AIAgentMinder
A governance framework for Claude Code. Slash commands, rules files, hooks, and shell scripts that add sprint execution, quality enforcement, and context management to Claude Code projects.
What this is: An installable template — no runtime dependency, no MCP server, no database. Built for a solo developer who wants Claude Code to execute autonomously while maintaining engineering discipline.
Command prefix: All commands use the aam- prefix (e.g., /aam-brief, /aam-handoff) to avoid collision with Claude Code built-ins and other plugins.
Quick Start
Install
# Recommended: npx installer (interactive — asks about your project, copies files)
npx aiagentminder init
# Or with all optional features enabled (no prompts)
npx aiagentminder init --all
# Or core governance only (no sprint planning, no architecture fitness)
npx aiagentminder init --core
Alternative: clone the repo and run /aam-setup from Claude Code.
After install
- Run
/aam-brief to create your product brief and strategy roadmap.
- Start building: "Read CLAUDE.md and docs/strategy-roadmap.md, then start Phase 1."
With sprint planning enabled: "Start a sprint for Phase 1."
What It Does
AIAgentMinder addresses three gaps in Claude Code:
- Autonomous execution. Sprint-driven development with spec approval gates, then fully autonomous execution — TDD, quality gates, self-review, PR creation, and merge without permission prompts.
- Quality enforcement. Mandatory TDD, quality gates, self-review (with optional cross-model validation), and scope checks that Claude cannot skip. Engineering practices that catch defects early.
- Context management. Real-time token monitoring, autonomous context cycling across sessions, and session handoff — preventing the quality degradation that happens as context fills up.
How It Works
Core files
| File | Purpose |
|---|
CLAUDE.md | Project identity, behavioral rules (auto-loaded every session) |
DECISIONS.md | Architectural decisions, rationale, known debt log |
docs/strategy-roadmap.md | Product brief — features, phases, out-of-scope items |
.claude/rules/*.md | Universal rules (auto-loaded every session) |
SPRINT.md | Active sprint header and issue table (optional) |
BACKLOG.md | Work inbox — capture future work via /aam-backlog |
.pr-pipeline.json | PR pipeline config — high-risk patterns, merge method, cross-model review |
Rules
Universal rules (loaded every session, all modes):
| Rule | What it does |
|---|
git-workflow.md | Branch naming, commit discipline, PR-only workflow |
tool-first.md | Use CLI/API tools instead of asking the user to do things manually |
correction-capture.md | Detects repeated wrong-first-approach patterns and proposes permanent rules |
context-cycling.md | Procedure for context cycling when the PreToolUse hook fires |
Mode-specific rules (loaded via session profile agents):
| Rule | Loaded by | What it does |
|---|
| Sprint workflow | sprint-executor | State machine sprint execution with mandatory quality checklist |
| Code quality | sprint-executor, dev, qa | TDD cycle, build-before-commit, small focused functions |
| Scope guardian | sprint-executor, dev | Checks new work against the roadmap before implementing |
| Approach-first | sprint-executor, dev | States intended approach before multi-file or architecture changes |
| Debug checkpoint | sprint-executor, dev, debug, hotfix | Stops debugging spirals after 3 failed attempts at the same error |
| Architecture fitness | sprint-executor, dev, qa | Structural constraints — file size limits, secrets detection, layer boundaries |
Session profiles
Use claude --agent <name> to load the right context for your task:
| Profile | Purpose |
|---|
sprint-executor | Full sprint state machine — all rules loaded |
dev | General development — TDD, architecture fitness, approach-first |
debug | Debugging — checkpoint pattern, triage |
hotfix | Minimal ceremony — debug checkpoint only |
qa | Quality review — code quality, architecture fitness |
Commands
lwalden