omcustom:analysis

Analyze project and auto-configure agents, skills, rules, and guides

SKILL.md

GitHub

Project Analysis Skill

Scan a project's tech stack, compare against installed agents/skills, and auto-configure missing items.

Options

--dry-run      Show what would be added without making changes
--verbose      Show detailed detection reasoning
--interview, -i   Run interactive architecture interview before file-based detection

Workflow

Step 0: Architecture Interview (--interview only)

When --interview flag is provided, conduct an interactive AI interview before file-based detection. This captures human context that file scanning cannot determine.

Interview flow (sequential, AI-guided):

  1. 프로젝트 유형: "이 프로젝트는 어떤 종류입니까?" → 옵션: web app, REST API, CLI tool, library, monorepo, data pipeline, mobile app

  2. 아키텍처 패턴: "어떤 아키텍처를 따르고 있습니까?" → 옵션: microservices, monolith, serverless, event-driven, layered, hexagonal

  3. 주요 언어: "주로 사용하는 프로그래밍 언어는?" → 자유 입력, 알려진 에이전트와 매칭

  4. 배포 대상: "어디에 배포합니까?" → 옵션: AWS, GCP, Azure, Vercel, on-premises, Docker/K8s, edge

  5. 팀 우선순위: "팀의 주요 관심사는?" → 옵션: performance, security, developer experience, cost, scalability

Interview results feed into Step 1 as weighted detection hints:

  • File evidence + interview agreement = confidence: high
  • File evidence only = confidence: medium (unchanged from current)
  • Interview only (no file evidence) = confidence: suggested

Integration with report:

Interview Insights (--interview):
  Project type: REST API (user-specified, confirmed by file scan)
  Architecture: microservices (user-specified)
  Deployment: AWS + Docker (confirmed by file scan)
  Team focus: security → sec-codeql-expert [suggested]

Suggested (from interview, no file evidence):
  ~ sec-codeql-expert  [suggested — no CodeQL config found]
  ~ de-kafka-expert     [suggested — no kafka deps found]

Step 1: Project Scan

Detect tech stack by checking indicator files and dependency manifests.

Indicator Files to Check Agent Skill
TypeScript tsconfig.json, *.ts, *.tsx lang-typescript-expert typescript-best-practices
React/Next.js next.config.*, package.json (next dep) fe-vercel-agent react-best-practices
Vue.js vue.config.*, *.vue fe-vuejs-agent -
Svelte svelte.config.*, *.svelte fe-svelte-agent -
Flutter/Dart pubspec.yaml, *.dart, lib/main.dart fe-flutter-agent -
Python pyproject.toml, requirements.txt, *.py lang-python-expert python-best-practices
FastAPI "fastapi" in imports/deps be-fastapi-expert fastapi-best-practices
Go go.mod, *.go lang-golang-expert go-best-practices
Go Backend go.mod + cmd/ or internal/ dirs be-go-backend-expert go-backend-best-practices
Rust Cargo.toml, *.rs lang-rust-expert rust-best-practices
Kotlin *.kt, build.gradle.kts lang-kotlin-expert kotlin-best-practices
Java *.java, pom.xml lang-java-expert -
Spring Boot spring-boot in deps be-springboot-expert springboot-best-practices
Express.js "express" in deps be-express-expert -
NestJS "@nestjs" in deps be-nestjs-expert -
Docker Dockerfile, compose.yml infra-docker-expert docker-best-practices
AWS CDK/SAM/CloudFormation files infra-aws-expert aws-best-practices
PostgreSQL *.sql, pg in deps db-postgres-expert postgres-best-practices
Redis redis in deps db-redis-expert redis-best-practices
Supabase supabase in deps/config db-supabase-expert supabase-postgres-best-practices
Airflow dags/*.py, airflow in deps de-airflow-expert airflow-best-practices
dbt dbt_project.yml de-dbt-expert dbt-best-practices
Kafka kafka in deps/config de-kafka-expert kafka-best-practices
Spark spark in deps/config de-spark-expert spark-best-practices
Snowflake snowflake in deps/config de-snowflake-expert snowflake-best-practices

Detection logic:

1. Read package.json / go.mod / Cargo.toml / pyproject.toml / pom.xml
2. Glob for indicator files (tsconfig.json, *.vue, Dockerfile, etc.)
3. Grep dependencies for framework/library names
4. For verbose mode: log each indicator found and confidence level

Step 2: Gap Analysis

Compare detected stack against what is already installed.

