| name | orlix-write-to-harness |
| description | Use before editing Orlix harness files such as AGENTS.md, docs/plans, docs/harness, ADR indexes, .agents/skills, .codex/agents, Codex rules, hooks, or harness-related repo guidance. Ensures changes preserve the repo as the source of truth, avoid duplicate architecture claims, and keep references current. |
Orlix Write To Harness
Use this skill before writing Codex harness files.
Rules
- Keep
AGENTS.md concise and route to canonical docs.
- Keep architecture truth in
docs/architecture, docs/adr, and docs/reference.
- Keep task state in
docs/plans/active/<task>/.
- Keep durable agent lessons in
docs/harness/MEMORY.md.
- Avoid duplicate or competing architecture claims.
- Update stale references after moving docs or renaming skills/agents.
- Keep the
OrlixOS-is-the-Kit decision consistent across ADRs, architecture docs, glossary, skills, agents, and AGENTS.md.
- Remove stale
OrlixKit routing unless it is explicitly documented as retired/forbidden.
- Prefer Codex-native surfaces:
AGENTS.md, .codex/agents, .agents/skills, .codex/rules, and .codex/hooks.
- Treat
rtk as an output wrapper in rules and hooks. Guarded bare commands and rtk-wrapped equivalents must have the same policy.
Verification
After harness edits, run a stale-reference scan, inspect skill/agent trigger descriptions for clear narrow scope, and verify paired bare/rtk command policy where rules are touched.
