analyze-project-architecture

What it does

Transforms raw project data into a structured architectural view by discovering modules, enriching descriptions from documentation, and persisting enriched data for consumption.

How it works

• Discovery: Run extension API discovery via the script suggested:
  • python3 .plan/execute-script.py plan-marshall:analyze-project-architecture:architecture discover --force
  • Output: .plan/project-architecture/derived-data.json (overwrites existing data).
• Step 2 initializes an enrichment file depending on existence of llm-enriched.json using:
  • python3 .plan/execute-script.py plan-marshall:analyze-project-architecture:architecture init or
  • architecture init --check to see if enrichment exists, with user prompting to replace or skip.
• Step 3 loads discovered data in TOON format via:
  • python3 .plan/execute-script.py plan-marshall:analyze-project-architecture:architecture derived.
• Step 4 enriches project description from README and module descriptions by:
  • python3 .plan/execute-script.py plan-marshall:analyze-project-architecture:architecture enrich project --description "{extracted project description}" --reasoning "{source: README.md introduction | inferred from module names | pom.xml description}".
• Step 5 collects module list via:
  • python3 .plan/execute-script.py plan-marshall:analyze-project-architecture:architecture modules.
• Step 6 enriches each module through 6a–6e, including reading module docs, determining purpose, writing module responsibility with reasoning, identifying key packages and dependencies, and assigning skills by profile with optional documentation profiling.
• Step 6e.3–6e.5 cover retrieving skills by domain, filtering based on module signals, and applying skills-by-profile including documentation as a separate profile when docs exist.

When to use it

Use when you need an LLN-based architectural analysis that converts raw module data into a structured, enriched description suitable for a solution-outline.

What it can touch

No specific touches are listed beyond the script-driven orchestration in the steps. The scripts reference modules, derived-data.json, llm-enriched.json, README.md, pom.xml, and package-info as sources for enrichment. The allowed-tools are Read, Write, Edit, Glob, Grep, Bash.

Caveats

License is MIT. The workflow requires user prompts in Step 2 when enrichment already exists and Step 1.5 only applies for Maven unless skipped. No outcomes are promised beyond the described steps and outputs. The skill enforces execution in sequence and does not skip modules without enrichment. All reasoning and descriptions are anchored to provided source material (README, module descriptions, derived data).