Codex's Instant Interrupts PR introduces a first-party, low-latency interruption primitive for the Codex coding agent, enabling safer human-in-the-loop control of long-running agent tasks.
state: seedheat: lowuncertainty: mediumconvergesscott: highagent-harnesses agent-orchestration coding-agents instant-interrupts human-in-the-loop codex-controlOpenAI Codex teamarmcat
What is this?
Codex is OpenAI's first-party coding agent (terminal and cloud), which the supplied changelog coverage describes as evolving from a repo-editing coding agent into a managed 'agent workspace' for long-running engineering work — adding persisted /goal workflows, automatic approval reviews before risky actions, explicit permission profiles, and in-app browser verification. The case claims repo PR #51812 by contributor 'armcat' adds 'Instant Interrupts', a low-latency harness-level interruption primitive for safer human-in-the-loop control; notably, none of the supplied snippets directly corroborate this specific PR or feature — they only establish Codex's broader direction toward long-running agents with human oversight (e.g., The New Stack's coverage of Codex continuing work while awaiting developer input). Whether the PR exists, what exactly it implements, and who armcat is are not established by the supplied material.
Why it matters to Scott
OpenAI Codex shipping 'Instant Interrupts' is a consequential other party arriving at the prompt-interrupt architecture Scott already defined and the Codex CLI gap he documented in Two Vendors One Primitive. The PR implements the low-latency harness-level interrupt primitive that his framework treats as the decisive signal — scheduled intent delivered into a warm, stateful process rather than a fresh run — directly validating his architectural analysis and the independent-convergence thesis.
ip:framework.prompt-interrupt-architectureip:source.two-vendors-one-primitive-ebookip:concept.interrupt-budgetip:concept.autonomy-budgetip:framework.long-running-agentsip:concept.human-over-the-loopradar:concept.agent-controlradar:concept.agent-harnessradar:concept.agent-harnessesradar:concept.human-in-the-loopradar:concept.agent-orchestrationradar:concept.long-running-orchestrationradar:concept.long-running-agentsradar:concept.codexradar:concept.openairadar:hardstop-kernel-agent-preemptionradar:stop-hook-coerced-api-changeradar:codex-thread-revert-supportradar:openai-codex-auto-reviewradar:charter-durable-agent-control-planeradar:grove-long-running-agent-protocol
queries asked of Scott's wikis
- agent interrupt / cancel / steering mechanisms in my coding-agent harnesses
- human-in-the-loop approval and pause patterns I've built or written about
- agent orchestration: mid-run control of long-running agent tasks
- cost of letting an agent run on a wrong trajectory (token waste, recovery patterns)
- safety boundary primitives in agent harness design (kill switches, sandboxing)
- Codex / OpenAI coding agents in my dev projects and work history
Measured heat
now 0 pts/hpeak 2 pts/hcomments 0/hpeers p16momentum: steady2 platformsage 99h
points/hour across evidence · reading as of 2026-10-12 02:59:37.977291+11:00 · deterministic, not a model opinion
How the heat travelled
pace: p37 vs 1247 stories at the 96h mark (now 99h old) — ahead of agentgate-signed-agent-receipts (1.3x), behind acs-local-skill-risk-catalog (0.8x)
Evidence (2) — ⭐ canonical anchor
Interpretation history
2026-10-09T01:43:09Z
grounded: converges/high — OpenAI Codex shipping 'Instant Interrupts' is a consequential other party arriving at the prompt-interrupt architecture Scott already defined and the Codex CLI
2026-10-09T01:34:26Z
case created — OpenAI Codex repo PR #51812 adds 'Instant Interrupts' — a direct harness-level control feature for the flagship coding agent, distinct from the Agents API case and highly relevant to agent orchestration and safety.
Decision trace
- 10-09 13:41feedback_briefingScott vote via UI
- 10-09 13:26attention_communicatedOpenAI Codex shipped 'Instant Interrupts' (PR #51812), a direct harness-level control feature for the flagship coding agent. This implements the low-latency prompt-interrupt architecture Sco
- 10-09 13:26attention_routeVendor convergence on the exact interrupt primitive Scott's framework treats as decisive. Ships today in the harness Scott uses. The Accountability.md skill and this PR together show the governan
- 10-09 13:18attention_routeVendor convergence on the exact interrupt primitive Scott's framework treats as decisive. Ships today in the harness Scott uses. The Accountability.md skill and this PR together show the governan
- 10-09 13:11attention_candidatecreate
- 10-09 12:43groundOpenAI Codex shipping 'Instant Interrupts' is a consequential other party arriving at the prompt-interrupt architecture Scott already defined and the Codex CLI gap he documented in Two Vendo
- 10-09 12:34createOpenAI Codex repo PR #51812 adds 'Instant Interrupts' — a direct harness-level control feature for the flagship coding agent, distinct from the Agents API case and highly relevant to agent o