* feat(crestodian): AI-first conversational onboarding with typed-op guardrails
Interactive `openclaw onboard` (and bare `openclaw` on a fresh install) now
opens the Crestodian conversation: detection-backed first-run proposal
(Claude Code/Codex logins, API keys), persona AI turns for every free-form
message (configless local-runtime fallback, 60s deadline, deterministic
degradation), approval-gated typed operations, chat-hosted channel setup
(`connect <channel>`), config get/schema read ops with secret redaction, and
a post-write validation hook that feeds schema errors back for a self-fix
turn. Adds the additive gateway `crestodian.chat` method so app clients run
the same conversation. Classic wizard stays behind --classic/explicit flags;
non-interactive automation unchanged; `--modern` becomes a deprecated alias
for `openclaw crestodian`.
* feat(macos): Crestodian chat onboarding and importance-ordered permissions
Replace the gateway step-wizard page with a Crestodian chat over the new
crestodian.chat method (works before any model auth exists), sort the
permissions page by importance with no scrolling, drop the redundant manual
refresh, and bump the onboarding version.
* feat(crestodian): run the custodian on the real agent loop with a ring-zero tool
Crestodian conversations now execute through the same embedded agent runner
as regular agents: a persistent agent session with a single construction-gated
`crestodian` tool wrapping the typed operations (read actions free; mutations
require approved=true asserted from explicit user consent, audited, with
post-write config validation fed back into the loop). The engine prefers the
loop (configured models or the Codex app-server fallback) and degrades to the
single-turn planner, then to deterministic commands. Setup approval seeds the
crestodian exec approval so local model harnesses can run; the configless
Codex backend config now enables exec and direct tool loading (it was
dead-on-arrival behind tools.exec.mode=deny and the tool-search index).
* test(crestodian): type the engine mock signatures for the core test lane
* fix(crestodian): map the advertised create_agent tool action
* fix(crestodian): host-verified approval arming and a macOS setup completion gate
Review findings: the model-supplied approved flag alone could authorize
ring-zero mutations (prompt injection / model error), and removing the macOS
wizard gate let users Next past the Crestodian page with nothing configured.
Mutating tool actions now also require host-verified consent (the engine arms
approval only when the user's actual message is an explicit yes), and local
macOS onboarding blocks advancing until setup authored the config, using the
same signal the old step wizard checked.
* fix(crestodian): bind approval to the exact proposed operation and gate dot navigation
A generic yes no longer authorizes arbitrary mutations: denied mutating tool
calls register a canonical operation fingerprint (host-owned, per session),
and an armed turn executes only the identical call, once. The denial message
is arming-aware so the approved turn self-heals in one roundtrip, and the
agent protocol pre-registers proposals. macOS onboarding page dots now honor
the same setup-completion gate as the Next button.
* fix(crestodian): redact sensitive wizard answers, skip logged-out CLIs, gate programmatic advance
Sensitive channel-wizard answers (tokens, passwords) are redacted from the
AI-visible conversation history; setup and the onboarding welcome never pick
or advertise a definitively logged-out CLI as the model; and macOS
handleNext() honors the page gates for programmatic callers (chat handoff)
just like the Next button.
* fix(crestodian): align the onboarding welcome's configured predicate with the app gate
A valid config carrying only a default model (partial/hand-written) now still
gets the first-run proposal instead of the ready guide, so the macOS setup
gate can always be satisfied from the conversation.
* fix(crestodian): armed turns can never mint their own executable proposal
An approval-mismatched call inside an armed turn no longer re-registers and
invites a retry (which let the model swap the approved operation for another
in the same turn); it voids the approval entirely and requires a fresh yes.
Proposals register only in unarmed turns, which the agent protocol already
does when proposing.
* fix(onboard): route any explicit setup flag to the classic wizard
* fix(ci): satisfy new lint rules, tool-display guard, and generated artifacts for crestodian
* chore(i18n): refresh native inventory after permissions copy wrap
* fix(crestodian): harden conversational onboarding
* docs(crestodian): document conversational onboarding
* test(crestodian): type embedded runner mock
* fix(crestodian): close onboarding security gaps
* chore: retrigger ci
Update OpenClaw ClawHub docs and user-facing copy for canonical owner-qualified skill routes.\n\nEvidence:\n- pnpm docs:list\n- pnpm test src/plugins/clawhub.test.ts src/cli/plugins-cli.install.test.ts src/gateway/server-methods/skills.clawhub.test.ts ui/src/ui/views/skills.test.ts\n- pnpm exec oxfmt --check --threads=1 docs/clawhub/cli.md docs/clawhub/publishing.md docs/cli/skills.md docs/help/faq.md docs/start/showcase.md docs/tools/creating-skills.md docs/tools/skills.md src/gateway/server-methods/skills.clawhub.test.ts src/plugins/clawhub.test.ts src/plugins/clawhub.ts ui/src/ui/views/skills.test.ts\n- git diff --check\n- exact-head hosted CI passed for 8530374388d8a73235b2ac8444b95a4a4c7d0f1c\n\nNote: repo-native scripts/pr prepare-run was attempted; local broad pnpm test was stopped after unrelated existing failures in agent/media/provider shards, while hosted exact-head CI and targeted ClawHub route/copy validation were green.
Fix live model inference edge cases across provider streaming, model switching, outbound delivery, and gateway tool resolution.
Includes live/provider issue fixes and leaves #89100 explicitly partial for the remaining FM-2 group routing case.
