ai dev methodology

Managing a hybrid squad of humans and AI agents leads to broken builds, race conditions, and undocumented shortcuts. This skill implements a markdown and git governance framework that treats AI agents as disciplined contributors, enforcing strict accountability and file-level coordination.

What it does

• Enforce governance by applying a hard authority ladder across project instructions, methodology rules, and user directions.
• Coordinate parallel agents using a lock protocol to prevent multiple agents from overwriting the same backlog items or files.
• Manage backlog ROI by calculating task priority against effort to ensure agents work on high-impact items first.
• Guard definitions of done by strictly linking task status to verified test passes and surgical code changes.
• Mitigate prompt injection by treating all external content, logs, and issue text as untrusted data rather than instructions.

How it works

1. Initialize project structure by creating the four planning layers from strategy docs down to granular epic backlogs.
2. Execute tasks by acquiring a file-lock, creating a PR-based implementation plan, and matching surgical code changes to existing styles.
3. Apply safety checks by validating all external inputs against the untrusted content rule before processing.
4. Verify and close by running pre-commit hooks and updating the Test: pass field before marking a task as done.

Frameworks & tools

This skill is platform-agnostic, designed for git-based repositories using Markdown for documentation. It integrates with any AI agent capable of reading SKILL.md or CLAUDE.md files.

Why this beats prompting it yourself

Manual prompts often fail to maintain state across long autonomous loops, leading to "done" tasks that haven't actually passed tests. This skill provides a rigid, encoded operating contract that prevents AI agreement bias and ensures every change is documented, locked, and verified.

Use cases

• Prevent two agents from working on the same feature simultaneously in a shared repo.
• Automate the sizing and prioritization of a messy backlog based on impact vs effort.
• Set up a secure autonomous coding loop that refuses to bypass pre-commit hooks or force-push to main.
• Audit a completed task against a 10-point definition of done checklist to ensure high assurance levels.

Known limitations

Does not support direct production deployments or overriding active file locks. Requires a specific repository folder structure to function at full capacity.