Verifiedon 2026.7.1
Action boundary
Before you act
- Expected result
- A least-privilege matrix distinguishes channel, gateway, operator, and service identities.
- Failure mode
- A shared administrator credential makes an agent action indistinguishable from a human action.
- Rollback
- Remove newly granted access and return to the previous documented authority matrix.
Identity is not intent
Authentication answers who connected; authorization answers what that identity may do; approval answers whether this specific action should proceed now. Keep those questions separate. Consult the current authentication and trusted-proxy documentation before changing gateway controls: reverse-proxy headers are not evidence of identity unless the proxy boundary is explicitly trusted.
Lab — the authority matrix
Make a table with rows for inbound channel, gateway process, administrator, operator, tool/service account, plugin, and reviewer. Columns: can read, can write, can send externally, can change configuration, can approve, and how the action is logged.
For every write, external send, privilege increase, plugin install, or security-control change, choose an approval rule: no approval only for reversible, bounded actions; explicit approval for irreversible, external, or privilege-changing actions. Approval must show the target, data summary, and impact—not merely a button labelled “continue.”
Learner artifact
Capture a redacted approval request showing actor, requested operation, target, scope, expiry, and approver decision. Do not capture message content or tokens.
Checkpoint
A tool has a broad service credential but the chat user is read-only. Is a read-only chat permission enough? No. The effective authority is the reachable tool credential. Narrow the credential, restrict the tool path, or introduce a human gate.
Test
Attempt a safe, non-production action with an unapproved identity. Record the denial and the audit event. If it succeeds, stop automation on that path and open a remediation finding; do not “test” by performing a destructive action.
Source provenanceVerification and sources
Review receipt rr_security_identities_approvals
- Outcome
- approved
- Method
- source-review
- Reviewer
- academy-security-source-review
- Reviewed
Evidence
- openclaw-docs-index — docs-authentication-2026-07-30; snapshot
35120aeb5dd4…
Limitations
- Approval covers the bounded 2026.7.1 learning exercises and cited security guidance; operators must validate controls against their own deployment and use disposable credentials in labs.
Lesson checkpoint
Ready to move on?
Mark this lesson complete when you can apply its outcome without relying on the examples above.