Promptify is a Claude Code skill that turns brief developer intent into a structured, context-aware task brief, then lets the user decide whether to enter execution.
When a short intent such as "fix the login failure message" is handed directly to an agent, common failures include unclear scope, incorrect context, missing test expectations, or high-risk edits starting too early. Promptify generates a compact brief before execution, making the goal, mode, context, requirements, and safety gate explicit.
Promptify is not an auto-executor or installer. Its core value is giving the user and agent the same task contract before work begins.
- A Claude Code skill package:
skills/promptify/SKILL.md. - A set of shared Markdown rules and templates:
skills/promptify/shared/. - Coverage for common development tasks: bugfix, feature, refactor, test, review, docs, planning, PRD, and long-running goal prompts.
Promptify is most useful when the task intent is short but the project context is complex. It is especially useful for work involving permissions, authentication, security, migrations, deletion, production risk, or other high-risk changes, and for tasks that need explicit modes such as review-only, plan-only, or prd-only.
It also fits teams that want a consistent agent intake format, or users who want a confirmable task contract before Claude Code executes. Promptify turns a short intent into a brief with a goal, mode, context, requirements, and any needed safety gate so the user and agent align before editing.
Promptify is less useful for tiny mechanical edits, tasks where the user has already written a complete implementation plan, situations where Claude Code should immediately enter a debugging loop, or requests that need new tool or platform capabilities.
Promptify relates to Claude Code as a front-door navigator relates to an execution engine: Promptify reduces misunderstanding, controls scope, and improves verifiability; Claude Code reads code, edits files, runs tests, and debugs. The trade-off is an extra workflow step, and Promptify does not improve Claude Code's underlying execution capability.
- Classifies short intent and routes it to the right template via
skills/promptify/shared/task-routing.md. - Discovers the smallest useful project context via
skills/promptify/shared/context-discovery.md, then generates a compact brief. - Matches the brief language to the user's input: Chinese input produces Chinese, English input produces English; technical identifiers, commands, and paths stay unchanged.
- Routes high-risk input such as deletion, migration, payment, permission, auth, security, production, mass update, rewrite, and purge into analysis-first mode; destructive edits require explicit confirmation.
- Defaults to prompt-first: output the brief first, then ask whether to execute. Users can explicitly request
prompt-only,review-only,plan-only,prd-only, orgoalmode.
promptify/
README.md
README.zh-CN.md
AGENTS.md
CLAUDE.md
docs/
adr/
0001-markdown-first-single-entry.md
out-of-scope/
codex-adapter.md
multi-skill-split.md
npm-cli-and-installer.md
web-ui-or-hosted-service.md
skills/
promptify/
SKILL.md
shared/
brief-standard.md
context-discovery.md
glossary.md
safety.md
template-authoring.md
task-routing.md
test-plan.md
templates/
bugfix.md
docs.md
feature.md
goal.md
plan.md
prd.md
refactor.md
review.md
task.md
test.md
skills/promptify/SKILL.md references resources in its sibling shared/ directory. Keep shared/ under skills/promptify/ so the resources are included when skills/promptify is installed as a Claude Code skill.
Promptify is distributed as a repository. It has no npm package, installer, or automatic host registration. Put the repository where Claude Code can load the skill.
Recommended: clone the repository to a stable location, then let Claude Code load its skills/ directory. For example:
git clone <repo-url> ~/promptify
ln -s ~/promptify/skills/promptify ~/.claude/skills/promptifyYou can also place the whole repository inside a project as a project-level skill. In either case, keep shared/ inside skills/promptify/.
Describe your development intent in Claude Code and ask it to use the promptify skill. For example:
Use promptify: fix the login failure message
promptify: review the current changes
promptify: turn the current discussion into a PRD
promptify: implementation plan for team template overrides
promptify: turn docs/superpowers/plans/2026-05-13-promptify-mvp.md into a long-running goal
The skill generates a compact brief first, then asks whether to execute. PRD mode generates the PRD and stops.
promptify: fix the login failure message
promptify: add CSV export
promptify: refactor order status calculation without behavior changes
promptify: add tests for payment callbacks
promptify: review the current branch, focusing on security and regression risk
promptify prompt-only: turn "optimize settings page load speed" into a brief; do not execute
promptify plan-only: plan the migration from REST to GraphQL
promptify prd-only: turn "team template overrides" into a PRD
promptify: delete the old permission table and migration scripts
promptify goal: keep progressing docs/plans/cache-cleanup.md
Input:
promptify: fix the login failure message
Example output:
Goal:
Fix the user-visible message shown when login fails, while preserving the existing login flow and error handling style.
Mode:
Generate the brief first and ask whether to enter execution.
Context:
Inspect the login entry point, error handling, existing tests, and login-related docs first.
Requirements:
Locate the source of the failed-login message and make the smallest scoped fix without changing the authentication flow. Add or update relevant tests, run the targeted verification command, and report changed files, behavior changes, verification results, risks, and follow-ups.
Input:
promptify: add CSV export
Example output:
Goal:
Add CSV export support for the existing list or report, following the current product interaction and permission boundaries.
