Promptify is a Claude Code skill repository. It contains:
- A single skill at
skills/promptify/SKILL.mdthat converts short developer intent into a structured task brief. - A set of Markdown shared rules and templates under
skills/promptify/shared/.
The repository must not contain a runtime service, web UI, database, telemetry, MCP server, npm CLI, or any non-Claude-Code adapter.
skills/promptify/SKILL.md: The Promptify skill instructions. References siblingshared/files inside the skill directory.skills/promptify/shared/brief-standard.md: Compact brief blocks, guided prompt-first behavior, execution modes, language rules.skills/promptify/shared/glossary.md: Canonical Promptify terminology for docs, templates, and generated briefs.skills/promptify/shared/task-routing.md: Task type cues, routing priority, examples.skills/promptify/shared/safety.md: High-risk signals, safety levels, confirmation behavior.skills/promptify/shared/context-discovery.md: Minimal-context exploration rules.skills/promptify/shared/evolution-loop.md: SIA-style generational loop contract for the evolve task type.skills/promptify/shared/template-authoring.md: Rules for creating and editing compact templates.skills/promptify/shared/test-plan.md: Manual QA checklist source.skills/promptify/shared/templates/*.md: Templates for task, bugfix, feature, prototype, data analysis, evolve, refactor, test, review, docs, plan, PRD, goal, handoff.docs/adr/: Architectural decision records for product boundaries and durable design decisions.docs/out-of-scope/: Rationale for product directions Promptify intentionally does not support.README.md,README.zh-CN.md: User-facing installation, usage, safety, QA, limitations.
- Keep the product Markdown-first. Do not reintroduce executable runtime code, host abstractions, install/uninstall CLIs, or multi-platform adapters.
- Put platform-neutral rules in
skills/promptify/shared/. Put skill-specific routing and behavior inskills/promptify/SKILL.md. - Skill references to
shared/...are relative toskills/promptify/; keepshared/inside thepromptifyskill directory. - Use
skills/promptify/shared/glossary.mdterminology consistently when editing docs, templates, or skill behavior. - Use
skills/promptify/shared/template-authoring.mdwhen changingskills/promptify/shared/templates/or the brief contract. - Match generated brief language to the user's input language while preserving technical identifiers in their original form.
- Treat the
promptifyskill as the only entry point. There are no slash commands. - Record durable product-boundary decisions in
docs/adr/; record rejected recurring feature directions indocs/out-of-scope/.
- Do not add web UI, hosted service, cloud sync, telemetry, external database, or MCP indexing.
- Do not add a Codex adapter, an npm package, or a CLI installer.
- High-risk signals: deletion, migration, payment, permission, auth, security, production, mass update, rewrite, purge.
- High-risk or destructive work must start with analysis and require explicit confirmation before edits.
- For review-only and plan-only modes, do not imply files were changed unless edits were explicitly requested later.
- For prd-only mode, produce the PRD and stop without publishing to an issue tracker or entering implementation.
After changing shared templates, the SKILL.md, or any of the markdown docs:
rg -n "目标:|模式:|上下文:|要求:" skills/promptify/shared/templates -g '!prd.md'
rg -n "问题陈述:|解决方案:|用户故事:|实现决策:|测试决策:|非目标:|补充说明:" skills/promptify/shared/templates/prd.md
rg -n "analysis-first|prompt-only|review-only|plan-only|prd-only|shared/templates" skills
rg -n "T[B]D|T[O]DO|implement late[r]|fill in detail[s]" skills/promptify/shared skills README.md README.zh-CN.md AGENTS.md CLAUDE.md
git diff --check HEADThe compact core-block scan should find the compact required blocks in each non-PRD template. The PRD scan should find the PRD core sections. The skill scan should find mode names and shared template references. The unfinished-marker scan should produce no output. If a check is not relevant to the files you changed, state why in the final report.
- Work on a feature branch or isolated worktree for multi-step changes.
- Keep commits scoped by concern.
- Do not revert unrelated user changes.