Custom instructions, per Claude Tag's own docs
Agent identity and standing instructions are explicitly two different primitives in Claude Tag's own model -- agent-identity.md says so directly: "How Claude behaves in channels (its standing instructions, plugins, and channel memory) is configured separately from its identity." This page grounds the directives table in the doc that actually describes that second primitive, docs/docs/claude.com/docs/claude-tag/admins/attach-to-scope.md ("Configure per-channel access").
1. Three instruction layers, written by different people
| Layer | Who writes it | Where |
|---|---|---|
| Custom instructions | Owner | The scope, in admin settings |
| Channel memory | Anyone in the channel | By telling Claude to remember |
| Task prompt | The requester | The message itself |
-- attach-to-scope.md, "Instruction layers." This site's directives.layer column is exactly that first row, made queryable: every seeded directive is custom_instructions, because CLAUDE.md is an Owner-authored, repo-scoped standing instruction, not something any individual session wrote for itself the way channel_memory or task_prompt would be.
2. Scopes inherit downward; narrower scopes add, they don't replace
"Bundles stack downward. A channel gets whatever is attached at Default Slack access, plus its workspace, plus anything attached to the channel itself." ... "Per-scope custom instructions are concatenated, Default Slack access first, then workspace, then channel. A channel's instructions add to, rather than replace, what's set above it."
-- attach-to-scope.md, "How scopes inherit" and "Custom instructions." This site's directives.scope column generalizes that exact inheritance chain to this repo: repo_wide directives are the Default-Slack-access analogue -- every session touching this repo inherits them, regardless of surface -- and surface_scoped directives are the channel analogue, narrower and additive, never overriding a repo-wide rule. dir_naming_ontology_enforcement and dir_rustc_wrapper_workaround are both surface_scoped for exactly this reason: they add a constraint on top of the repo-wide dir_cloud_sandbox_rule, they don't replace it.
3. What this site adds
Claude Tag's instruction-layer model lives inside one Slack-scoped admin console. This site factors the same "Custom instructions" layer out into a small typed primitive -- directives in D1 -- populated with this repo's own real, already-enforced standing instructions from CLAUDE.md, so "how Claude behaves is configured separately from who it is" stops being a documentation sentence about Slack scopes and becomes a live, queryable catalog of every standing rule this actual codebase's own sessions already inherit. See the live board.