LintGate

Proactive professional cognition stack for LLM coding agents. Fires on every Write, Edit, and Bash via PostToolUse hooks. 118 MCP tools for on-demand analysis.

First Steps

1. Orient: build_theory_pack — get the project's theory digest (~500-1500 tokens)
2. Check: controlplane_run — 7-channel supervision mesh (lint + tests + deps + git + behavior + structure + test_effectiveness)
3. Reflect: constraint_check — state your constraints before acting; prediction_register — register falsifiable predictions

Key Concept

LintGate is not just a linter. It supervises the agent's reasoning strategy, not just its code output. The ControlPlane runs 6 independent channels in parallel — code quality, test health, dependency integrity, git hygiene, behavioral drift detection, and codebase structural analysis — with a coherence engine that reads the agreement and disagreement across channels as diagnostic signal. Theory extraction grounds findings in the project's own stated values. The Architecture of Inquiry closes the loop — behavioral discoveries flow back into the project's living semantic model. Habit Mode manages the context window itself — detecting sustained execution phases and producing structured compaction snapshots that turn context loss into context refinement.

Setup

bash setup.sh    # venv, install, hook, MCP config, agent integration — one command

This detects which LLM agents are on the system and generates config files at the appropriate integration depth.

Reference

• AGENTS.md — tool reference + self-integration for all agents (110 tools by cognitive mode)
• .claude/CLAUDE.md — Claude Code cognitive context: epistemic state, dispositions, guardrails, managed sections for living context. This is the deepest integration — behavioral discoveries flow back as patches.
• .claude/rules/inquiry.md — Architecture of Inquiry protocol
• docs/design.md — full architecture and design philosophy

For Claude Code, all three files work together. For other agents, AGENTS.md is the primary entry point and integrate.sh generates an appropriate cognitive context file.