Skip to content
Back to skills

torque

ASecurity

Torque: a Claude Code skill that converts GSD (project management) and Superpowers (engineering discipline) into one coherent workflow. Prevents double-planning, conflicting subagent models, and overlapping design capture by providing a single routing decision at task start.

  • 6 stars
  • 0 votes
  • 0 copies
  • 12 views
  • Added May 28, 2026
ai-agentstestingdebuggingcode-reviewgitapifrontendsecurity

Works with

  • claude code
  • api

Security analysis

A100/100

Pro scans all 7 files and shows the line behind each finding

Scanned May 28, 2026

npx -y skills add mattjaikaran/torque --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of torque?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for torque
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/mattjaikaran-torque/badge)](https://www.skillsdirectory.com/skills/mattjaikaran-torque)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: torque
version: 0.3.0
description: "Torque: a Claude Code skill that converts GSD (project management) and Superpowers (engineering discipline) into one coherent workflow. Prevents double-planning, conflicting subagent models, and overlapping design capture by providing a single routing decision at task start."
---

# Torque — GSD + Superpowers, United

## Core Rule

Check for `.planning/` directory in the project root. This single check determines routing.

## Primary Routing Table

| Scenario | `.planning/` exists? | System | Entry Point |
|----------|---------------------|--------|-------------|
| New project (greenfield) | No → creates it | GSD | `/gsd:new-project` |
| Brownfield onboarding | No → creates it | GSD | `/gsd:map-codebase` then `/gsd:new-project` |
| New feature in GSD project | Yes | GSD | `/gsd:discuss-phase` → `/gsd:spec-phase` → `/gsd:plan-phase` → `/gsd:execute-phase` |
| AI/LLM integration phase | Yes | GSD | `/gsd:ai-integration-phase` → `/gsd:plan-phase` → `/gsd:execute-phase` → `/gsd:eval-review` |
| Quick task in GSD project | Yes | GSD | `/gsd:quick` |
| Trivial fix in GSD project | Yes | GSD | `/gsd:fast` (inline, no subagents) |
| Standalone feature (no GSD) | No | SP | **REQUIRED:** `brainstorming` → `writing-plans` → `subagent-driven-development` |
| PRD-driven feature | Either | SP → GSD | `write-a-prd` → `prd-to-issues` → `/gsd:new-project` or `/gsd:add-phase` |
| Bug fix (any context) | Either | SP first | **REQUIRED:** `systematic-debugging` ALWAYS first, then route fix through appropriate system |
| Code review (send) | Either | SP | **REQUIRED:** `requesting-code-review` (fills GSD's missing review step) |
| Code review (receive) | Either | SP | `receiving-code-review` — verify feedback before implementing blindly |
| Stress-test a plan | Either | SP | `grill-me` — interview until all ambiguity is resolved |
| Architecture improvement | Either | SP | `improve-codebase-architecture` — surface friction, propose refactors as RFC issues |
| Ship / PR creation | Yes | GSD | `/gsd:ship` or `/gsd:pr-branch` (filters `.planning/` commits) |
| Ship (standalone) | No | SP | `finishing-a-development-branch` |
| Resume work | Yes | GSD | `/gsd:resume-work` |
| Resume work | No | SP | Re-read plan files in `docs/plans/` |

## Extended Routing (GSD 1.34+)

| Scenario | Entry Point | Notes |
|----------|-------------|-------|
| Spec refinement | `/gsd:spec-phase` | Socratic clarification of WHAT a phase delivers — produces SPEC.md before planning |
| AI/LLM design contract | `/gsd:ai-integration-phase` | Framework selection, implementation guidance, evaluation strategy → AI-SPEC.md |
| AI eval audit | `/gsd:eval-review` | Retroactive audit of AI phase evaluation coverage |
| Spike (feasibility test) | `/gsd:spike` → `/gsd:spike-wrap-up` | Throwaway experiment to validate approach before committing |
| Sketch (UI mockup) | `/gsd:sketch` → `/gsd:sketch-wrap-up` | Rapid HTML mockups with multi-variant exploration |
| UI/frontend phase | `/gsd:ui-phase` → `/gsd:ui-review` | Design contract (UI-SPEC.md) before implementation |
| Security audit | `/gsd:secure-phase` | Verify threat mitigations post-implementation |
| Codebase intelligence | `/gsd:intel` | Query, refresh, or inspect `.planning/intel/` |
| Code review + auto-fix | `/gsd:code-review` → `/gsd:code-review-fix` | Two-step: find issues, then fix atomically |
| Autonomous execution | `/gsd:autonomous` | Runs all remaining phases: discuss → plan → execute |
| Parallel workstreams | `/gsd:workstreams` | Manage concurrent work across features |
| Persistent debugging | `/gsd:debug` | 4-phase debugging with knowledge base persistence |
| Phase assumptions | `/gsd:list-phase-assumptions` | Surface Claude's assumptions before planning |
| Nyquist validation | `/gsd:validate-phase` | Audit and fill validation gaps for a completed phase |
| Milestone audit | `/gsd:audit-milestone` → `/gsd:plan-milestone-gaps` | Verify intent delivered, fill gaps |
| Ship / PR creation | `/gsd:ship` | Create PR, run review, prepare for merge |
| Clean PR branch | `/gsd:pr-branch` | Filter out `.planning/` commits for clean code review |
| Safe revert | `/gsd:undo` | Roll back phase or plan commits using phase manifest |
| Project statistics | `/gsd:stats` | Phases, plans, git metrics, timeline |
| Session report | `/gsd:session-report` | Token usage estimates and work summary |
| Docs generation | `/gsd:docs-update` | Generate/update project docs verified against codebase |
| Changelog | `changelog-generator` (SP) | Auto-generate user-facing changelogs from git commits |
| Diagnostics | `/gsd:health` | Diagnose and repair `.planning/` directory issues |
| Post-mortem | `/gsd:forensics` | Investigate failed GSD workflows |
| Learnings extraction | `/gsd:extract_learnings` | Extract decisions, patterns, surprises from completed phases |

