Correctless plugin
A Claude Code plugin for development with separated roles: spec, review, TDD, QA and verification by different agents
- 66 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- Hooks block and allow file writes by phase, and `/cauto` and `/cchores` carry work through to a PR
- Commands run several agents and test runs, so token spend is high
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add plugins/correctlessDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- Hooks block and allow file writes by phase, and `/cauto` and `/cchores` carry work through to a PR
- Commands run several agents and test runs, so token spend is high
This is third-party code. Review the repository files before installing.
What it does
Correctless builds a pipeline around the agent so the same AI does not plan, write, test and check itself. The main command chain: `/cspec` fixes what counts as correct, `/creview` attacks the spec before code exists, `/ctdd` drives development through tests, `/cverify` checks code against the spec and `/cdocs` updates documentation. The agent that writes tests does not know the implementation plan, and the QA agent did not write the code. Shell hooks hold the phases: `workflow-gate.sh` blocks source edits in the red stage and all writes during QA, `audit-trail.sh` keeps a log and `sensitive-file-guard.sh` protects marked paths. First setup runs with `/csetup`. The project accumulates memory: postmortems, QA and audit findings, deferred review items and antipatterns feed future specs and reviews. For critical work there is a raised intensity with six agents reviewing the spec. `/cauto` runs from an approved spec to a PR, `/cchores` picks a suitable issue and opens one PR, and `/cstatus` and `/cmetrics` show phase, cost and findings.
Who it is for. Developers and teams on Claude Code who care about trusting agent-written code and want independent checks at each stage.
Good fit when
- A feature is complex and testable rules are needed before code is written
- The agent's tests mirror the implementation too closely and red tests are wanted before the plan
- Caught bugs should turn into antipatterns and rules for later specs
- A run from an approved spec to a PR is needed with boundaries and a decision log
Not a fit when
- A one-line fix or quick prototype: a multi-agent pipeline will not pay off
- The project has no runnable test suite and the plugin needs one
- There is no Claude Max subscription, which the token spend assumes
Example request
/cspec Add CSV report export with a 10,000-row limit and write the correctness rules, then run `/creview`Limitations
It needs Claude Code, a Claude Max subscription, `jq` 1.7 or newer, Bash 4 or newer and a project with a runnable test suite. Many agents per feature mean noticeable token spend, especially at high and critical intensity. The Bash hooks block edits by phase and can be inconvenient when you need to bypass a task quickly by hand. Critical intensity adds Alloy modeling and live red-team assessment, which need separate preparation. After a plugin update, Claude Code is restarted and `/csetup` is run again.
How to disable. Run `/plugin uninstall correctless` and `/plugin marketplace remove correctless`. For a clone install, delete `.claude/skills/workflow` and the hooks from `.claude/settings.json`.
FAQ
How do I install Correctless in Claude Code?
Run `/plugin marketplace add joshft/correctless`, then `/plugin install correctless` and `/csetup`. The alternative is to clone the repository into `.claude/skills/workflow` and run `setup`.
Which commands make the main cycle?
`/cspec`, `/creview`, `/ctdd`, `/cverify` and `/cdocs`: spec, attack on it, tests and implementation, verification and docs. There are also `/cdebug`, `/cauto` and `/cchores`.
How do the hooks keep phases?
`workflow-gate.sh` blocks source edits at the red stage and all writes during QA, and `workflow-advance.sh` alone owns phase transitions. Order then does not depend on prompt discipline.
How do I raise check strictness?
In `.correctless/config/workflow-config.json` under `workflow` set `"intensity": "high"` or `"critical"`. The default is standard intensity.
How do I update the plugin?
Remove the plugin and marketplace with `/plugin uninstall` and `/plugin marketplace remove`, add them again, restart Claude Code and run `/csetup`. For a clone, `git pull` and `./setup` are enough.
Discussion
No questions or reviews yet.
Related
PluginsA plugin that turns Claude Code into a 16 agent team: documentation first planning, TDD and 9 pre-merge quality gates.
A plugin for Claude Code: you read one spec and parallel agents ship it to a commit with green tests.
A Claude Code working configuration: a planner with per-milestone review, sub-agents and skills for analysis, refactoring and docs sync
Claude Code plugin with 17 skills: task sizing, design review, TDD, fresh-context subagents and a PR after three human decisions.
A ready-made Claude Code setup: agents, slash commands, skills, safety hooks, rules and MCP connections in one install
Claude Code kit with eight phase agents, 21 skills and hooks that keep the agent inside set limits.
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| Correctless (this card) | no | 66 | |
| Superpowers | no | 316,679 | general development and planning skills, whereas this enforces strict role separation of agents and phases through hooks |
| TDD Guard | no | 2,375 | a single hook that enforces TDD, whereas this is a full pipeline from spec to verification |
README in short
The README sets out why the project exists: the same agent should not review its own code. It then covers quick start, phases, hooks, intensity, automation, updating, uninstalling and the license.