From aisa
Use this skill when initializing or rebuilding AI-assisted development configuration for a project. Scans the project tech stack and generates a complete .claude/ setup (CLAUDE.md, skills, agents, learnings, cache) from scratch.
How this skill is triggered — by the user, by Claude, or both
Slash command
/aisa:aisa-init [specs-path][specs-path]The summary Claude sees in its skill listing — used to decide when to auto-load this skill
Build the complete `.claude/` configuration for this project from scratch.
Build the complete .claude/ configuration for this project from scratch.
Specs location: $ARGUMENTS (default: specs/ or openspec/ — auto-detected if not specified)
Before starting the pipeline, check whether an AI configuration already exists in this project.
Run the following checks:
test -f CLAUDE.md && echo "CLAUDE.md exists" || echo "CLAUDE.md missing"
test -d .claude && echo ".claude/ exists" || echo ".claude/ missing"
test -d .claude/skills && echo "skills/ exists" || echo "skills/ missing"
test -d .claude/agents && echo "agents/ exists" || echo "agents/ missing"
test -d .claude/learnings && echo "learnings/ exists" || echo "learnings/ missing"
test -d .claude/cache && echo "cache/ exists" || echo "cache/ missing"
If ANY of CLAUDE.md, .claude/skills/, or .claude/agents/ already exist, stop and present the following choice to the user:
Existing AI configuration detected:
✅ CLAUDE.md [present / missing]
✅ .claude/skills/ (N skills)
✅ .claude/agents/ (N agents)
What would you like to do?
1. Audit — review what exists, report gaps, suggest improvements (non-destructive)
2. Rebuild — remove existing .claude/ and CLAUDE.md, regenerate from scratch
⚠️ Option 2 will DELETE all existing skills, agents, and learnings. This cannot be undone.
Wait for user choice:
aisa-audit). Do NOT proceed with setup.Read the full pipeline specification in REFERENCE.md (in this skill's directory) and execute it.
The pipeline has 6 phases with 2 mandatory quality gates:
Phase 1 — Discovery → scan project structure, docs/, specs/, code, learnings
Phase 2 — Architecture Design → propose skills & agents across technical/business/design
Phase 3 — Architecture Critique ← QUALITY GATE (must pass before generating)
Phase 4 — Generation → produce all skill, agent, learnings, and CLAUDE.md files
Phase 5 — Generation Critique ← QUALITY GATE (must pass before wiring)
Phase 6 — Wiring & Validation → write files, verify references, commit
.claude/learnings/ infrastructure for continuous knowledge captureEvery skill and agent you generate MUST pass the principle checklists defined in
.claude/skills/aisa-principles/SKILL.md (Skill Principles P1-P3, Agent Principles A1-A6).
Validate before completing Phase 5. Exception: openspec-* skills are exempt from Quality Gates.
If any generated file fails these checks → fix it before moving to Phase 6. Do not defer.
Every phase MUST produce a visible section header and summary before pausing or proceeding. Use this structure (adapt content to findings):
## Phase 1 — Discovery Report
{project structure, tech stack, specs found, existing learnings — 4-8 lines}
[PAUSE — awaiting approval to proceed to Phase 2]
## Phase 2 — Architecture Design
{N skills proposed across technical/business/design dimensions; agents: N proposed or "none (Rule 7 not met)"}
## Phase 3 — Architecture Critique ← QUALITY GATE
### Critique Findings
{specificity check, domain coverage, principle compliance — list any issues found}
### Simulation
{pick one proposed skill, walk through a representative task to verify it adds value}
### Verdict
{PASS / FAIL with revision required}
[PAUSE] **Please review the Architecture Critique above and confirm you'd like to proceed to file generation, or request revisions.**
## Phase 4 — Generation
{list of files being generated with brief rationale for each — e.g., .claude/skills/foo/SKILL.md, .claude/agents/bar.md, .claude/learnings/log.md, CLAUDE.md}
## Phase 5 — Generation Critique ← QUALITY GATE
### Critique Findings
{principle checks P1-P3/A1-A6 per generated file; spot-check one file with verify command}
### Verdict
{PASS / FAIL}
[PAUSE — awaiting approval to proceed to Phase 6]
## Phase 6 — Wiring & Validation
{files written; references verified; commit summary}
## Learnings
{discoveries captured in .claude/learnings/log.md, or "No new patterns detected — learnings
infrastructure created for future entries"}
Before delivering any phase output, perform an internal verification:
## Learnings section appears in the final output (even if it states "No new patterns detected")If any item fails, correct it before presenting output.
After Phase 1 (Discovery Report) — present findings, wait for approval. After Phase 3 (Architecture Critique) — present critique, wait for approval. After Phase 5 (Generation Critique) — present quality scores, wait for approval.
During Phase 2 (Architecture Design), assess the planned topology size and recommend an
execution mode for aisa-sync lifecycle. Include in CLAUDE.md if Agent Teams are warranted.
Creates:
.claude/skills/ — project-specific portable expertise files.claude/agents/ — autonomous executor definitions.claude/learnings/log.md — learning journal.claude/learnings/README.md — learning system docs.claude/cache/snapshot.json — initial cache for incremental evolutionCLAUDE.md — project configuration with spec-driven workflowAfter Phase 6 (Wiring & Validation) completes, run the following verification steps:
verify-setup.js by searching for **/verify-setup.js (Glob).node <plugin-path>/scripts/verify-setup.js health --project-root . --markdown
node <plugin-path>/scripts/verify-setup.js validate --project-root . --markdown
## Setup Complete — Verification Report
### Files Created
[list all files]
### Health Check
[paste output verbatim]
### Principle Compliance
[paste output verbatim]
### Overall Verdict
- Both scripts exit 0 → "Setup verified — all checks pass ✅"
- Any FAIL items → list each one explicitly
### Next Steps
- To add more skills: see docs
- If issues found: run `/aisa-lint`
- For scoped updates later: run `/aisa-update <area>`
npx claudepluginhub rnagrodzki/ai-setup-automation --plugin aisaCreates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.