Agent Instructions

This is an OMH agent-instructions workflow skill, projected for Agent Skills hosts (Claude Code, Codex, Cursor, opencode, OpenClaw, pi).

Why This Exists

agent-instructions exists because the file every handoff target reads first had no owner: rules-distill extracts rule candidates, codebase-onboarding builds a human reading path, and context aligns terminology, and each reads that file without writing it.

First Steps

• List every instruction file and which agent reads it before proposing a section.
• Run or ask for the observed run of each command before marking it verified.

Do Not Use When

• The ask is distilling repeated lessons into reviewed rule candidates; use omh-rules-distill.
• The ask is a reading path through the codebase for a new engineer; use omh-codebase-onboarding.
• The ask is project terminology alignment -- reviewing the terms this project uses and correcting inconsistent vocabulary; use ulw-context.
• The ask is user-facing product documentation; use omh-docs.

Examples

Good example:

• Prompt: update our CLAUDE.md
• Expected behavior: Inventory the instruction files, run or collect each build and test command, mark each verified or unverified, and replace only the marked region with the commands, the generated-file map, the gates, and the costed pitfalls, refusing any count or line number.
• Why: An instruction file that states an unrun command or a stale count sends every later agent the wrong way.

Bad example:

• Prompt: add that the suite has 4,100 tests and the router is at line 212 of chat.py
• Expected behavior: Refuse both values, and write the command that counts the tests and the symbol that locates the router instead.
• Why: Both numbers were already wrong by the next merge.

Completion Checklist