OpenClaw resolves web-search secret references for Ollama
OpenClaw is resolving stored web-search credentials before Ollama-backed runs, tightening the seam between local inference and external search services.
Front Page
A small, severe wire desk for agent infrastructure.
Today’s agent-infrastructure wire is about the seams operators inherit: credential resolution, secret redaction, multimodal content, provider metadata, reply delivery, and controls for scheduled work. Quiet repositories remain useful signals, but silence earns a watchlist—not a headline dressed as certainty.
The wire
Hermes Agent adds a Russian desktop locale
Hermes Agent is extending its desktop interface with Russian localization, a concrete expansion of the project’s user-facing surface beyond the English-language operator base.
Paperclip redacts agent configuration secrets from API responses
Paperclip is removing agent credentials from server read and mutation responses, narrowing an exposure path in the control plane that manages office-agent configuration.
Inspected changes:
- Server redact agent config secrets in read and mutation responses
- Validate issue dispatch origin before heartbeat execution
- Runtime worktree policy gating
- Task Watchdog stop letting routine timestamp churn re-flip the stop fingerprint
- Subagents detach finance events instead of deleting them on agent removal
Nanobot hardens its loop for multimodal message content
Nanobot now guards text-only string handling when message content arrives as multimodal lists, preventing image-and-text turns from falling into a scalar-content assumption.
ZeroClaw preserves Gemini thought signatures across tool calls
ZeroClaw is retaining Gemini thought signatures in tool-call history, keeping provider-required reasoning metadata intact as agent turns cross tool boundaries.
NanoClaw retries replies lost during output re-wrapping
NanoClaw is rescuing undelivered replies after its re-wrap nudge while suppressing redundant recaps, targeting a delivery failure at the agent-to-chat boundary.
Inspected changes:
- Agent Runner rescue undelivered unwrapped replies after the re-wrap nudge, with recap suppression
- /add-audit skill — opt-in local audit log for the ncl surface
- Guard seam: one decision function for every privileged action (guarded-actions phase 2)
- Tasks: one-door delivery — send_message is the only path out of a task session
- Agent Runner stall watchdog to recover from hung in-flight tools
PicoClaw adds skill toggles and an immediate cron run control
PicoClaw is bringing skill enable/disable state and a Run Now action into its operator surface, making scheduled automation easier to inspect and intervene in.
OpenFang remains on maintenance watch after a quiet public activity sweep
OpenFang’s current public surfaces did not yield a fresh PR or issue strong enough to support a product claim. Its continued chart presence is useful demand context, but the evidence warrants a watchlist rather than…
IronClaw refreshes Google OAuth tokens in canary runs
IronClaw’s canary workflow now refreshes Google OAuth tokens per run, making integration tests less dependent on stale credentials and more representative of live authentication behavior.
Inspected changes:
NullClaw proposal would reuse Grok CLI login sessions as a provider
A NullClaw issue proposes running Grok through existing grok-cli login sessions, raising a useful boundary question around convenience, credential ownership, and unmetered provider access.
Hermes Agent keeps desktop activity visible through context compression
Hermes Agent’s desktop indicator now survives context compression, closing a small but consequential observability gap while long-running sessions compact their history.
IronClaw’s public signal narrows to release machinery
IronClaw’s freshest inspected artifact is release housekeeping rather than a product or operator change. The useful signal is the absence: deployment motion is visible, but the public evidence does not support a larger…
Inspected changes:
Nanobot stops its dream-memory loop from committing empty work
Nanobot’s periodic memory process now skips no-op commits, reducing repository noise and making actual memory changes easier to audit.
Inspected changes:
NanoClaw narrows duplicate-send detection to the active turn
NanoClaw now scopes identical-send detection to the turn in flight, reducing the risk that old output suppresses a legitimate response in a later agent turn.
Inspected changes:
NullClaw report warns shared A2A bearer routes may cross caller boundaries
A NullClaw bug report says a shared bearer route can reuse task and context state across callers, putting identity separation—not transport reachability—at the center of its A2A risk.
OpenClaw separates avatar limits from its agent-list gateway response
OpenClaw’s gateway now treats inline avatar payloads as a separate size-boundary problem, reducing the chance that agent discovery turns into an oversized-response path.
Inspected changes:
OpenFang adds MiniMax M3 to its model catalog
OpenFang is wiring MiniMax M3 into its provider catalog, widening model choice while its public repository remains comparatively quiet on broader operator-facing work.