[codex] Redirect terminal usage loops - #35698
Conversation
2a57ea3 to
13211a5
Compare
13211a5 to
8a4e9a3
Compare
8a4e9a3 to
6d3e3b8
Compare
Rebased onto upstream/main as a single commit to refresh PR NousResearch#35698.
6d3e3b8 to
918401f
Compare
|
Rebased and refreshed this upstream mirror at head Current state:
I attempted the actual merge through |
|
Closing as a redundant intermediate of a cumulative stack. These PRs form a stack where each is a strict superset of the previous: #35632 ⊂ #35673 ⊂ #35694 ⊂ #35698 ⊂ #35702. Both ends have now been refreshed onto current |
Summary
Root Cause
Terminal usage failures such as
meshctl task view,meshctl task status --bad-flag, and similar variants all looked distinct to the existing exact-argument guardrails. The agent received only a broad same-tool failure warning, so it could keep guessing new argument forms inside the same CLI surface.Validation
./scripts/run_tests.sh tests/tools/test_read_loop_detection.py tests/agent/test_tool_guardrails.py tests/run_agent/test_tool_call_guardrail_runtime.pypython -m ruff check agent/tool_guardrails.py run_agent.py tests/agent/test_tool_guardrails.py tests/run_agent/test_tool_call_guardrail_runtime.py tests/tools/test_read_loop_detection.pygit -c core.fsmonitor=false diff --check