prompt-update

name	prompt-update
description	Classify new content (rule, pattern, workflow step) and write it to the correct documentation layer. Use when user wants to add prompt documentation or mentions /prompt-update command.

/prompt-update - Classify & Place Documentation Content

Classify a piece of new content and write it to the correct file(s) in the project's documentation layers.

Usage

/prompt-update "always use eq() not any() in Mockito"
/prompt-update "test-review should catch Fake storage assertions"
/prompt-update                          # Interactive — asks what to add

Workflow

1. Parse input. If no content provided, ask the user what to add.
2. Understand the content. What does it mean? Is it a principle, a detection pattern, a code example, a workflow step?
3. Classify targets using .claude/templates/documentation/prompt-update-classification.md. Walk ALL 4 decision questions explicitly — principle, detection pattern, code example, layer-specific context. Show the classification before writing:

   1. Principle? → .claude/guidelines/frontend-rules.md (new section)
   2. Detection? → code-smells-routing-table.md (smell table) + scan-duplication.md (A-check)
   3. Template?  → new templates/refactoring/extract-tailwind-class.md
   4. Layer-specific? → n/a
   

   Most updates touch 2+ files. If only one target is identified, double-check — a rule without detection won't be enforced, a detection without a template won't guide the fix.
4. Check for duplication — grep key terms across .claude/rules/, .claude/tech/, .claude/agents/, .claude/skills/, .claude/templates/. If a duplicate exists, report it and ask: skip, merge, or replace?
5. Write to all target files in one pass. Match each file's existing style.
6. Tech-agnostic gate — for every write targeting a universal file (.claude/rules/, .claude/templates/), scan the content you just wrote for tech leaks. See "Tech-Agnostic Verification" in the classification template. If a leak is found, rephrase in universal terms first — only relocate to tech binding when the content is inherently tech-specific.
7. Product-agnostic gate — after writing, re-read every edit you made (all layers, not just universal) and scrub product-specific information: product/company name, business-domain terms, external vendor/integration names, concrete domain entity names, and story/feature names or numbers. Replace each with the generic concept it represents. See "Product-Agnostic Verification" in the classification template. The prompt library must be reusable across products; product instances belong only in ProductSpecification/.
8. Report what was written and where (see Output format in classification template).
9. Verify with /skill-creator — see Impact Assessment in classification template.

Constraints

• Never create new rules files — content fits existing files
• Never place fenced code blocks in agents or skills — extract to templates
• Never edit skill files directly — delegate to /skill-creator
• Always check for duplication before writing
• Preserve existing formatting and section structure of target files
• When the principle already exists in a higher layer, don't re-state it — add only the enforcement mechanism

Templates

• .claude/templates/documentation/prompt-update-classification.md — classification guide, decision process, examples, impact assessment

Next Steps

1. /prompt-refactor {target-file} — verify the file is still structurally clean.
2. /skill-creator — verify affected skills respect the new content (when Impact Assessment says to run it).