AIAgentMinder
AIAgentMinder (AAM) is an opinionated governance layer for Claude Code — a minder for your AI coding agent. Built for the solo developer, it gives Claude Code the engineering discipline a team would normally supply, so AI-assisted work stays tested, reviewed, and in scope. It's a set of skills, subagents, and hooks that let you enforce your standards without repeating your instructions, stop a session before context degradation sets in, and run autonomous sprints that span hours and multiple sessions without losing the thread.
Basically, I wanted Claude Code to have the skills I use daily as a developer — and figured a few product-manager tasks belonged in there too. This plugin lets you effectively manage AI development of large, complex projects that can't be finished in one to five sessions. When a project needs 100+ PRs over the span of months, you need consistency from your tooling, self-updating memory, the ability to change direction quickly, fast quashing of the defects that inevitably arise, and deterministic enforcement of quality.
Installs easily into existing or new repos. Works from a back-of-the-napkin sketch of an idea, or a fully spec'd plan.
The 30-second case
- Run a sprint, walk away.
Start a sprint for Phase 2. AAM plans, takes your approval at the spec gate, then executes item-by-item in isolated git worktrees until the PRs are ready for review or even merged.
- A different model reviews your code before you do.
/aiagentminder:self-review runs the change through five specialist lenses (security, performance, API design, cost, UX friction) plus a cross-model judge pass.
- Clean cross-session resume.
/aiagentminder:handoff writes the resume state into Claude Code's native Auto Memory; in the next session, "resume work" picks up exactly where you stopped.
- Plan and decision discipline by default.
scope-check keeps work aligned with the roadmap, DECISIONS.md logging keeps the record honest, BACKLOG.md captures everything you can't do right now.
Quick Start
Step 1 - Install
In Claude Code:
/plugin marketplace add lwalden/AIAgentMinder
/plugin install aiagentminder@lwalden-aiagentminder
Step 2 - Configure
In your target project:
/aiagentminder:setup
/aiagentminder:setup runs interactively — it fingerprints your codebase, asks for project identity, writes a starter CLAUDE.md, DECISIONS.md, BACKLOG.md, and docs/strategy-roadmap.md, and seeds .claude/rules/ and .gitignore.
Suggested Next Steps
/aiagentminder:brief to draft a product brief and strategy roadmap interactively.- Start building. Start a new session with 'claude --agent sprint-master' then prompt 'start sprint 1'
Reference
Skills (slash commands)
| Command | Purpose |
|---|
/aiagentminder:setup | Bootstrap a project (one-shot install / re-sync) |
/aiagentminder:brief | Interview-driven product brief and strategy roadmap |
/aiagentminder:revise | Mid-stream roadmap revision with DECISIONS.md logging |
/aiagentminder:scope-check | Compare proposed work against the roadmap; promote to backlog if out of scope |
/aiagentminder:milestone | Phase health assessment — progress, timeline, scope drift |
/aiagentminder:tdd | Guided TDD — plan, tracer bullet, RED-GREEN loop, refactor |
/aiagentminder:grill | Plan interrogation — walk every decision branch before implementing |
/aiagentminder:triage | Structured bug triage — reproduce, diagnose, fix plan, GitHub issue |
/aiagentminder:quality-gate | Pre-PR quality checklist (build, tests, coverage, lint, security) |
/aiagentminder:self-review | Pre-PR review via 5 specialist sub-agents + cross-model judge pass |
/aiagentminder:pr-pipeline | Autonomous PR review-fix-test-merge pipeline |
/aiagentminder:handoff | Session checkpoint — writes "Next Session" into Auto Memory |
/aiagentminder:retrospective | Sprint metrics and adaptive sizing guidance |
/aiagentminder:backlog | Capture, list, promote BACKLOG.md items |
/aiagentminder:sync-issues | Optional GitHub Issues bridge |
Sub-agents (Task tool dispatch)
