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.
Daily Edition
illmethinks ClawCharts river for 2026-07-12
Ten source-clustered briefs on current agent-infrastructure trust boundaries, delivery paths, and operator controls.
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.
More:
- fix(server): redact agent config secrets in read and mutation responses
- Validate issue dispatch origin before heartbeat execution
- Fix runtime worktree policy gating
- fix(task-watchdog): stop letting routine timestamp churn re-flip the stop fingerprint
- fix(agents): 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.
More:
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.
More:
- fix(agent): preserve Gemini thought signatures in tool-call history
- docs(maintainers): fix dashboard workflow link
- feat(zerocode): show active resolved log path in Doctor diagnostics (#8650)
- [Feature]: Add user-behavior E2E coverage for first-run setup
- feat(memory): add Hindsight as a first-class memory backend
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.
More:
- fix(agent-runner): rescue undelivered unwrapped replies after the re-wrap nudge, with recap suppression
- feat: /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
- fix(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…
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.
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.