Mode:
Generate a feature brief first, then execute after user confirmation.
Context:
Inspect the relevant list view, data query layer, permission checks, existing export behavior, and test coverage first.
Requirements:
Clarify exported fields, filters, file naming, and empty-state behavior. Reuse existing data and permission logic, avoid adding a separate export service, add targeted tests, and report verification commands, behavior changes, and remaining risks.
Input:
promptify: refactor order status calculation without behavior changes
Example output:
Goal:
Refactor order status calculation to reduce complexity while preserving existing external behavior and data results.
Mode:
Generate a refactor brief first, then execute after confirmation.
Context:
Inspect order status entry points, callers, edge cases, existing tests, and historical docs first.
Requirements:
Lock down current behavior and test baseline before making small structural changes. Do not change status enums, API responses, or database structure. Add missing edge-case tests, and report refactor scope, verification results, and compatibility risks.
Input:
promptify: review the current branch, focusing on security and regression risk
Example output:
Goal:
Review the current branch changes, prioritizing security, permission, data consistency, and regression risks.
Mode:
review-only; output review findings only and do not edit files.
Context:
Inspect git diff, related tests, permission boundaries, external input handling, and high-risk modules first.
Requirements:
List findings by severity with file locations, impact, and actionable fixes. If no clear issue is found, state residual risk and recommended additional verification.
Input:
promptify prd-only: turn "team template overrides" into a PRD
Example output:
Problem Statement:
Teams want to maintain overrideable brief template rules for specific projects or teams while keeping the default templates available.
Solution:
Define template override loading order, scope, conflict handling, and fallback behavior.
User Stories:
As a team maintainer, I can provide custom templates for team projects and keep using defaults when no override is configured.
Implementation Decisions:
Prefer a Markdown-first override convention without introducing a service, database, or runtime sync.
Testing Decisions:
Use documentation examples and template checks to cover default behavior, override behavior, override conflicts, and missing fallback.
Non-Goals:
No web UI, cloud sync, permission system, or cross-platform adapter.
Notes:
PRD mode stops after generating the document and does not automatically enter implementation.
Input:
promptify: delete the old permission table and migration scripts
Example output:
Goal:
Evaluate the deletion request for the old permission table and migration scripts, confirming impact radius and rollback path.
Mode:
analysis-first; this involves deletion, permission, and migration, so any destructive edits require explicit confirmation.
Context:
Inspect database migration history, permission callers, deployment flow, backup strategy, and test coverage first.
Requirements:
First produce an impact analysis, dependency list, verification plan, and rollback plan. Do not delete files, modify migrations, or remove permission paths without explicit user confirmation.
Input:
promptify goal: keep progressing docs/plans/cache-cleanup.md
Example output:
Goal:
Turn docs/plans/cache-cleanup.md into a durable goal prompt for staged cache cleanup work.
Mode:
goal; establish a long-running objective, phase checkpoints, and per-turn execution boundaries.
Context:
Read the plan document, cache modules, test commands, and known risks first.
Requirements:
Break the work into verifiable small steps, record the current phase, next step, blockers, and verification results, and avoid expanding each turn into refactors outside the plan.
| Task | Template |
|---|---|
| General task | skills/promptify/shared/templates/task.md |
| Bugfix / debug | skills/promptify/shared/templates/bugfix.md |
| Feature | skills/promptify/shared/templates/feature.md |
| Refactor | skills/promptify/shared/templates/refactor.md |
| Test | skills/promptify/shared/templates/test.md |
| Review | skills/promptify/shared/templates/review.md |
| Docs | skills/promptify/shared/templates/docs.md |
| PRD | skills/promptify/shared/templates/prd.md |
| Plan | skills/promptify/shared/templates/plan.md |
| Goal prompt | skills/promptify/shared/templates/goal.md |
See skills/promptify/shared/safety.md. High-risk tasks such as deletion, migration, payment, permission, auth, security, production, mass update, rewrite, and purge enter analysis-first mode, and destructive edits require explicit confirmation.
Promptify's boundaries are recorded in docs/adr/0001-markdown-first-single-entry.md. Out-of-scope directions are recorded in docs/out-of-scope/, including npm CLI, Codex adapter, web UI / hosted service, and multi-skill splitting.
When maintaining templates, refer to skills/promptify/shared/glossary.md and skills/promptify/shared/template-authoring.md to keep terminology, mode semantics, and safety gates consistent.
Versions and changes are maintained in CHANGELOG.md. Version numbers follow SemVer: documentation or template touch-ups increment PATCH, new templates, modes, or notable workflow behavior increment MINOR, and breaking layout or behavior changes increment MAJOR.
After changing shared templates or SKILL.md, run at least:
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 first command should find compact core blocks in every compact brief template. The second should find the PRD core sections. The third should find mode names and shared template references. The unfinished-marker scan should produce no output.
- No npm CLI, installer, or automatic host registration.
- No Codex adapter; this repository targets Claude Code skills only.
- No web UI, hosted service, cloud sync, telemetry, or MCP indexer.
- Execution-stage capability depends on Claude Code itself, and by default requires user confirmation after brief generation.