Bitter Frontier What coding agents just made possible, where the bottleneck moved, and what deserves attention.

Evidence record / omnigent

A dated record of one change, kept so the writing that cites it can be checked. Compiled from the sources listed below by the research run, not written for reading. The judgment lives in the issues that cite it, below.

2026-10-13-omnigent-correction-on-a-policy-server-outage-claude-code-and-codex-tool-calls-get-an-explicit-ask

Correction: on a policy-server outage, Claude Code and Codex tool calls get an explicit ask, not no opinion, and response checks now give up after two minutes. The September read of this, taken from the PR body, was wrong in the part that matters. Since v0.13.0, when a claude-native or codex-native session cannot reach the Omnigent policy server (after up to 30 seconds of retries), the PreToolUse hook does not abstain. It returns an explicit ask, and Omnigent’s comments and tests say that choice was made because abstaining “fails open in bypassPermissions/acceptEdits”. Prompt submission still blocks, PostToolUse fails open, and devin-native and kimi-native sessions deny. One v0.18.0 change completes the picture: policy evaluations that cannot park on a human, the response and tool-result phases, now give up after at most 120 seconds and fail open “exactly as before”, instead of holding a turn open for hours.

Channel: tagged-release. Half: both. Date: 2026-10-13 (re-read at v0.18.0; outage behavior since v0.13.0; #8665 first ships in v0.18.0).

Operator consequence: The layering answer for an outage: Omnigent pushes the decision into the harness’s own prompt rather than letting a bypass-mode harness run silently, provided the harness honors a hook ask in that mode. Confirm that on the Claude Code and Codex versions you run before relying on it, and keep a harness-native deny list for anything that must never run. Response and tool-result policies (output checks, redaction after the fact) are advisory under failure and now give up within two minutes.

Receipt

Finding metadata

Confidence
high

Run: 2026-10-13-profile-refresh-2026-09-21_2026-10-13-frontier-v0

Finding ID: 2026-10-13-omnigent-correction-on-a-policy-server-outage-claude-code-and-codex-tool-calls-get-an-explicit-ask

Profile citations

Source links

Primary links, including exact changelog lines when available.

  1. Omnigent
    tagged commit file2026-10-13-omnigent-correction-on-a-policy-server-outage-claude-code-and-codex-tool-calls-get-an-explicit-askomnigent-ai/omnigent / omnigent/native/native_policy_hook.py#L543-L584tagged commit file2026-10-13-omnigent-correction-on-a-policy-server-outage-claude-code-and-codex-tool-calls-get-an-explicit-askomnigent-ai/omnigent / tests/test_native_policy_hook.py#L378-L390tagged commit file2026-10-13-omnigent-correction-on-a-policy-server-outage-claude-code-and-codex-tool-calls-get-an-explicit-askomnigent-ai/omnigent / omnigent/native/native_policy_hook.py#L540-L580merged pr2026-10-13-omnigent-correction-on-a-policy-server-outage-claude-code-and-codex-tool-calls-get-an-explicit-askgithub.com/omnigent-ai/omnigent/pull/6429merged pr2026-10-13-omnigent-correction-on-a-policy-server-outage-claude-code-and-codex-tool-calls-get-an-explicit-askgithub.com/omnigent-ai/omnigent/pull/8665tagged commit file2026-10-13-omnigent-correction-on-a-policy-server-outage-claude-code-and-codex-tool-calls-get-an-explicit-askomnigent-ai/omnigent / omnigent/policies/types.py#L66-L80

Versioned source: run artifact