Gangsta Agents plugin
A plugin for spec-first development: six phases, subagent roles and a decision ledger, with a human approving each phase.
- 87 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- The plugin launches subagents and parallel code execution, but each phase waits for approval.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add plugins/gangsta-agentsDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The plugin launches subagents and parallel code execution, but each phase waits for approval.
This is third-party code. Review the repository files before installing.
What it does
The `gangsta` plugin installs a set of skills and six subagents for spec-driven development. Work follows a pipeline of six phases: reconnaissance of code and requirements, a debate on the solution between a proposer and an opponent, a spec written without code, a split into work packages, parallel TDD execution and a final reconciliation. Each phase ends with a gate and the agent moves on only after a human approves. Extra skills cover debugging, TDD, isolated git worktrees, code review by an inspector and verification before claiming completion. A family ledger keeps successful reasoning paths, failures and accumulated prohibitions between tasks.
Who it is for. Developers running large tasks in Claude Code or Cursor who want to agree the spec first and hand code to subagents only after that.
Good fit when
- The feature is large and a spec approved by a human comes first.
- An idea should be tested by a debate between two agents before any code is written.
- Work is split into packages and run in parallel with mandatory TDD.
- A shared memory of past failures and decisions is wanted for a team of agents.
Not a fit when
- A two-line fix: six phases with approval will only slow it down.
- The agent is meant to run without stops: every phase waits for a human reply.
- The team does not use specs and tests as the basis of its process.
Example request
Run reconnaissance on this repository, then debate the approach to the new module and stop for spec approval.Limitations
The plugin imposes a strict process and spends many tokens on subagents and approval phases, so it is excessive for small tasks. The principle that the spec is law requires code to follow the document and any process failure sends work back to the spec phase. Installation differs by agent: commands for Claude Code, Cursor, Codex, OpenCode and Gemini CLI vary, and for Codex and OpenCode the instructions are read from a link. The author does not describe behavior with non-English specs.
How to disable. In Claude Code run `/plugin uninstall gangsta@gangsta-marketplace`. Remove skills installed through `npx skills` from the agent skills directory and clean the accumulated ledger in the project by hand.
FAQ
How do I install the plugin in Claude Code?
Run `/plugin marketplace add kucherenko/gangsta`, then `/plugin install gangsta@gangsta-marketplace`.
Which phases does a task pass through?
Reconnaissance, debate, spec, work package split, execution and final reconciliation. Each phase waits for human approval.
What do the subagents do?
Six roles: an associate, a devil's advocate, a proposer, a soldier, a synthesizer and an inspector who reviews the work.
What does the ledger store?
Successful reasoning paths, failures with a diagnosis and accumulated prohibitions that later tasks reuse.
Does the plugin work outside Claude Code?
Yes. The README gives commands for Cursor, Codex, OpenCode, Gemini CLI, Copilot and Antigravity.
Discussion
No questions or reviews yet.
Related
PluginsClaude Code plugin with 60 subagents and 62 team presets, plus a browser designer for planning agent teams visually.
A multi-agent harness that verifies results by artifacts: stop hooks, an independent judge and an event log for Claude Code, Codex, Cursor and more
A skill plus hooks that keep a coding agent on task and cut speculative defenses, unread checksums and scope creep
A plugin with one skill for designing, debugging and evaluating AI agents, based on Bojie Li book on agent engineering.
A plugin marketplace for agents: reflexion, code and PR review, git, TDD, DDD, specs and subagent-driven development
A Claude Code plugin with four working-method skills: classify the task, act in steps, verify the result and generate domain adapters
README in short
The README explains the philosophy, the six-phase pipeline, the role hierarchy, the ledger and five governance rules, then lists installation for each agent.