Contractor First Execution — Development skill for Claude Code

Contractor-First Execution

An agent skill for engineering work where wrong assumptions cost more than a short investigation.

It makes the agent inspect the repository before asking discoverable questions, state falsifiable assumptions, identify real blockers, and scale its process to the blast radius. Small fixes stay small. Work involving schemas, permissions, personal data, money, migrations, deployments, publication, or other shared boundaries gets an explicit execution contract.

The operating loop

1. Investigate: read the relevant implementation, tests, configuration, dependencies, and adjacent conventions.
2. Contract: restate the observable outcome, separate facts from assumptions, ask at most three genuinely blocking questions, and name the files and proof.
3. Execute: continue immediately when the work is already authorized and no blocker remains.
4. Stop on contradiction: if new evidence invalidates the plan, surface it instead of silently changing direction.

The skill has two lanes:

• Fast lane for obvious, reversible, low-risk corrections.
• Contract lane for material ambiguity or high-impact boundaries.

Install

Clone the repository into your personal agent skills directory:

git clone https://github.com/OthmanAdi/contractor-first-execution "$env:USERPROFILE\.agents\skills\contractor-first-execution"

Restart or refresh your agent host so it discovers the new `SKILL.md`.

Use

The skill should trigger naturally for requests such as:

• “Inspect the repository before coding and tell me your assumptions.”
• “Plan this migration, including rollback and tests.”
• “Do not ask me anything you can discover in the code.”
• “Implement this once you have checked the API and compatibility boundary.”

The full instructions live in [SKILL.md](SKILL.md). The public fixt