Daily Edition Sources +6

Agent Control Hints Become Runtime Contracts

Fresh OpenClaw, Codex, and Hermes changes show agent projects turning speed modes, chat metadata, and worker instructions into explicit runtime surfaces.

Diagram Punk poster showing OpenClaw fastMode auto, Codex typed passthrough, Hermes injected guidance, and a not-a-shared-standard caveat.
Diagram Punkcontrol hints are contracts now, but not a shared standard.
repos openclaw/openclaw + 2 more evidence
6 source signals 3 repos 3 linked commits
Evidence: 3 linked commits / June 23, 2026 / Daily Edition
Open Edition Evidence below

openclaw/openclaw commit 2b75806197 adds one-turn fastMode: "auto" support to chat.send, making a speed hint time-bound protocol behavior rather than a persistent hidden toggle.

Facts

  • OpenClaw now documents fastMode: "auto" with a default 60-second cutoff and per-model or per-request fastAutoOnSeconds overrides.
  • openai/codex commit 5b95745eae replaces generic ResponseItem.metadata with typed internal_chat_message_metadata_passthrough.
  • NousResearch/hermes-agent commit 84e1d31e5 moves kanban worker and orchestrator rules from bundled skills into injected guidance.

Evidence

The receipts are OpenClaw gateway protocol text and runner code, Codex protocol schemas and Rust models, and the Hermes kanban guidance diff. Each turns an implicit control hint into a named interface the runtime can preserve, route, or display.

Context

For agent operators, the practical question is no longer only whether a model can act. It is whether speed, provenance, and delegated work are represented as contracts that clients and maintainers can inspect before something goes wrong.

Limits

These are independent patches, not a shared standard. They do not prove that the controls are stable across clients. Watch whether future releases expose these contracts in user-facing diagnostics, logs, and compatibility promises.

Evidence Trail

Receipts below the story

The article above is the public narrative. This section keeps the source trail, limits, and reporting notes on the same page.

Edition
DateJune 23, 2026
LaneDaily Edition
Confidence87%
Sources6
Reposopenclaw/openclaw, openai/codex, NousResearch/hermes-agent

Primary Evidence

Evidence Limits

  • These commits are from independent projects; they do not prove a common standard.
  • The evidence shows protocol, metadata, and guidance surfaces, not user adoption.
  • The OpenClaw evidence is broad and includes many files; the article relies only on the fast-mode auto protocol and runner behavior.
  • The Hermes evidence concerns kanban worker guidance, not all forms of multi-agent orchestration.

Atlas Context

Open Atlas ›
Agent AtlasChapter 4

Loops, Workflows, and Recovery

Repeated model-tool interaction is only one control structure. Production runtimes combine probabilistic loops with host-defined workflows, queues, interrupts, terminal states, and recovery operations.
Letters & Corrections

Send a note to the desk

Corrections, missing context, or a follow-up lead.

Same Edition

Open issue ›
Reported +12 Jun 23, 2026

Google Antigravity After the Transition

google-antigravity/antigravity-cli + google-gemini/gemini-cli evidence
12 sources 2 repos 2 linked commits