knowledge-skills¶
Autonomous knowledge management for keeping AI context files (CLAUDE.md, AGENTS.md) up to date with recent codebase changes. Scans merged PRs from the last N days, dispatches parallel extraction agents to identify knowledge items, synthesizes proposed updates, runs an adversarial review, and produces a git-apply-able patch for human review.
The skill is forge-agnostic (GitHub via gh CLI, GitLab via glab CLI) and designed for CI pipeline execution. It produces artifacts (patch file, run report, extraction data) but never creates PRs/MRs itself — external tooling handles forge interactions.
Plugin Details
- Version: 0.1.0
- Author: opendatahub-io
- License: Apache-2.0
- Category: Documentation
- Repository: opendatahub-io/knowledge-skills
- Tags: knowledge context claude-md agents-md pr-analysis automation
Skills¶
| Skill | Description | Invocable |
|---|---|---|
/knowledge.repo |
Scan merged PRs and propose updates to AI context files (CLAUDE.md, AGENTS.md) as a git-apply-able patch |
Installation¶
/plugin install knowledge-skills@opendatahub-skills
Architecture¶
A 7-phase pipeline with early-exit gates at each phase: Setup → Fetch PRs → Extract (parallel haiku agents, waves of 10) → Synthesize (opus) → Review (opus, context-isolated) → Revise (opus, conditional) → Artifacts.
Key architectural patterns: - Separation of concerns: SKILL.md orchestrates, scripts handle deterministic work, agent prompts handle judgment - Adversarial review: review agent is context-isolated from synthesis agent (doesn't see rationale, only diff + raw extractions) - Stateless: no tracking of prior runs. Rejected proposals are handled by adding guidance to context files, not by building state - Per-PR extraction: each PR gets its own agent context to avoid overflow, dispatched in parallel waves