* feat(browser): add optional vision understanding to screenshot tool
* fix(browser): wrap vision output as external content, enforce maxBytes, forward auth profiles
* fix(browser): remove no-op scope/attachments config, drop profile pass-through lacking runtime support
* feat(media-understanding): add profile/preferredProfile to DescribeImageFileWithModelParams and forward to describeImage
* style(browser): add curly braces to satisfy eslint curly rule
* fix(browser): correct tools.browser.enabled help text to match actual behavior
* fix(browser): thread agentDir/workspaceDir from plugin tool context into browser vision
* refactor(browser): move vision config from tools.browser to browser.models
The browser plugin's vision configuration now lives on the top-level
`browser` config namespace (browser.models, browser.visionEnabled,
browser.visionPrompt, etc.) instead of `tools.browser`. This aligns
with the plugin's existing config location and avoids confusion between
tool-level and plugin-level settings.
- Remove tools.browser from ToolsSchema and ToolsConfig
- Add models/vision* fields to BrowserConfig and its zod schema
- Update getBrowserVisionConfig to read from cfg.browser
- Update schema help, labels, and quality test
- Update vision.test.ts to use new config shape
* docs(browser): add screenshot vision configuration section
Document the new browser.models config for automatic screenshot
description via vision models, enabling text-only main models to
reason about web page content.
* fix(browser): remove deliverable media markers from vision result, drop unused import
P1: Vision-success path no longer exposes the raw screenshot as
deliverable media (removes MEDIA: line and details.media.mediaUrl).
This prevents channel delivery from auto-sending sensitive page content
when the intended output is a text description.
P2: Remove unused ToolsMediaUnderstandingSchema import that would fail
noUnusedLocals typecheck.
* fix(browser): add command/args fields to browser models schema
The browser vision model schema uses .strict(), so CLI-type entries
with command/args were rejected by TypeScript. Add these fields to
align with MediaUnderstandingModelSchema.
* chore(browser): remove debug console.log statements
* fix(browser): harden screenshot vision result against MEDIA: directive injection and restore image sanitization on failure fallback
ClawSweeper #84247 review round 2:
P1 (security, high): neutralize line-start MEDIA: directives in vision descriptions
before wrapping with wrapExternalContent. The agent media extractor scans every
browser tool-result text block via splitMediaFromOutput which treats line-start
MEDIA: as a trusted local-media delivery directive, and browser is on the
trusted-media allowlist. Without neutralization, page or vision-provider output
containing 'MEDIA:/tmp/secret.png' could synthesize a channel-deliverable media
artifact from untrusted content. wrapExternalContent itself does not strip
line-start directives. Introduce neutralizeMediaDirectives in vision.ts that
prepends '[neutralized] ' to any line whose trimStart() begins with MEDIA:
(case-insensitive), defanging the parser anchor while keeping the original
text human-readable.
P2 (compatibility): pass resolveRuntimeImageSanitization() to imageResultFromFile
in the vision-failure catch fallback. The non-vision screenshot path already
forwards this option (d5cc0d53b7) so configured agents.defaults.imageMaxDimensionPx
takes effect. Without this fix, any provider timeout/error silently bypasses the
sanitization guard and returns a raw full-resolution screenshot.
Regression coverage:
- vision.test.ts: 6 unit cases for neutralizeMediaDirectives (no-op fast path,
mid-line MEDIA: untouched, line-start defanged, leading-whitespace defanged,
case-insensitive, multiple directives per blob).
- browser-tool.test.ts: 2 integration cases that drive the full screenshot
tool execute path:
- 'neutralizes MEDIA: directives in vision text and does not attach media'
asserts no line matches /^\s*MEDIA:/i in returned text, secret path text
is preserved verbatim, details.media is absent, and imageResultFromFile
is not called on the success path.
- 'preserves screenshot image sanitization on vision failure fallback'
mocks describeImageFileWithModel to reject and asserts the fallback
imageResultFromFile call receives imageSanitization: {maxDimensionPx:1600}
plus the 'browser screenshot vision failed' extraText.
* fix(browser): apply clawsweeper fallback media fix from PR #84247
* refactor: reuse media image understanding for browser screenshots
* refactor: use structured media delivery
* test: update music completion media instruction expectation
* fix: trim buffered reply directive padding
* test: refresh codex prompt snapshots for message media aliases
---------
Co-authored-by: scotthuang <scotthuang@tencent.com>
Co-authored-by: Peter Steinberger <steipete@gmail.com>
Add an owned Kysely dialect for native node:sqlite, raise the Node 22 floor to 22.16+ for StatementSync.columns(), and cover select/returning/stale insert id behavior.
Replaced 60 typography characters (curly quotes, apostrophes, em/en
dashes, non-breaking hyphens) with ASCII equivalents per
docs/CLAUDE.md heading and content hygiene rules.
- docs/start/openclaw.md: 10 chars; removed the duplicate '# Building
a personal assistant with OpenClaw' H1 (Mintlify renders title from
frontmatter).
- docs/platforms/mac/remote.md: 10 chars; removed the duplicate
'# Remote OpenClaw (macOS ⇄ remote host)' H1 (the U+21C4 codepoint
and parens both produced brittle anchors).
- docs/tools/thinking.md: 10 chars
- docs/reference/templates/BOOTSTRAP.md: 10 chars (kept the in-body
'# BOOTSTRAP.md - Hello, World' heading because the page is a
template whose content is meant to be copied verbatim into a
workspace BOOTSTRAP.md).
- docs/plugins/sdk-provider-plugins.md: 10 chars
- docs/platforms/macos.md: 10 chars
