fix(agent): seed CLAWBOX.md so the agent knows device conventions - #82
Merged
KrasimirKralev merged 2 commits intoApr 19, 2026
Merged
Conversation
…x conventions Root cause: the OpenClaw agent's session-start context (AGENTS.md, BOOTSTRAP.md, etc.) knows nothing about ClawBox-specific device conventions. When the user asks 'is skill X installed?', the agent falls back to guessing paths and looks in /home/clawbox/.npm-global/lib/node_modules/openclaw/skills — the package's built-in skills directory, not where the App Store writes user-installed skills (~/.openclaw/workspace/skills/). It then reports 'skill not installed' despite the file being on disk. Same class of problem for browser automation (agent doesn't know browser_* MCP tools exist) and for skill installation (agent would copy files manually instead of going through app_install). Adds config/clawbox-workspace-guide.md — a concise markdown file that documents: - Where user-installed skills live (~/.openclaw/workspace/skills/) and explicitly where they do NOT live (the npm-global path) - That skill install/uninstall must go through app_install / app_uninstall (so the App Store UI stays consistent and gateway reload actually fires) - The full browser_* MCP tool set with a standard launch → screenshot → click/type → screenshot workflow, plus the 'do not use ui_open_app(browser)' caveat (that opens the Setup panel) - The ui_*, app_*, webapp_* tools inventory for UI operations - File-system + network pointers (project dir, config path, device URLs) gateway-pre-start.sh now seeds this file as ~/.openclaw/workspace/CLAWBOX.md whenever: - The workspace dir exists (so we don't race the first gateway boot before OpenClaw has initialized it) - The shipped template differs from the on-disk copy (cmp -s), so edits the agent or user made locally aren't clobbered on every restart Also appends a one-line reference to AGENTS.md (if it exists and doesn't already mention CLAWBOX.md), so the agent's default session-start bootstrap flow naturally loads our guide. Verified on Jetson: first run seeded both files, second run logged 'skipping' for both (idempotent), syntax clean via bash -n.
|
Warning Rate limit exceeded
Your organization is not enrolled in usage-based pricing. Contact your admin to enable usage-based pricing to continue reviews beyond the rate limit, or try again in 50 minutes and 33 seconds. ⌛ How to resolve this issue?After the wait time has elapsed, a review can be triggered using the We recommend that you space out your commits to avoid hitting the rate limit. 🚦 How do rate limits work?CodeRabbit enforces hourly rate limits for each developer per organization. Our paid plans have higher rate limits than the trial, open-source and free plans. In all cases, we re-allow further reviews after a brief timeout. Please see our FAQ for further information. ℹ️ Review info⚙️ Run configurationConfiguration used: Path: .coderabbit.yaml Review profile: ASSERTIVE Plan: Pro Run ID: 📒 Files selected for processing (2)
📝 WalkthroughWalkthroughA new ClawBox workspace integration guide is added as documentation, along with a shell script modification that automatically seeds this guide into the workspace during gateway startup, conditionally appending references to session-start context files. Changes
Estimated code review effort🎯 2 (Simple) | ⏱️ ~10 minutes Poem
🚥 Pre-merge checks | ✅ 3✅ Passed checks (3 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 3
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In `@config/clawbox-workspace-guide.md`:
- Line 54: The doc entry for the ui_open_app built-in is using the wrong Browser
Setup app ID; update the description to use the actual app ID "browser" (not
"browser-setup") so agents call ui_open_app with appId === "browser", matching
the MCP special-case that detects Browser Setup by checking appId === "browser";
ensure the table row for `ui_open_app` mentions the supported app IDs and
explicitly lists "browser" for Browser Setup to avoid mismatched calls.
- Around line 9-13: Update the doc to say the agent should check the configured
workspace value (agents.defaults.workspace) as resolved by the openclaw config
resolver (the logic in openclaw-config that reads ~/.openclaw/openclaw.json and
falls back to ~/.openclaw/workspace) rather than assuming only the default path,
and clarify that built-in npm package skills
(/home/clawbox/.npm-global/…/openclaw/skills) are still not to be checked;
explicitly mention agents.defaults.workspace and the config resolution so
readers know where the runtime path comes from.
In `@scripts/gateway-pre-start.sh`:
- Around line 178-184: The script currently hard-codes CLAWBOX_WORKSPACE and
always replaces CLAWBOX.md when sources differ, which overwrites local edits;
change it to use the resolved workspace value (the same source used by
openclaw-config.ts — i.e., use the agents.defaults.workspace resolution instead
of "/home/clawbox/.openclaw/workspace") and make seeding idempotent by only
installing CLAWBOX.md if the destination does not exist (do not copy when files
differ), ensure variables CLAWBOX_WORKSPACE, CLAWBOX_GUIDE_SRC and
CLAWBOX_GUIDE_DST remain properly quoted, and add minimal error handling around
install (check source exists and return non-zero on failure) so the script
preserves local edits and uses the canonical workspace location.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Path: .coderabbit.yaml
Review profile: ASSERTIVE
Plan: Pro
Run ID: 747b7067-f91e-4167-a7b8-7a076068af54
📒 Files selected for processing (2)
config/clawbox-workspace-guide.mdscripts/gateway-pre-start.sh
Contributor
Author
|
@coderabbitai review |
✅ Actions performedReview triggered.
