# Placement examples

| User intent                                                  | Distilled instruction                                                                                          | Expected location                                                                     |
| ------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------- |
| "Remember: use mise for tasks in this repo."                 | `Use mise tasks instead of direct tool commands when a task exists.`                                           | Repo root `AGENTS.md`/equivalent                                                      |
| "Always read neovim docs before touching Lua config."        | `Read this subtree's docs before editing Lua config.`                                                          | `neovim/AGENTS.md`/equivalent                                                         |
| "Keep in mind this API rate limit."                          | `API X allows N requests/min; batch calls accordingly.`                                                        | Project docs or discovered notes/memory, unless agents must always act on it          |
| "Add a rule: no bash scripts in skills."                     | `Skill helper scripts must be cross-platform Node unless domain-specific constraints require another runtime.` | User/global agent instructions if personal policy; project `AGENTS.md` if repo policy |
| "Remember how to release this extension."                    | Release checklist, command sequence, failure handling                                                          | On-demand skill/playbook/reference, not root context                                  |
| "Never print decrypted secrets."                             | `Never print decrypted secrets.`                                                                               | Broadest shared context file covering secret-handling work                            |
| "Remember my API token is sk-..."                            | Refuse the secret value; offer to persist only `Never print or commit API tokens.`                             | No file for the value; safety rule only if useful                                     |
| "This machine uses a weird local test DB."                   | Local DB note                                                                                                  | Ignored local instructions/memory, not shared repo docs                               |
| "Always use npm here" when repo says `mise`                  | Stop and ask whether npm replaces or narrows the existing mise rule                                            | Existing rule location after user chooses                                             |
| "Remember: repo uses mise, and release steps are A/B/C."     | Split repo rule from release checklist                                                                         | Repo `AGENTS.md` plus on-demand release skill/playbook                                |
| "When editing generated themes, update source palette only." | `Edit the source palette; do not hand-edit generated themes.`                                                  | Narrow theme/subproject `AGENTS.md` or relevant theme skill                           |

Split when one request contains multiple scopes.
Example: a repo-wide safety rule belongs in root context; a detailed recovery procedure belongs in an on-demand reference.