agentic-coding-harness

Automated PRD-to-commit coding workflows are driven by generating PRD-scoped workpacks that agents execute autonomously, applying quality gates like typecheck, tests, and commit-on-green rules. You initiate this by running a slash command with your project goal to generate the workpack.

A workpack in autonomous coding is a PRD-driven collection of scoped task files that structures end-to-end project execution. It allows agents to autonomously plan, build, and verify software by converting high-level goals into actionable, queued tasks without manual orchestration.

Enforcing file size limits and typecheck rules during autonomous development is handled by built-in technical quality gates, which include 300-lines-per-file limits, directory size constraints, and mandatory typecheck and test execution before applying commit-on-green discipline.

Generating AGENTS.md files and extracting prompt templates from an existing codebase is supported through Python CLI instruments. These tools automate the creation and maintenance of AGENTS.md files and extract reusable prompt templates for both autonomous and human-in-the-loop runs.

Autonomous coding works for both fresh project builds and codebase audits, applying PRD-driven workflows to new development and existing code analysis. It supports state machine visualization, session logging, and operational tasks across both project types.

Maintaining context and reproducibility across autonomous coding sessions is achieved through structured session logging and AGENTS.md updates. This production-grade workflow reduces context loss by recording session starts, ends, and operational states for every autonomous run.