| # AGENTS.md — Global Codex Instructions | |
| These instructions apply to Codex in this workflow. | |
| Before any task, always read: | |
| - `WORKFLOW.md` | |
| - `PROJECT_STATE.md` | |
| - `TASK_REGISTRY.md` | |
| - project-specific `AGENTS.md` when working in `projects/<project>` | |
| ## Mandatory Rules | |
| - Apply Reuse First before every modification. | |
| - Write a short plan before coding. | |
| - Never create a large module unless clearly necessary. | |
| - Never modify files outside the requested scope. | |
| - Never store secrets, tokens, keys or credentials. | |
| - Never hardcode API keys. | |
| - Always write a task report. | |
| - Always update `PROJECT_STATE.md`. | |
| - Always explain which tests were run. | |
| - Always report tests that were not run and why. | |
| - Use skills in `/skills` when relevant. | |
| - Prefer small, reviewable changes. | |
| - Stop and ask when a change could destroy data or expand scope. | |
| ## Required Pre-Coding Output | |
| Before coding, always output: | |
| ```markdown | |
| ## Task Understanding | |
| Goal: | |
| Non-goals: | |
| Expected output: | |
| ## Existing Code Search | |
| Commands used: | |
| Files found: | |
| Reusable components: | |
| Missing components: | |
| ## Reuse Decision | |
| Decision: reuse / modify / build / stop | |
| Reason: | |
| ## Minimal Design | |
| Files to edit: | |
| Tests to run: | |
| Risks: | |
| ``` | |
| ## Quality Bar | |
| Codex must complete the full loop: understand, search, decide reuse, design minimally, implement only assigned files, test, check leakage/security, report, update state and prepare review. | |