Capture the current conversation, a source, or a synthesis into your personal knowledge base as a well-formed wiki note. Writes correct frontmatter with today's date, picks the right bucket (sources/external, reports/, concepts/, questions/), cites existing source IDs, and runs just refresh to reindex afterward. Use when the user says "capture this", "save this to the KB", "add this as a note", "/kb-capture", or asks to file a finished research output.
Download an external documentation site into your personal knowledge base as a doc pack and source note. Use when the user says "/kb-ingest <url>", "ingest these docs", "add these docs to the KB", or "download reference docs". Runs webdown crawl via the knowledge repo's ingest-pack recipe, then creates a MANIFEST.md and source note automatically.
Routing and retrieval policy for your personal knowledge base at ~/code/knowledge (or the knowledge-base: path in the active project CLAUDE.md). Use when the user asks about past research, prior notes, captured sources, topics they have studied, or anything that might live in a personal wiki of synthesized notes. Triggers on phrases like "what do we know about", "check my notes", "any research on", "from the KB", "have I looked into", "do I have notes on". Tells you to read the cheap index/ files first, then query via the qmd MCP server (lex for known terms, vec for fuzzy), then open full notes only once the shortlist is small. Also the entry point for answering "is my KB healthy?" — points at just check.
Enforce strict red-green-refactor TDD for any task. Use when implementing features, fixing bugs, adding functionality, or building new modules. Routes to the full 7-stage pipeline for new modules with 3+ behaviors, or runs inline red-green-refactor for bug fixes and small changes. Triggers on: "use TDD", "fix this bug", "add a feature", "implement", "run the pipeline", "TDD pipeline", "build a module with TDD", or any coding task in a project with TDD in its CLAUDE.md. Agents inherit the session model; pass `--model <name>` to pin one for the run.
Multi-model code review with cross-validation. Orchestrates independent Claude and GPT-5.5 reviews, cross-validates findings, outputs merged fix list. Use when reviewing code changes, auditing files, or wanting a second opinion. Triggers on: "cross-review", "multi-model review", "review with codex", "get a second opinion on this code". Pass `--model <name>` to run the Claude-side agents on a specific model.
Add TDD pipeline configuration template to this project's CLAUDE.md. Run this to set up the pipeline for a new project.