Detect (DE) · Authorization
Authorization at the In-agent (forensic only) layer
advisory · Bypassable by language alone
If this concern is breached, how do we know?
What this cell does
Tool descriptions logged with each call (forensic).
Deliberately empty
This cell has no artifact, and that is the argument rather than a gap. At this layer, for this concern, nothing is enforced. Populating it with an enforcement claim would invert what the framework is saying.
Cell notes
Sentinels, Authorization / In-agent
Control. Tool descriptions logged with each call (forensic).
Strength. Forensic only.
Tooling
- - Whatever logs your agent runtime emits per tool call. For Claude Code, see
~/.claude/sessions/.
Files in this directory
(None. The capture is a property of the agent runtime; the shipping is handled by ../../identity/in-agent/ship-sessions.sh.)
Verification
Find a known tool call in the shipped session log and confirm its description was captured.
Common mistakes
- - Trusting tool descriptions in the session log to be authoritative for what the tool will actually do. Tool-description rug-pull (see
../../supply-chain/) means the description at call time may differ from the description at approval time.
Citation
NIST CSF 2.0 DE.CM-09.
Crosswalk
| NIST CSF 2 0 | DE.CM-09 |
|---|
Cite this cell:
https://agenticcovenants.com/detect/authorization/in-agent/