project-scaffold

Forge reusable skills from accumulated project-memory lessons. Use when the user wants to consolidate, distill, or extract durable skills out of project-memory/ — turning lessons-learned into agent-loadable skills that prevent the project from repeating its own mistakes. Trigger phrases: 'forge a skill', 'forge memory', 'distill lessons', 'extract skills from memory', 'consolidate project-memory', 'turn lessons into skills', 'evolve our skills', 'memory curator', 'what skill should this become'. Trigger proactively when project-memory wrap-up has produced 3+ new lessons since the last forge, when the user is about to open a PR after several captured lessons, or when a recurring theme emerges across 2+ session files. Reads project-memory/lesson-learned/ (and other subsystems), clusters lessons ≥7 days old, then merges them into umbrella skills under the host project's detected skill-loading path.

Use when the user wants to capture, organize, or retrieve project knowledge as a git-committable memory system at project-memory/ at the project root. Trigger proactively on phrases: 'lesson learned', 'save lesson', 'wrap up session', 'postmortem', 'retro', 'capture this', 'what did we learn', 'why did this fail', 'have we seen this before', 'add to lesson', 'end of session', 'project memory', 'team memory'. Also trigger when a mission is ending and no lesson has been captured yet, or when the user asks a question whose answer likely lives in prior work. Backed by qmd for hybrid BM25 + vector search over every subsystem under project-memory/.

Use when setting up a new project, auditing an existing one, or aligning a project so the same project-level skills and agent guide work under Claude Code, Codex, and Pi Agent. Trigger on phrases like "make this project canonical", "set up canonical skills layout", "ensure canonical skills style", "migrate CLAUDE.md to AGENTS.md", "make claude/codex/pi share the same agent guide here", "standardize the agent layout for this repo", or whenever the user opens a fresh project and wants the five project invariants (skills/ real dir, .claude/skills symlink, .agents/skills symlink, AGENTS.md canonical, CLAUDE.md = @AGENTS.md) to hold. Also trigger when the user mentions only one of these symptoms (drifted CLAUDE.md vs AGENTS.md, .claude/skills or .agents/skills as a real directory, multiple skills directories) — the goal is to converge to the canonical shape every time, not just react to the symptom they named.

Author or update a project's AGENTS.md to install behavioral guidelines that govern any agent acting in that project. Use when the user wants to configure project-level agent behavior, set up or extend AGENTS.md, add a behavioral floor, prepend a coding-discipline preamble, or apply a named behavior pack to a project. Bundles the Karpathy behavioral preamble as one reference pack and supports adding more. Trigger phrases include "set up AGENTS.md", "configure agent behavior", "add behavioral guidelines", "add karpathy preamble", "install behavior pack", "create AGENTS.md", "control how agents behave in this repo".