Full-Auto Agentic Coding Workflow

Problem it addresses

AI coding assistants can jump straight into implementation, leaving requirements, specs, test adequacy, and merge decisions to humans. This skill frames the flow from idea to pull request as an engineering process with limited human checkpoints.

How it works

• On trigger, it silently checks the environment for Product-Manager-Skills, such as discovery-process or prd-development, and for cc-sdd signals like kiro-discovery or /kiro- commands.
• It checks project basics: CLAUDE.md, SPEC.md, .gitignore, and whether a test command can run; missing items are surfaced or degraded conservatively.
• It grades task size, routes small fixes through a shorter path, and uses requirement discovery, specification, autonomous implementation, testing, and PR review for larger work.
• The four human checkpoints are: confirm requirements, choose the technical approach, judge test sufficiency, and review the PR before merge.

Boundaries

It is useful when the project has enough context for an agentic coding loop, but it does not replace architecture review, security validation, or business acceptance. If tests, boundaries, or supporting skills are missing, the workflow falls back to more cautious prompting and manual follow-up.