What problem does it solve? Teams running multi-stage AI agent workflows lose hard-won prompt knowledge when successful or failed prompts are never captured. This Skill standardizes how prompts are saved into a structured prompt library so the Friday prompt-review ritual produces consistent, reusable entries instead of ad-hoc notes. ## Core Features & Use Cases - Win-mode entries: Writes a full library entry at prompts/library/<stage>/<topic>.md with verbatim prompt body, why-it-works paragraph, success metric, failure mode, model affinity, and invocation example. - Loss-mode anti-patterns: Appends a one-line failure note to prompts/library/_anti-patterns.md so losing prompts stay cheap to record and easy to avoid later. - Update mode and overwrite protection: Appends dated update blocks to existing entries and refuses to overwrite files without explicit confirmation, offering a -v2 variant instead. - Use Case: After a planning workflow where a Codex prompt produced a perfect step-by-step plan, tell the agent to save the prompt; it validates the stage and slug, drafts the entry from the template, and writes prompts/library/plan/jwt-validation.md ready to commit. ## Quick Start Ask the agent to save the prompt that just worked as a win entry for the plan stage under the topic jwt-validation in the prompt library.
