Project Architect

Output Contract

Produce a plan that is immediately executable, with:

• Requirements (functional + non-functional)
• Task breakdown (small, dependency-aware)
• Skill opportunities (what to automate/templatize)
• Runbook (commands + expected artifacts)
• Risks / unknowns / decision points
• Acceptance tests (verifiable)

Steps

1. Extract requirements

- What must exist when done? - What must NOT change? - Edge cases and failure modes.

1. Decompose into tasks

- Right-size: one agent iteration per task. - Separate investigation from implementation. - Include required tests/checks per task (what to verify).

1. Identify skill opportunities

- Look for repeated patterns that will recur across tasks. - Propose minimal new skills (avoid over-forging). - For each candidate skill, provide: - name, type (operational|meta|composed) - 1-2 sentence description (with triggers) - inputs/outputs - example prompts

1. Write runbook + acceptance

- Commands to run locally (build/test/lint/typecheck). - Acceptance checks that are machine-verifiable when possible.

Notes

• If this project uses Forgeloop: keep the plan aligned with IMPLEMENTATION_PLAN.md and Forgeloop's “single unit of work” iteration model.