Create, shorten, or reorganize AGENTS.md and its references when agent guidance changes. Enable only the requested documentation, ADR, or task workflows.
Read existing instructions and inspect only the areas relevant to the requested change.
Enable only the requested workflows. Preserve existing paths, policies, and languages; migrate files or change global configuration only when requested.
Keep AGENTS.md to actionable repository constraints and conditional links. Include rules that affect agent decisions, not generic advice or facts easily found in code.
Keep one source per rule. Remove duplication, contradictions, and obsolete references without losing intentional constraints.
Load detailed references only for the task that needs them. Do not route agents through several indexes or require reading every skill.
For a new layout, keep AGENTS.md at the root and create only needed references under .agents/docs/, .agents/adr/, or .agents/todo/.
Require correcting inaccurate guidance, not producing text after every edit. Add an ADR only when its rationale has lasting value, and a task only when work needs tracking.
Define agent-content and human-documentation languages separately when needed. An English skill does not impose English output on every project.
Keep task history and procedures out of AGENTS.md. Use specialized skills only for the artifacts needed; do not enable them all as a side effect.
Check changed paths and links. Do not add explanatory files or usage tutorials unless requested.