# Read-only discovery authorization

- Goal:
- Authorized mode: Learn
- Starts:
- Expires:
- Human owner:
- Sources/targets:
- Allowed observations:
- Explicitly forbidden:
- Rate/time limits:
- Sensitive-data handling:
- Required output:
- Stop conditions:

## Default allowed observations

System identity/version, hardware inventory, interfaces/routes, listeners, package versions, service state, selected logs, filesystem types/mounts, numeric ownership, SMART data, configured backup job status, and configuration reads that do not expose secret values.

Authorization to observe locally is not consent to send every value to a model provider. Keep the full evidence in the human-controlled raw store. The agent-facing output replaces WAN addresses, MAC addresses, serials/WWNs, household/account identifiers, and query-level activity with opaque IDs or aggregates unless `Sensitive-data handling` explicitly names an approved field and provider. Stop if an allowed command unexpectedly emits credentials, personal content, browsing history, or an unapproved raw identity map.

## Default forbidden actions

Writes, restarts, reloads, package changes, filesystem repair, SMART destructive tests, secret-value extraction, public scanning, authentication attempts against unknown accounts, and any attempt to bypass a permission denial.

Unavailable approval means deny. A failed read does not authorize privilege escalation.