procoder logo

procoder

Senior-developer discipline for AI coding agents. A commit gate that counts unchecked as failing, quality controllers that refuse to call unfinished work done, and a lessons loop that closes each escaped bug's class. One Go binary, no runtime deps, works with 20+ agents.

version 3.4.0works with 20+ agents
Website GitHub

What is it?

What it is

Procoder is a Go‑based AI coding assistant plugin that turns any agent into a senior developer by enforcing commit‑gate quality, self‑serve tooling, and a self‑learning loop.

Why it exists

AI agents often accept “done” without rigorous review. Procoder adds a refusal‑based controller that blocks unsafe, unformatted, or unchecked work, ensuring every change survives a senior‑dev standard before it can be committed.

Who should use it

Go developers, AI-assisted coders using Claude Code, Cursor, Windsurf, Cline, or other supported agents, engineering teams enforcing code quality and commit gates, repositories seeking automated static analysis, security scanning, and documentation health checks

Who should avoid it

Non-Go projects without supported language formatters, teams unwilling to adopt structured quality chains or commit gates, users requiring real-time MCP server integration, projects incompatible with Apache-2.0 licensing

How it works

A quick walkthrough in plain English

How procoder works

Step 1 of 3

You interact with it

Open procoder, send a request, or connect it to your stack.

Features

Commit gate (procoder check) - formatting, git hygiene, secrets, lint, CI/infra checks
Quality chain with refusing controllers - spec interview, plan, backlog, sprints, todo
Self-learning loop - pre-PR review, lessons ledger, copilot-leak integration
Test domain - runs canonical test suites per ecosystem with policy enforcement
Release controller - version sync, changelog, tree clean, gate clean, suite green
Ten security and quality domains - security, lint, maintainability, performance, docs, testing, CI, infra, GitOps
Code index - ctags + SCIP for symbol navigation (find, refs, callers, impact)
Senior engineering habits - build ladder principles, debt tracking, audit command
P-CONTROL mode - binary computes/reports, agent acts, no code modification behind back
Multi-agent support - Claude Code, Cursor, Windsurf, Cline, Kilo Code, Roo, Kiro, Codex CLI, Copilot CLI, Gemini, OpenCode
Self-serve tools (check, format, lint, scan, query) and forced lifecycle hooks
Single Go binary with no runtime dependencies, cross-platform, air-gapped compatible
Repository-scoped configuration via .procoder/ directory with project-level overrides
Comprehensive formatting support for 16+ languages

Advantages

  • Consistent results across CLI, git hooks, and CI pipelines
  • Agent remains in full control - nothing modifies code behind the agent's back
  • One binary replaces multiple plugins and tools
  • Scope-limited checks respect non-adopted repositories
  • Controllers that refuse instead of just advising
  • Pre-PR self-review catches issues before human review
  • Unchecked items are never reported as clean (counts as failing)
  • Self-learning from escaped bugs creates permanent improvements
  • Works with 20+ different AI coding agents
  • Zero runtime dependencies enables air-gapped and CI environments
  • All findings delivered in one turn with fixed content
  • Multi-language formatting with project config priority
  • Audit command for onboarding existing codebases

Disadvantages

  • External dependencies for some security scans (gitleaks, semgrep, osv-scanner)
  • Performance benchmarking limited to Go only
  • May be overly restrictive for small or experimental projects
  • Version 3.4.0 indicates still maturing
  • Complexity could overwhelm new users or small teams
  • Full features require explicit repository adoption
  • Secrets detection relies on external tooling
  • Tight integration means less flexibility for preferred tools
  • Some checks may false-positive on unconventional project structures

Installation

FAQ

How do I install Procoder?

For Claude Code, run `/plugin marketplace add azrtydxb/procoder`, then `/plugin install procoder`, and finally `/procoder:init`. Procoder also ships adapters for many other agents.

Which AI coding agents does Procoder support?

It works with 20+ agents including Claude Code, Cursor, Windsurf, Cline, Kilo Code, Roo, Kiro, Codex CLI, Copilot CLI, Gemini, OpenCode, and any agent that reads `AGENTS.md`.

What does the `procoder check` command do?

It runs a commit gate that checks formatting, git hygiene (conflict markers, junk files, oversized files, AI-attribution lines), secrets, lint, CI and infra hygiene, and documentation health. It reports findings and blocks if there are blocking issues.

How does Procoder prevent the agent from modifying code without approval?

Procoder uses a P-CONTROL model: the binary computes and reports findings, but the agent acts on them. Nothing modifies code, files, or state behind the agent's back. The only exception is `procoder prune --apply`, which removes superseded plugin copies from the cache.

What is the self-learning loop?

It's a pre-PR self-review that catches reviewer-class findings before a PR exists. Any bug that escapes becomes a "lessons" ledger entry, requiring an adaptation (linter rule, rubric line, pinning test) before the work counts as done. Copilot's auto-review findings can also be collected and filed as issues.

How does Procoder handle repositories that haven't adopted it?

In a repository without `.procoder/` or an `AGENTS.md` naming Procoder, only checks that are universally true run (secrets, oversized files, conflict markers, junk, AI-attribution lines). Content-reading checks only see lines your commit wrote. Procoder does not apply its conventions unasked.

Loading documentation…
View on GitHub

Featured in Videos

YouTube tutorials and walkthroughs for procoder

Alternatives

Similar projects ranked by category, topics, and text overlap.

Compare
procoder | MushyBook