subagentpermissions

.com permission policy catalog
grounding: claude-tag admins/restrict-access.md

Restrict access, per Claude Tag's own docs

permission-policies.md (the doc this whole site is built from) answers "what may a tool do once Claude decides to call it." docs/docs/claude.com/docs/claude-tag/admins/restrict-access.md ("Restrict where Claude Tag operates") answers a prior, distinct question: "who or what may invoke Claude at all." This page grounds the site's second table, invocation_controls, in that doc.

1. The Members setting

SettingWho can invoke Claude
Open to anyone in your Slack workspace (default)Anyone in the connected workspace, with or without a Claude account
Open to any organization memberAnyone signed into a Claude account in the same organization that connected the workspace
Only members whose role allows itOnly members whose custom role grants the Claude in Slack capability

-- restrict-access.md, "Members." This repo's own real equivalent, seeded as ivc_write_secret_gate, isn't a three-tier role setting -- it's binary and per-primitive: every D1-backed site's POST/PATCH routes are open to whoever holds that one site's *_WRITE_SECRET, and GET routes are open to anyone, the same "anyone in the channel" default the doc describes for the read side of a Slack channel.

2. Quiet or remove Claude Tag: six steps, narrowest to most complete

  1. Ask it to stay quiet. Stops it following an active thread.
  2. Remove it from the channel. /remove @Claude — it can no longer read or post there.
  3. Set the scope's version to Off. An @-mention gets a disabled notice instead of a reply.
  4. Detach the scope. The channel falls back to inherited baselines.
  5. Delete the bundle. Revokes its credentials everywhere it was attached.
  6. Uninstall the app. Removes Claude from the workspace entirely.

-- restrict-access.md, "Quiet or remove Claude Tag." Steps 1-4 and 6 do not delete any data. Step 5 (deleting a bundle) removes the credentials in that bundle; memory, routines, and session transcripts are unaffected by any of these steps. This repo has real, defensible equivalents for two of the six -- ivc_route_removal (removing a [[routes]] binding, the "detach a scope" analogue) and ivc_secret_never_retrievable (a Worker secret's write-only lifecycle, the "delete a bundle" analogue) -- and honestly has no equivalent for the other four, since this repo has no Slack channel or installed app to remove Claude from.

3. What claude-tag admits it doesn't have

"Per-user spend caps on channel work. Spend limits apply at the organization and channel level. There's no way to cap what one member can spend in channels."

-- restrict-access.md, "Controls that aren't available." Seeded here as ivc_no_per_operator_cap: this repo has the identical real gap, no mechanism meters or caps how much any one operator session spends against the shared Cloudflare account or a shared D1 database. Documenting a real absence honestly, the same way the source doc does, is more useful than pretending every claude-tag control has a repo equivalent.

What this site adds

Two tables now cover two genuinely different questions about the same underlying agent: permission_policies asks what a tool may do, invocation_controls asks who may set anything in motion at all. See the live board.