corpus-forge

Run the corpus-forge data-improvement chat loop. Use when the user wants to fortify low-confidence or metadata-poor entries — find the weakest, talk through the fix, and commit edits via MCP.

Check whether a newer corpus-forge release is available via the MCP check_update tool and recommend the channel-appropriate upgrade command. Use when the user asks about updating corpus-forge or when the server's instructions advertise a newer version.

Synthesize a project's AGENTS.md by combining local inspection with corpus-forge cross-corpus retrieval. Use when the user asks to "init agents", "generate AGENTS.md", "synthesize project conventions", or starts a brand-new repo that should ship with grounded agent guidance.

Run Nightly inside Claude Code — pick the next task from the priority cascade (resumed plans, unblocked tasks, accepted RFCs, ranked GitHub issues), execute on an isolated worktree, delegate to specialist sub-agents via the Task tool, land as a PR or local proposal, disclose uncertainty, render briefing. Phase 3 — autonomous-for-a-bounded-backlog.

Search a corpus-forge training corpus via its MCP server. Use when the user asks a question that should be grounded in indexed content (codebases, docs, conversations) rather than answered from model knowledge.