance/skill_miner.py --dry-run
Related
soul-refinement — daily SOUL.md refinement (companion mining)save-lesson — one-off bug-fix lesson capturehermes-agent-skill-authoring — how to author upstreamable SKILL.md filesorch-skill-lifecycle — the orchestrator's skill intake pipeline
