What problem does it solve? Large or risky code changes often die to context/token limits, lose their plan between sessions, or ship half-broken when one chat tries to do everything. This Skill turns a big job into an ordered pack of small, self-contained prompts — each run in its own fresh chat, each leaving the project in a working, verifiable state — so work survives context limits and relays cleanly across chats and tools. ## Core Features & Use Cases - Pack authoring (Mode A): Performs read-only architecture reconnaissance and a pre-flight constraint scan, locks decisions and scope fences, then decomposes the work into risk-rated prompts with guardrails, verification matrices, and commit messages, saved to docs/<TOPIC>_PROMPT_PACK.md. - Pack execution (Mode B): Runs exactly one pack prompt per chat with a read-first protocol, file:line verification against current code, strict scope guardrails, and a report-and-stop loop that never commits without explicit approval. - Handoff briefings (Mode C): Writes paste-ready, self-contained briefings to resume a dying chat or relay work to another tool such as Codex via AGENTS.md. - Use Case: You need to migrate a backend across many files but keep running out of context. Ask for a prompt pack, receive a sequenced P1→P8 plan, then paste each prompt into a fresh chat, verify, commit, and continue until the migration ships. ## Quick Start Ask the AI to make a prompt pack for your large change, for example: "Break this onboarding rebuild into a prompt pack I can run one prompt per fresh chat."
