Allow
It runs
The call passed every check. The decision and its reason are recorded.
Layer 06
A policy matters only if it is checked when the agent acts. StartiQOS checks every tool call before your agent can run it, in a fixed order, and records the result.
The Checks
Each check depends on the ones before it, so the order never changes.
01
If it is on, nothing runs.
02
Shell and code execution are refused everywhere.
03
The agent’s token must be valid and unexpired.
04
The chain back to the original request must be intact.
05
The agent may hold no more than its parent.
06
The capability must be in the approved policy. If it is not, the call is paused for a person.
07
The combined reach must stay within the limits.
08
The session’s taint must be within the capability’s bound.
09
Arguments must match the expected shape. If they do not, the call is paused for a person.
The Verdict
Every call ends in one of three answers, and every answer is recorded.
Allow
The call passed every check. The decision and its reason are recorded.
Escalate
The agent is paused. The approver can allow it once, add it to the policy, or deny it.
Deny
The call failed a check that a person should not override in the moment, such as a broken chain or a blocked combination.
Questions
Under 0.05 ms at the 99th percentile in our tests (0.046 ms). The model call your agent makes takes far longer.
It never allows silently. In fail-closed mode it denies; in fail-open mode it pauses the call for a person.
Yes. It is the first check on every call, so switching it on stops every agent at once.
Changing any entry breaks the hash chain, which verification reports. That makes the ledger tamper-evident, not tamper-proof.
Early access starts with a two-week shadow-mode pilot. Nothing is blocked. You get the gap between the access each agent holds and the access it uses, and every combination that could move data out.