second-brain

Self-healing wiki maintenance agent. Runs a 5-phase consolidation cycle: Audit → Deduplicate → Relate → Enrich → Reindex. Dispatched automatically by reindex or manually via the second-brain plugin. <example> Context: User just ran /second-brain:improve which created several new wiki/learnings/ entries. user: "those new learnings need to be linked into the entity pages" assistant: "I'll dispatch the knowledge-maintainer agent to scan the new learnings, identify entities they touch, and add cross-references." </example> <example> Context: Reindex found broken wiki-links and duplicate slugs. user: "fix the wiki issues" assistant: "Let me dispatch the knowledge-maintainer to resolve broken links, merge duplicates, and rebuild the index." </example>

Subagent for deep code quality review and adversarial-critic duties. Goes beyond the automatic quality gate to find architectural issues, performance problems, subtle bugs, and serves as the fresh-context critic in the /second-brain:improve learning pipeline (prevents same-context judge-and-author bias). <example> Context: User just finished a non-trivial refactor and wants more than a syntax review. user: "give this a proper second opinion before I merge" assistant: "I'll dispatch the quality-reviewer agent for a deep architectural pass — logic errors, leaky abstractions, edge cases, the works." </example> <example> Context: The /second-brain:improve skill has extracted a candidate learning and needs a fresh-context critic to score it. user: [running /second-brain:improve, candidate proposal generated] assistant: "Dispatching quality-reviewer with only the proposal text and destination file (no transcript) to score independently." </example>

Adversarial validation of the second-brain plugin. Questions architectural layers from a position of maximum skepticism, drilling from general to granular. Rotates focus across runs using git-change priority, history rotation, and perspective variety so different questions are asked each time.

Detect existing AI-context files on the host (CLAUDE.md, AGENTS.md, .cursorrules, .windsurfrules, .continuerules, .aider.conf.yml, instructions.md) at $HOME and the current repo, and propose USER.md preferences plus PROJECT.md content. Prompts to confirm each chunk. Idempotent.

Manual deep-dive on the current/most-recent session. Proposes up to 3 grounded "pin" candidates the user can accept/reject/edit. No autonomous critic.

Health-check the second-brain wiki and PROJECT.md cross-references. Finds orphan wiki pages, dead [[wiki-links]], and broken Cross-references slugs. Read-only by default — offers fixes interactively.

Search the second-brain wiki for pages relevant to a question. Thin wrapper around the knowledge_search MCP tool — returns candidate pages and lets Claude synthesize a cited answer from them.

Scaffold the v1.0 hot tier — USER.md, projects/<slug>/PROJECT.md, projects.jsonl — for the active repo. Idempotent.

Show second-brain hot-tier and wiki health at a glance. Reports USER.md size, active PROJECT.md size, projects.jsonl project count, wiki page counts per category, and index.md status.

Detect installed plugin version vs the version in plugin.json, run idempotent migrations between them, and update the installed-version marker. Safe to run anytime — no-op when already at current version. Use after pulling a new plugin release.

7 hooks across 5 events