nour-command

Use for any non-trivial Nour task that needs context recovery, end-to-end execution, MAX EFFORT, backlog reconciliation, deep research, audit-and-fix work, or proof-backed completion. Also use when No

SKILL.md

GitHub

NOUR COMMAND

This skill is a router, not a second policy.

  1. Read ../../../AGENTS.md; it is the engineering/safety authority.
  2. Read ../../../NOUR-COMMAND.md; apply its execution framework.
  3. If the task is non-trivial, read only the relevant entries in ../../../docs/agent-os/NOUR-COMMAND-MODES.md.
  4. Load the nearest app AGENTS.md before editing app code.
  5. Invoke any narrower repo skill whose trigger applies; this router does not replace domain skills.
  6. Define the task-specific completion contract before consequential execution.
  7. Recover context before asking Nour to repeat prior work.
  8. Verify current remote/production truth before trusting a prior session, plan, or local checkout.
  9. Prefer finish/wire/repair/consolidate/adopt/delete before adding another implementation.
  10. Execute in dependency order and run authorized work to empty.
  11. Try to falsify apparent success before declaring completion.
  12. Simplify after a successful change and turn material failures into regression protection.
  13. Report demonstrated outcomes, exact proof, real remaining work, and exact blockers.

When AGENTS.md or an app-specific instruction conflicts with this skill, the canonical policy wins.

For Claude, NOUR-COMMAND.md is imported by root CLAUDE.md, so the main framework is already in context. This skill exists to make the mode-router invocation explicit and to force loading only the relevant deep-mode details.