Untrusted content
Instruction-like text in the material an agent reads.
The ClawMetry detector library
Every finding has a story.
See what can go unnoticed in an agent workflow, what ClawMetry detects, and what to check next.
18 detectorsDetailed video explainersInspectable evidence
For people running agents such as Claude Code, Codex and OpenClaw. Coverage depends on the telemetry and configuration available to ClawMetry.
Sensitive actions
File changes, credentials, destinations and privilege requests.
File blast radius
One small task. Thirty files changed.
See when an agent task reaches more files than expected.
Explore the detector →Credential access
Your agent opened .env.
Catch a sensitive file read inside an ordinary task.
Explore the detector →Network egress
A new host. A reason to look.
Notice unfamiliar destinations in captured agent commands.
Explore the detector →Privilege change
The task just asked for more power.
Catch recorded commands that request elevated access.
Explore the detector →The agent environment
Executable configuration and changes to the components an agent loads.
Repository config execution
The checkout names a program.
Find executable configuration that a normal tool call can hide.
Explore the detector →Agent config tamper
Who changed the agent rules?
Notice hooks and changes to the files that guide an agent.
Explore the detector →Package install scripts
Install the package. Run its script.
See executable lifecycle hooks before overlooking them in a manifest.
Explore the detector →Agent component change
A new tool joined the agent.
Track added or changed MCP servers, skills and plugins.
Explore the detector →Work that goes off track
Loops, repeated errors and activity that deserves a closer look.
Stuck loop
Still running. Going nowhere.
Catch the repeated action hiding inside a busy session.
Explore the detector →No progress
Busy does not mean progress.
Spot long stretches of activity without a recorded edit or completion.
Explore the detector →Repeated tool failure
The same tool. Another failure.
Find the failing step that retries keep burying.
Explore the detector →Action discrepancy
The step failed. The agent moved on.
Notice when a failure is followed by a different action.
Explore the detector →Work that stops quietly
Capacity refusals, unanswered requests and repeated restarts.
Rate limited
Waiting on the provider.
Separate capacity refusals from productive agent work.
Explore the detector →Blocked on user
Your agent is waiting for you.
Find the approval or unanswered question holding up a session.
Explore the detector →Repeated restarts
Back online. Then back again.
Spot a session that keeps starting over.
Explore the detector →Inspection incomplete
A scan limit should be visible.
Know when credential inspection could not cover the full payload.
Explore the detector →Patterns across agents
Shared unusual actions across independent sessions.