Released: 2026-05-01
Price: $4.99
Category: orchestration
What It Does
Autonomous Skill Curator gives OpenClaw operators a structured, approval-gated way to keep their skill library healthy over time. It scores every skill for quality, identifies stale or risky skills, surfaces duplicate clusters, and produces a curation plan — all without touching a single file without your say-so.
The skill is advisory-first by design: it generates a JSON or Markdown plan for human review and does not mutate the skill library autonomously. There is no --apply flag. Every recommended change must be actioned by the operator after reviewing the plan.
Key Features
Quality Scoring
Each skill is scored against configurable criteria covering script presence, SKILL.md metadata completeness, file freshness, and risk signals. Skills below the --min-score threshold are flagged for review rather than action.
Duplicate Detection
Clustering logic groups skills with similar names or overlapping described functionality, making consolidation candidates visible at a glance.
Stale File Detection
Skills with no recent activity or missing core files are highlighted so they can be prioritised in the review queue.
Risk Flagging
Known risk patterns (skills that shadow core tools, unusual file counts) are noted in the plan output for Warlock review.
Rollback-Safe Output
The output plan is structured for use as an approval-gated work order. Each recommendation includes a rollback note so any change applied can be reversed cleanly.
Audit Logging
Every curation run produces a timestamped log of recommendations made, accepted, and rejected — creating a traceable record of how the skill library evolved.
Usage
python3 scripts/autonomous_skill_curator.py /path/to/skills
python3 scripts/autonomous_skill_curator.py /path/to/skills --out curation-plan.json
python3 scripts/autonomous_skill_curator.py /path/to/skills --min-score 60
Requirements
- OpenClaw v2026.3.23+
- Python 3.8+
- No third-party dependencies — Python stdlib only
- No API keys required · No network access needed
QA and Security Notes
Viper QA passed (6/6) and Warlock security cleared this release on 2026-05-01. No subprocess, no shell execution, no eval, no exec, no network I/O. risk_score=0.