agenticreq audit · no spec · no mock run yet

Result: FAIL    5 gaps · 0 sign-offs left · 4 to check · 2 passed
audit without a spec (framework: openai-agents): built-in defaults, not rules you wrote.
To check YOUR rules: `agenticreq init`, fill spec.yml, then `agenticreq check`.
checked: bounds, fail_closed, tool_inventory_report, unbounded_tool_policy, structured_output,
    write_safety, output_size_limits, trust_boundary, egress_policy, secrets_policy,
    self_modification, sandbox_policy, trace_policy, audit_log, version_pinning,
    environment_separation
not read from code: sign-off items (kill switch, log immutability and the rest a person
    confirms)
needs a spec (Level 1): prohibited, limits, approval_required, escalation, untrusted_inputs,
    declared_delegation, role_boundary

GAPS FOUND IN CODE (5)
  ✗ write_safety [reporter]  Write tool save_report runs without approval.
  ✗ bounds.max_cost_usd      No max_cost_usd set: write a value, or not_needed with a reason.
  ✗ bounds.max_duration_s    No max_duration_s set: write a value, or not_needed with a reason.
  ✗ structured_output        Agent output is free text, not a validated structure: no typed /
                             schema-validated output found (response_format, output_type,
                             with_structured_output, output_pydantic and the like); not required at
                             the medium default, required at high / critical.
  ✗ environment_separation   Configuration is not separated by environment: no configuration is read
                             from the environment (os.environ / getenv / dotenv / a Settings class);
                             endpoints, keys and models are fixed in code.

CHECK WITH A MOCK RUN (4)
  ? fail_closed       Code pattern only (no swallowed exception in 2 tool bodies). Add a mock that
                      returns an error, then run: agenticreq mock-run
  ? audit_log         Code pattern only: audit hook (OpenAI Agents SDK traces every run by default
                      and nothing disables it (set_tracing_disabled /
                      RunConfig(tracing_disabled=True))). Add a mock that exercises it, then run:
                      agenticreq mock-run
  ? business_rules    No spec given, so the Level 1 rules did not run. Run: agenticreq init
  ? runtime_evidence  The agent has not run yet, so nothing was seen at run time. Run: agenticreq
                      mock-run

PASSED (2)
  ✓ bounds.max_steps  Found in code: 5 (agent.py:26). Also: Found in code: 10 (agent.py:27).
  ✓ version_pinning   Found in code: gpt-4o-2024-08-06 (dated suffix -YYYY-MM-DD, agent.py:19).
