Intelligent development orchestrator for Claude Code — takes a GitHub issue or freeform description and drives it end-to-end to a merged PR.
superfix is a Claude Code skill that acts as the single front-door for all development work. You give it a GitHub issue number (#42) or describe what needs doing in plain English; it classifies the problem, assembles the right sub-skill set, and manages the full pipeline — exploration, planning, expert review, implementation, testing, code review, and PR — without requiring you to invoke individual skills manually or break the work into steps yourself.
It supersedes the older change-pipeline skill.
Invoke superfix (via /superfix in Claude Code) whenever you have:
- A GitHub issue number:
/superfix #42,/fix #42,/implement #42 - A freeform description:
/superfix "the login button is broken on mobile" - No arguments — it will ask for an issue number or description
The skill triggers automatically on the keywords superfix, issue #N, fix #N, or implement #N.
Reads the full issue — body, labels, all comments — and classifies it:
| Classification | Types |
|---|---|
| Problem type | bug, performance, ux, architecture, security, feature |
| Complexity tier | Simple, Moderate, Complex (provisional; confirmed after exploration) |
Posts a Problem Profile as a GitHub comment if an issue was referenced. Keeps the issue body as a living spec and adds timestamped progress comments at each phase transition.
Runs type-aware exploration before writing any code:
- bug — systematic debugging + execution path tracing
- performance — hot-path mapping + call graph + baseline measurement
- ux — component tree, design tokens, layout patterns
- architecture — full system mapping, dependency graph, structural analysis
- security — auth/data flow tracing + pre-plan security scan
- feature — quick lookup (Simple) or full explorer + architect (Moderate/Complex)
Type-specific design step (brainstorming, threat modelling, UX iteration), then a written plan saved to docs/superpowers/plans/YYYY-MM-DD-<slug>.md plus a context file passed to every subagent. All Moderate and Complex work goes through a blocking expert panel review before implementation.
- Simple — runs inline on the main thread
- Moderate/Complex — dispatches an autonomous agent with the context file and plan
Execute sequence for every type includes: TDD (where applicable), implementation, tests, code review, security audit, silent-failure hunting, type design analysis (if new types introduced), simplification, and verification before completion.
Pushes the branch, creates a PR, posts a closing comment on the GitHub issue, and closes the issue once the PR is confirmed merged.
Applied automatically — no user configuration needed:
| Task | Model |
|---|---|
| File reads, symbol lookups, quick checks | Haiku |
| Exploration, planning, implementation, review | Sonnet |
| Expert panel, complex architecture, ambiguous triage | Opus |
This repo is structured as a Claude Code installable plugin with a skill manifest.
# From your project root (requires Claude Code with plugin support)
claude skill install https://github.com/thedanielmay/superfix-skillThe skill manifest is at the repo root (SKILL.md). The implementation lives in skills/superfix/SKILL.md.
SKILL.md Root manifest (skills CLI discovery)
skills/
superfix/
SKILL.md Full skill implementation and phase definitions
The entire skill is prompt-based — no compiled code, no runtime dependencies. All behaviour is defined in skills/superfix/SKILL.md.
Active. The skill is used in production and supersedes change-pipeline. The classification rules, phase sequences, and skill inventory are kept up to date in the SKILL.md files.