We Started With Four Open Agent CLI Questions. We Ended With Two Real Wins. The agentpolicy project released v0.1.0, a local-first, Cedar-based policy decision point for AI agents that returns allow, deny, ask, rate-limit or redact decisions for every agent action, now available on PyPI. In a Tier-1 field test, the team got live harnesses working for Claude Code and Codex — capturing request/response traffic and audit trails, blocking destructive commands and escalating production writes — while documenting honest noes for Cursor and Gemini, whose CLI could not run locally due to a classifier-only gemma route and an OAuth tier gate. From the agentpolicy https://github.com/agentic-suite/agentsec-agentpolicy v0.1.0 release: a local-first, Cedar-based policy decision point for AI agents decide allow, deny, ask, rate-limit, or redact for every agent action, with shipped rule packs . Now on PyPI https://pypi.org/project/agentsec-agentpolicy/0.1.0/ . This story is the Tier-1 part of the v0.1.0 field test report https://github.com/agentic-suite/agentsec-agentpolicy/blob/main/docs/field-test/v0.1.0/FIELD TEST REPORT.md . AI disclosure: this article was drafted and edited with AI assistance. The field test, the defects, the fixes, and every number are real and linked below. At one point, the entire Tier-1 CLI section of the field test was still unresolved. That was the honest state. Claude Code had not run live yet. Codex had not run live yet. Cursor and Gemini had not been closed out either. If I had stopped there, the report would have been accurate but incomplete. So I kept going case by case, with one rule: do not claim a harness until it runs with evidence on this box. That persistence changed the release in a meaningful way. Claude Code and Codex both went live. Cursor and Gemini both got real investigations and honest noes. We shipped the two we could prove and left the other two out. The Tier-1 set in the plan was four CLIs: Claude Code, Cursor, Codex, and Gemini. Early in the sweep, none of them belonged in a green release note. The report had to say that plainly. This is where the field-test discipline mattered. A case was only real if it ran through the actual harness and left evidence behind: verdict, assertions, transcripts, audit trail, and captured LLM traffic. If it did not, it stayed unproven. That kept me from turning "probably works" into "supported." Claude Code was the first real win, and it took more than a config toggle. The working harness is in scripts/fieldtest/t1 claude.sh https://github.com/agentic-suite/agentsec-agentpolicy/blob/main/scripts/fieldtest/t1 claude.sh , with the hook entrypoint in t1 evaluate.py https://github.com/agentic-suite/agentsec-agentpolicy/blob/main/scripts/fieldtest/t1 evaluate.py and the shared policy in t1 policy.yaml https://github.com/agentic-suite/agentsec-agentpolicy/blob/main/scripts/fieldtest/t1 policy.yaml . Three things had to be figured out: /v1/messages with a real msg body, so the CLI could talk to it directly. --settings explicitly because cwd settings files do not load headless the way I assumed. t1 proxy.py Once that wiring was real, the case passed live: a read executed, a prod write escalated without creating the side-effect file, a destructive command was denied, and the audit trail recorded all three decisions. Six request/response pairs were captured. Codex needed a different investigation. Its harness lives in scripts/fieldtest/t1 codex.sh https://github.com/agentic-suite/agentsec-agentpolicy/blob/main/scripts/fieldtest/t1 codex.sh . The key findings were different from Claude's: /v1/responses natively. codex exec reads stdin. Left alone in a harness, it quietly waits forever. Closing stdin with