1. List existing agents:  ls .claude/agents/*.md
2. List existing skills:  find .claude/skills -name "SKILL.md"
3. For each detected indicator:
   a. Check if required agent file exists → mark MISSING or PRESENT
   b. Check if required skill directory exists → mark MISSING or PRESENT
4. Build two lists:
   - missing_agents[]   — agents needed but not present
   - missing_skills[]   — skills needed but not present
5. (Optional) Build unused list for suggestions:
   - Agents present but no indicator matched → flag for review

Step 3: Auto-Configure

Apply changes for all missing items (skip in --dry-run mode).

For each missing agent:
  - If agent exists in templates/.claude/agents/ → copy to .claude/agents/
  - Else → delegate to mgr-creator with detected domain context

For each missing skill:
  - If skill exists in templates/.claude/skills/ → copy to .claude/skills/
  - Else → log as "skill not available in templates, manual setup needed"

Rules:
  - Keep all existing rules (they are universal, never remove)

Guides:
  - Verify templates/guides/ directory has relevant reference docs
  - Log missing guide topics as suggestions only (no auto-copy)

Step 4: Report

Output a structured summary after the run.

[analysis] Project: <detected project name or path>

Tech Stack Detected:
  - TypeScript (tsconfig.json found)
  - React/Next.js (next in package.json deps)
  - Docker (Dockerfile found)

Agents:
  + lang-typescript-expert  [added]
  + fe-vercel-agent          [added]
  ~ infra-docker-expert      [already present, skipped]

Skills:
  + typescript-best-practices  [added]
  + react-best-practices        [added]
  ~ docker-best-practices       [already present, skipped]

Rules:   no changes (universal rules kept as-is)

Guides:  react/ — present
         docker/ — present
         typescript/ — present

Suggestions:
  - infra-aws-expert not detected (no CDK/SAM files found)
  - de-* agents not detected (no pipeline indicators found)

Summary: 2 agents added, 2 skills added, 0 removed

--dry-run output prefixes all additions with [would add] instead of [added] and makes no file changes.

--verbose output adds a Detection section before the report:

Detection Details:
  tsconfig.json            → TypeScript confirmed
  package.json[next]       → Next.js confirmed (confidence: high)
  package.json[react]      → React confirmed (confidence: high)
  Dockerfile               → Docker confirmed
  no go.mod found          → Go skipped
  no Cargo.toml found      → Rust skipped

Step 5: Adaptive Harness Integration

After analysis completes, offer adaptive-harness optimization:

[Analysis Complete] Tech stack detected. Optimize harness for this project?
├── Yes → Run /omcustom:adaptive-harness --optimize (deactivate unused, suggest missing)
├── Dry-run → Run /omcustom:adaptive-harness --optimize --dry-run (show changes only)
└── Skip → Keep current harness configuration

If user selects Yes or Dry-run, invoke the adaptive-harness skill with the analysis results as context. The project profile generated by analysis feeds directly into adaptive-harness optimization.

Integration

Skill Integration
adaptive-harness Called after analysis to optimize harness based on detected stack

Example Invocation

/analysis
/analysis --dry-run
/analysis --verbose
/analysis --dry-run --verbose

Notes

  • Always run --dry-run first on unfamiliar projects to preview changes
  • Agents and skills are additive only — existing items are never removed automatically
  • For stacks not in the detection table, delegate to mgr-creator for dynamic agent creation
  • Rules are never auto-removed; they are project-universal