A Claude Code plugin marketplace that ships the ai-setup-automation plugin for creating and maintaining AI-ready project configurations.
- Detects your tech stack and scaffolds a
.claude/directory withCLAUDE.md, skills, agents, and settings - Provides 11 skills in three categories — Essentials, Analysis, and Utilities — covering the full lifecycle: setup, review, sync, checks, and post-incident learning
- Manages a cache layer to reduce token consumption by 60–80% on repeated audits
- Keeps your AI configuration in sync with your codebase as it evolves
| Requirement | Version | Notes |
|---|---|---|
| Claude Code | — | This is a Claude Code plugin marketplace |
| Node.js | >= 16 | For cache-snapshot.js and verify-setup.js scripts. Uses built-in modules, no npm install needed |
| git | — | Assumed for most features |
- Open Claude Code and run
/plugin - Go to Marketplaces → Add marketplace → enter
rnagrodzki/ai-setup-automation - Go to Discover → select
aisa→ Install
/plugin marketplace add rnagrodzki/ai-setup-automation
/plugin install aisa@ai-setup-automation
See docs/getting-started.md for a full first-use walkthrough.
Open /plugin, go to the Marketplaces tab, and toggle auto-update for ai-setup-automation.
/plugin marketplace update ai-setup-automation
/plugin update aisa@ai-setup-automation
If you installed the plugin before this naming fix, uninstall the old entry and reinstall:
/plugin uninstall ai-setup-automation@ai-setup-automation
/plugin install aisa@ai-setup-automation
- Navigate to your project directory
- Start Claude Code
- Run
/aisa-init - Follow the interactive prompts
The skill detects your tech stack, presents a setup plan for your approval, and scaffolds the full .claude/ directory.
| Skill | Description |
|---|---|
/aisa-init |
Scaffold a new .claude/ directory — detects tech stack, generates CLAUDE.md, skills, agents, learnings journal, and cache. One-time setup for new projects |
/aisa-sync |
Full maintenance cycle — runs drift detection, harvests learnings, identifies expansion needs, applies prioritized updates to .claude/ skills and agents |
/aisa-postmortem |
Incident-to-prevention pipeline — gathers context from conversation or git history, maps root causes to skill gaps, encodes lessons into .claude/ skills and learnings log |
| Skill | Description |
|---|---|
/aisa-audit |
Deep read-only review of .claude/ content accuracy — mechanical validation, outdated code example detection, skill specificity check, coverage gap analysis. Reports HEALTHY / NEEDS_ATTENTION / CRITICAL |
/aisa-inspect |
Quick read-only drift scan — compares .claude/ skills and agents against current codebase state, reports CURRENT / OUTDATED / STALE / CRITICAL per file. Run weekly |
/aisa-lint |
Validate skill and agent structure against architectural principles (P1-P3, A1-A6) — checks frontmatter, self-learning directives, PCIDCI pattern. Does not verify codebase accuracy |
/aisa-update |
Targeted skill/agent update after a specific code change — uses git diff to scope impact, updates only affected .claude/ files, flags out-of-scope drift without fixing it |
| Skill | Description |
|---|---|
/aisa-harvest |
Promote ACTIVE entries from .claude/learnings/log.md into skill gotchas, new skills, and documentation. Run when log has 10+ entries or oldest entry exceeds 2 weeks |
/aisa-cache |
Manage the .claude/cache/ snapshot used by inspect and sync — sub-commands: rebuild (default), status, invalidate. Reduces token use by 60-80% |
/aisa-spec-check |
Check openspec CLI availability, project initialization status, and version currency — suggests install/init/update with user confirmation |
Behavioral regression tests use promptfoo to validate plugin skills against expected outputs. Tests require Task and gum for interactive selection.
| Command | Description |
|---|---|
task test |
Run all skill tests |
task test:skill |
Select one or more skills to test (interactive) |
task test:skill -- aisa-lint |
Test a specific skill directly |
task test:skill -- aisa-lint aisa-cache |
Test multiple skills directly |
task test:view |
Open the promptfoo web UI to inspect results |
| Document | Description |
|---|---|
| Getting Started | Installation, first use, what gets created |
| Architecture | Repository structure, plugin system, name resolution |
| Plugin: ai-setup-automation | Skills reference, cadence, lifecycle, execution modes, principles |
| Adding Skills | Create custom skills for your project |
| Adding Commands | Create custom slash commands |
| Adding Hooks | Set up automated actions on session events |
A GitHub Actions workflow runs on every pull request targeting main and verifies that modified plugins have their version field bumped in plugin.json. The check:
- Detects which plugins have changed files in the PR
- Compares the
plugin.jsonversion against the base branch - Fails if a plugin's files changed but its version was not incremented
To skip the check when a version bump is intentionally not needed, add the skip-version-check label to the pull request.
This happens when the plugin name registered in the marketplace doesn't match the identity in plugin.json. Clear the cache, restart, and reinstall:
rm -rf ~/.claude/plugins/cache/ai-setup-automationThen restart Claude Code and run:
/plugin install aisa@ai-setup-automation
The version field in plugin.json must be bumped for Claude Code to detect a new version. If the version hasn't changed, Claude Code uses the cached copy. See the CI Checks section — every PR that modifies plugin files must bump the version.
Open /plugin, go to the Marketplaces tab, and verify auto-update is toggled on for ai-setup-automation. Auto-update is off by default for third-party marketplaces.
Large repositories may exceed the default git timeout. Set the environment variable before starting Claude Code:
export CLAUDE_CODE_PLUGIN_GIT_TIMEOUT_MS=300000