Constraint-Driven Development interviews the user in four questions, each with a default, and records the answers in CONSTRAINTS.md at the repository root. The file holds a floor that is always enforced, numbered rules with the command that checks each one, measured metrics held by ratchets and a table of exceptions with owners and expiry dates.
The skill also installs a de facto tool for each chosen dimension, places each check by cost in the build, test, review and ship stages, and watches the diff for five moves that lower the bar.
What You Get
- A CONSTRAINTS.md with floor, enforced rules, measured metrics and exceptions
- A tool and command table for the common quality dimensions
- A diff-scoped floor-guard reference script and a quality-gate script for Semgrep and axe results
- A verification checklist that each enforced command fails on a known violation
Common Use Cases
- Setting up quality gates for a new project
- Choosing thresholds when no number exists
- Stopping an agent from silencing checks to reach green
- Guarding an autonomous build loop
What Makes It Unique
Most quality skills describe what good looks like in prose. This one records the project’s own bar with numbers and a command per rule, ranks checks by whether the agent can pass them with code that does not work, and requires at least one external check such as axe-core or osv-scanner.
Under The Hood
The SKILL.md loads when quality standards are discussed. It tells the agent to read package manifests, linter configs, CI files and coverage output, then to write CONSTRAINTS.md and update AGENTS.md or CLAUDE.md. It names install commands for npm, pip, pipx and brew tools. Bundled files are quality-gate.mjs, a Node script that reads Semgrep or axe JSON and sets an exit code, and a floor-guard reference that runs git commands over the diff.