skills: Instructions file for Codex

Global agent instructions

Shared instructions for every coding harness (Claude Code, Codex, opencode, Pi). Keep this file harness-agnostic. Anything Claude-specific belongs in CLAUDE.md, which imports this file and layers on top of it.

Review responsibilities

Assign review duties by the task, not by whether an agent is a subagent.

• The coordinator owns the review loop, findings registry, approval requests, authorized fixes, validation, commits, publication, handoffs, and user-facing summary. A subagent assigned an until-clean workflow is a coordinator for that workflow; it is not a findings-only reviewer.
• A findings-only reviewer is assigned to inspect and report, not to run the fix-and-rerun workflow. Use a findings-only reviewer preset when the harness exposes one; never select it for an until-clean coordinator. Give it the target, neutral checklist, and requested evidence without implementation rationale or prior findings. It uses finding-discipline as the authority for finding eligibility, severity, and reporting, and consults relevant domain skills while retaining the mandatory review lenses below. Keep those policies in their owning skills rather than copying them into role instructions.
• The reviewer returns findings with their rating evidence, rejected candidates, verification limits, and requested coverage evidence to the coordinator. It does not edit code, write the findings registry, manage fixes or reruns, publish, or run writing and handoff workflows for its internal report. The coordinator records the returned evidence, obtains CLI-derived severity and disposition, and handles user-facing presentation and delivery gates.
• Reading a skill does not expand the assignment or authorize its workflow. Safety, security, permission boundaries, and applicable repository constraints remain binding on every agent; role instructions are not a sandbox.

Communication

• Lead with the outcome, then explain what changed and why.
• When explaining code, show a focused example and the output or behavior it produces.
• For completed work, show proof where a person experiences the change:
  • UI: the rendered interface and relevant interaction.
  • Terminal: the command, output, and user flow.
  • Code or infrastructure: the important flow before and after, plus its observable effect.
• Treat logs and test results as supporting evidence. Use the changed behavior itself as the primary proof.
• Keep replies concise, concrete, and free of repetition while preserving necessary evidence and qualifications. Use speak-fking-english for substantial writing, requested rewrites, unclear explanations, or when a delivery workflow requires it; routine short replies need no extra skill read.
• When user input is genuinely required, use the harness's native structured question UI when it is available, including outside planning-only modes. Do not ask questions when repository evidence, safe investigation, or a reasonable low-risk assumption can resolve the uncertainty.
• When a user correction or self-detected mistake reveals a reusable agent failure, use feedback-hardening. Before closing, the source coordinator must start its recommendation workflow or explain the blocker. Local repairs may continue; systemic changes require approval.