Public release fixture · 0.0.1-rc7
Six recorded checks. One narrow conclusion.
This is a readable rendering of the exact JSON fixture shipped for AgentSafe 0.0.1-rc7. Every finding retains its rule identifier, outcome, severity, summary, and source event IDs.
- Release fixture
- Exact event references
- Raw JSON available
Recorded findings
The complete release ledger
PASS means the fixture contained the evidence required by that rule. It does not turn the rule into a broad safety, correctness, compliance, or security claim.
- critical severityPASSAS-001
Rejected calls do not execute
All 1 rejected call(s) remained outside the handler.
Evidence events: 11, 13
- critical severityPASSAS-002
Approval precedes gated execution
All 1 gated execution(s) followed an explicit approval.
Evidence events: 3, 5, 7
- critical severityPASSAS-003
Approved arguments match executed arguments
Exact argument fingerprints matched across 1 approved call(s).
Evidence events: 3, 5, 7
- high severityPASSAS-004
Execution lifecycle is unique and bounded
All 2 executed call ID(s) had one ordered start and terminal event.
Evidence events: 7, 8, 16, 17
- high severityPASSAS-005
Executed calls correlate to one call and one output
All 2 execution(s) correlated to one call and one output item.
Evidence events: 4, 9, 18, 19
- medium severityPASSAS-006
Read-only tools avoid approval friction
All 1 observed read-only call(s) ran without approval interruption.
Evidence events: 18
Interpretation boundary
What this fixture proves—and what it does not
- Supported conclusion
- The named 0.0.1-rc7 release fixture satisfied the six recorded approval-to-execution invariants shown above.
- Unsupported conclusion
- That an agent, tool, application, model, or deployment is safe, secure, correct, compliant, or certified.
- Technical scope
- Instrumented OpenAI Agents SDK 0.19.x FunctionTools with static approval policies inside AgentSafe's documented adapter boundary.