LLM²Working preview · All tools · Assistant access

Agent readiness / declared inventory

See the permissions.
Prepare the review.

Record owners, capabilities and named resources. Keep missing control, logging and revocation evidence visible, then prepare a queue for human review.

Declared inventory only · no account connection · no action execution

Loading fictional example…

Works in browser memory. No data is uploaded. References and confirmations are declarations; this prototype does not authenticate an agent, verify effective permissions or enforce approvals. Do not enter credentials or secrets.

Declared agents, capabilities and evidence

Inventory review

Preparing the local example

The form enables when its local engine and fixture load.

Loading local assets…

Separate the questions

Evidence is not authorization.

A complete set of declarations can still require priority human review. Send, purchase, transfer, delete or delegate capabilities with confidential or regulated resources remain explicit review items. No result grants permission for a particular action.

MCP readOnlyHint is descriptive metadata. Wildcard and unknown scope remain unresolved even for a read capability. Supplied verification flags do not independently prove effective controls.

Know the coverage

Start with a declared inventory.

Agents outside this inventory are not assessed. Local labels cannot prove human-versus-agent attribution. Opaque provider role names do not distinguish direct, inherited, app-only or delegated access.

Ask separately for the sponsor, administrator and provider export timestamp. Queue dates use only the next review date you supply. This prototype discovers nothing, schedules nothing and changes no permissions.