Chat logs are disposable; the library is the memory. This turns a prompt that worked into a reusable entry, strips the case-specific facts out of it, and refuses to overwrite anything.
Maintain my prompt library, stored at [LIBRARY LOCATION — where the library document lives, e.g., a Word file in a named folder]. I will hand you a working prompt (or a project that used one); genericize it and add it as a new entry, or revise an existing entry. First inventory the existing library so the new entry matches the others exactly and doesn't duplicate one.
Entry format, matched exactly to what's already there: a titled section for each prompt with (1) a one-word-or-phrase name and a short parenthetical on its role/audience; (2) a Purpose paragraph — what the deliverable is, who uses it, where in the workflow it runs, what it needs; (3) the copy-paste prompt itself in a visually distinct block; (4) Guardrails — numbered failure-modes-to-avoid, each an actual observed problem stated as a behavior to demand; and (5) Style Points for the deliverable. End the document with a dated log table; add a row for every addition or revision, noting where the prompt was distilled from.
Genericizing rule: strip every [CASE/PROJECT-SPECIFIC FACT — names, numbers, parties, dates] and keep only the shape of the task, expressed as placeholders in the form [NOUN CATEGORY — explanation of what the placeholder is FOR and how it relates to the rest of the prompt]. The user replaces the whole bracket; every placeholder must justify why the prompt needs that input, not merely name it.
Preservation discipline: never overwrite the library file — save every material change as a new versioned copy whose filename states the version and what changed. Match the existing document's own formatting by editing its structure directly, not by regenerating it through a converter that would flatten the styling. After saving, show me the new entry rendered so I can eyeball it before it's final.
Finally, at the end of every project — not just this one — study what we did for lessons learned and append them to [MEMORY STORE — the skill file or notes that persist across sessions], and proactively flag the moments when it's worth creating a new entry, a new skill, or some other durable memory structure. Ask before any large-token pass, and offer a leaner route as part of asking.