scaffoldlisted

# scaffold — deterministic scaffolding for any coding agent This Skill drives `@mohantn/scaffold-core`, the `scaffold` CLI. The CLI is LLM-agnostic: it never calls a model. Your three jobs each turn are: 1. Build an **intent manifest** describing the boilerplate to scaffold, in TOON against the published schema. 2. Run `scaffold generate` against the manifest — it renders files, injects marker-based registrations into existing files, and writes a TOON report on stdout. 3. Fill every `AI_IMPLEMENTATION_START/END` block the report marks as `empty: true` using your Edit tool. The deterministic half (file rendering, marker-based injection, hash-trail idempotency) is the CLI's. The probabilistic half (turning this user's natural-language ask into a manifest, then implementing the join logic in each `AI_IMPLEMENTATION` block) stays in you. Do not invent boilerplate you could have asked the CLI to render — that re-introduces the latency and drift this tool exists to remove. ## One-time setup (first time this Skill runs in a target repo) Four hook scripts ship in the toolkit's `hooks/` directory: `pre-tool-use.mjs`, `post-tool-use.mjs`, `stop.mjs`, and `user-prompt-submit.mjs`. Together they make both phase-3 completion *and* the "always go through `scaffold generate`" rule deterministic instead of prose-based: - `PreToolUse` (`pre-tool-use.mjs`) is the hard gate. It fires *before* a `Write` or `Edit` tool call runs and blocks it outright when the target is a pack-owned file be