FIELD NOTE / X
A hook sees a route, not a consequence.
The short film, the complete written thought, and the evidence behind it.
The X conversation link will follow its public release.
A hook sees a route, not a consequence.
Video caption
A hook sees a route, not a consequence. The customer feels the mutation, not the callback count. Probe shell, MCP, delegation and service routes separately. Narration uses Eric's authorized AI voice clone. #EricFieldNotes
Full written post / accessibility read
If a coding agent has a pre-call hook, it is tempting to call the production tenant protected. But that hook sees matching host calls. A shell command, an MCP tool, a child agent or a service credential may reach the same effect by another route.
In a disposable tenant fixture, suppose the direct tool is denied while a delegated worker holds a service token. The dashboard can show a perfect denial record and the target can still change. That is a route inventory failure, not proof that a particular vendor hook is broken.
For each path record the host invocation, callback matcher, tool, principal, service endpoint and target receipt. Mark observed, denied, or unknown. Send a harmless canary through each route and independently read the target. Unknown stays unknown until a real probe runs.
Write the invariant as staging identity cannot mutate production. Enforce it in the target service, then use hooks to catch mistakes early. Re-run the route canary when a connector or credential changes, because a new authority path can bypass an old callback without touching its code.
Narration uses Eric's authorized AI voice clone.
#EricFieldNotes
Four-beat scene transcript
1. A hook sees a route, not a consequence.
If a coding agent has a pre-call hook, it is tempting to call the production tenant protected. But that hook sees matching host calls. A shell command, an MCP tool, a child agent or a service credential may reach the same effect by another route.
Visual: A green callback log can miss a second authority path.
2. The missing lane still has credentials.
In a disposable tenant fixture, suppose the direct tool is denied while a delegated worker holds a service token. The dashboard can show a perfect denial record and the target can still change. That is a route inventory failure, not proof that a particular vendor hook is broken.
Visual: The customer feels the mutation, not the callback count.
3. Build an authority table.
For each path record the host invocation, callback matcher, tool, principal, service endpoint and target receipt. Mark observed, denied, or unknown. Send a harmless canary through each route and independently read the target. Unknown stays unknown until a real probe runs.
Visual: List every path from instruction to protected effect.
4. Inventory effects before claiming control.
Write the invariant as staging identity cannot mutate production. Enforce it in the target service, then use hooks to catch mistakes early. Re-run the route canary when a connector or credential changes, because a new authority path can bypass an old callback without touching its code.
Visual: Probe shell, MCP, delegation and service routes separately.
Research and claim limits
The examples identified as illustrative or simulated are design probes, not reported incidents. Vendor specifications do not establish workload performance.