## Idea Capture & Backlog

| Scenario | Entry Point | Notes |
|----------|-------------|-------|
| Quick note | `/gsd:note` | Zero-friction idea capture — append, list, or promote |
| Add todo | `/gsd:add-todo` | Capture task from conversation context |
| Check todos | `/gsd:check-todos` | List pending todos and select one to work |
| Backlog idea | `/gsd:add-backlog` | Park an idea in 999.x numbering |
| Review backlog | `/gsd:review-backlog` | Promote backlog items to active milestone |
| Plant seed | `/gsd:plant-seed` | Forward-looking idea with trigger conditions |
| Explore idea | `/gsd:explore` | Socratic ideation — think before committing |

## Browser & E2E Testing

| Scenario | Entry Point | Notes |
|----------|-------------|-------|
| Test web app | `webapp-testing` (SP) | Playwright-based: screenshots, form fills, link checks |
| Full browser automation | `playwright-skill` (SP) | Auto-detect dev servers, write test scripts |

## Two-Layer Model

- **GSD = macro layer**: project structure, milestones, phases, plans, state tracking, pause/resume
- **Superpowers = micro layer**: TDD, debugging methodology, verification, code review per task

GSD orchestrates WHAT to build and in what order. Superpowers enforce HOW to build each piece correctly.

## Torque Commands

| Command | Purpose |
|---------|---------|
| `/torque:status` | Where am I? Single view across GSD + SP state |
| `/torque:check` | Verify GSD + SP installation and detect misconfigurations |
| `/torque:next` | Smart next-action — reads state, suggests one command |
| `/torque:migrate` | Upgrade path to [Meridian](https://github.com/mattjaikaran/meridian) |

## Integration Rules (Summary)

1. **Plan location**: GSD uses `.planning/phases/`, SP uses `docs/plans/`. Never mix.
2. **Brainstorming**: Use `discuss-phase` for GSD work, `brainstorming` for standalone only. Never both.
3. **Subagent execution**: GSD `execute-phase` OR SP `subagent-driven-development`. Never nested.
4. **Debugging**: SP `systematic-debugging` runs first for all bugs. GSD deviation Rule 1 covers only incidental executor bugs.
5. **Code review after GSD**: Run `requesting-code-review` after GSD verify-work, before merge.
6. **Code review receiving**: When receiving review feedback, use `receiving-code-review` to verify suggestions before implementing blindly.
7. **UI discipline**: For frontend phases, run `/gsd:ui-phase` to produce UI-SPEC.md before implementation, then `/gsd:ui-review` after.
8. **AI discipline**: For AI/LLM phases, run `/gsd:ai-integration-phase` to produce AI-SPEC.md before implementation, then `/gsd:eval-review` after.
9. **Spec before plan**: For medium+ features, run `/gsd:spec-phase` between discuss and plan to lock falsifiable requirements.
10. **Spike before commit**: When feasibility is uncertain, run `/gsd:spike` before discuss-phase. Wrap findings with `/gsd:spike-wrap-up`.
11. **Ship cleanly**: Use `/gsd:ship` or `/gsd:pr-branch` to create PRs that exclude `.planning/` commits from code review.
12. **Learnings persist**: GSD captures execution patterns automatically. Review with `/gsd:intel` to inform future phases.

## Anti-Patterns (Never Do These)

- Run `brainstorming` inside a GSD phase (use `discuss-phase`)
- Create `docs/plans/` in a GSD project (use `.planning/phases/`)
- Nest SP `subagent-driven-development` inside GSD `execute-phase`
- Skip `systematic-debugging` for bugs in GSD projects
- Create `.planning/` manually (use `/gsd:new-project`)
- Use `/gsd:quick` for multi-phase work (use full pipeline)
- Merge without code review (GSD has no review step built in)
- Run both `discuss-phase` AND `brainstorming` (pick one based on `.planning/` existence)
- Implement review feedback without verifying it first (use `receiving-code-review`)
- Skip spec refinement for ambiguous phases (use `/gsd:spec-phase` between discuss and plan)
- Build AI features without evaluation strategy (use `/gsd:ai-integration-phase` first)
- Push `.planning/` commits in PRs (use `/gsd:pr-branch` to filter them out)
- Commit to a large approach without spiking first (use `/gsd:spike` for uncertain feasibility)

## Complexity Escalation

```
Trivial (typo, config) → /gsd:fast
Small (< 30 min, < 5 files) → /gsd:quick
Medium (single phase) → discuss → spec → plan → execute
Large (multi-phase) → new-project or add-phase pipeline
Uncertain feasibility → /gsd:spike first, then route by size
AI/LLM integration → /gsd:ai-integration-phase → plan → execute → eval-review
Autonomous (hands-off) → /gsd:autonomous
```

For detailed workflows and conflict resolution, load files from `references/`. Workflow guides are in `references/workflows.md`.

Files in this skill

  • CHANGELOG.md4.3 KB
  • SKILL.md9.4 KB
  • references/comparison.md4.9 KB
  • references/conflict-resolution.md6.4 KB
  • references/decision-flowchart.md11.6 KB
  • references/integration-rules.md8 KB
  • references/workflows.md12 KB

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.