|
…esolve workspace, seed-if-missing Three findings on PR ID-Robots#82, all addressed: 1. CLAWBOX.md documented the ui_open_app app ID as 'browser-setup' in the tools table, but the MCP special-case in clawbox-mcp.ts checks appId === 'browser' for the Browser Setup panel. An agent following the doc literally would have passed an unknown app ID. The table entry now lists the actual built-in IDs (chat, files, settings, store, vnc, terminal, browser) and explicitly notes that 'browser' opens the Setup panel — not a real browsing session — and points back to the browser_* section for real web browsing. 2. The skills-path section and the pre-start script both hard-coded ~/.openclaw/workspace/skills/ as the answer to 'where do skills live?'. The actual source of truth is agents.defaults.workspace in openclaw.json (with a fallback to ~/.openclaw/workspace when unset), which is what getSkillsDir() on the ClawBox API side reads. Guide now describes that resolution and gateway-pre-start.sh computes the workspace via Python: reads agents.defaults.workspace, expands ~, treats relative paths as relative to ~/.openclaw, and falls back to the default on missing/invalid config. 3. CLAWBOX.md was being overwritten on every gateway start whenever its content differed from the shipped template (cmp -s check). That clobbers any local customization the user or agent has made. Switched to seed-if-missing: only write when the file doesn't exist. Operators who want to pull in an updated template can delete the on-disk copy; next gateway start re-seeds.
Contributor
Author
|
@coderabbitai review |
✅ Actions performedReview triggered.
|
KrasimirKralev
added a commit
to KrasimirKralev/clawbox
that referenced
this pull request
Apr 19, 2026
14 tests had been red on beta since before the Gemma/provider-switch work in PRs ID-Robots#82/ID-Robots#83 landed. Three separate root causes: 1. Test mocks drifted from the configure route's real imports. The route now calls `applyModelOverrideToAllAgentSessions` and `parseFullyQualifiedModel` (added in ID-Robots#83 for session sweeping), plus four functions from `@/lib/llamacpp` and one from `@/lib/local-ai-runtime`. Tests mocked none of these, so the first request-time call to any of them threw and the route returned 500 — producing the "expected 200 to be 500" cascade. Added the missing mocks with real-shape implementations. 2. vitest-under-bun clears mock implementations along with call history in `vi.clearAllMocks()`. Factory defaults set inside `vi.mock(...)` survive `vi.resetModules` but not `mockClear`. Tests passed in isolation, failed in sequence. Re-apply the implementations in `beforeEach` so each test starts with a consistent mock surface. 3. The ai-models-step component test relied on `llamaCppIsActive` defaulting true, but the panel only sets it true when `currentProviderId === "llamacpp"`. Without that prop the "Gemma 4 is already configured" pill never rendered. Added the prop. Extras during simplifier pass: - Mirror real `parseFullyQualifiedModel` logic byte-for-byte (`idx <= 0 || idx === fq.length - 1`) so trailing-slash inputs reject as expected. Fixed drift in `chat-model.test.ts` too. - Extract shared proxy-URL constant via `vi.hoisted` to avoid duplicating the magic string across two mock factories. Result: 93/93 files, 1069/1069 tests pass — CI signal restored.
2 tasks
KrasimirKralev
added a commit
that referenced
this pull request
Apr 19, 2026
…90) 14 tests had been red on beta since before the Gemma/provider-switch work in PRs #82/#83 landed. Three separate root causes: 1. Test mocks drifted from the configure route's real imports. The route now calls `applyModelOverrideToAllAgentSessions` and `parseFullyQualifiedModel` (added in #83 for session sweeping), plus four functions from `@/lib/llamacpp` and one from `@/lib/local-ai-runtime`. Tests mocked none of these, so the first request-time call to any of them threw and the route returned 500 — producing the "expected 200 to be 500" cascade. Added the missing mocks with real-shape implementations. 2. vitest-under-bun clears mock implementations along with call history in `vi.clearAllMocks()`. Factory defaults set inside `vi.mock(...)` survive `vi.resetModules` but not `mockClear`. Tests passed in isolation, failed in sequence. Re-apply the implementations in `beforeEach` so each test starts with a consistent mock surface. 3. The ai-models-step component test relied on `llamaCppIsActive` defaulting true, but the panel only sets it true when `currentProviderId === "llamacpp"`. Without that prop the "Gemma 4 is already configured" pill never rendered. Added the prop. Extras during simplifier pass: - Mirror real `parseFullyQualifiedModel` logic byte-for-byte (`idx <= 0 || idx === fq.length - 1`) so trailing-slash inputs reject as expected. Fixed drift in `chat-model.test.ts` too. - Extract shared proxy-URL constant via `vi.hoisted` to avoid duplicating the magic string across two mock factories. Result: 93/93 files, 1069/1069 tests pass — CI signal restored.
This was referenced Aug 18, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
The OpenClaw agent's session-start context (AGENTS.md, BOOTSTRAP.md, etc.) has no ClawBox-specific guidance. Observed failure mode earlier today: user installed
self-improving-agentvia the App Store, asked the agent to confirm — the agentread-checked/home/clawbox/.npm-global/lib/node_modules/openclaw/skills(the npm package's built-in skills dir), saw nothing, and confidently reported the skill as not installed. The skill was actually at~/.openclaw/workspace/skills/self-improving-agent/the whole time.Same class of problem applies to the desktop-Chromium integration and to skill installs:
browser_*MCP tools exist → defaults toui_open_app("browser")which opens the Setup panel, not a browsing session.app_install— which skips the App Store registration + gateway reload the user needs to see / uninstall it from the UI.What changed
New
config/clawbox-workspace-guide.mddocuments:~/.openclaw/workspace/skills/) and do NOT live (the npm-global path)app_install/app_uninstall(so the App Store UI stays consistent + gateway reload actually fires)browser_*MCP tool inventory with the standard launch → screenshot → click/type workflow, plus the "do not useui_open_app("browser")" caveatui_*/app_*/webapp_*tools inventoryscripts/gateway-pre-start.shnow seeds this file as~/.openclaw/workspace/CLAWBOX.mdwhen:cmp -s) — so later agent/user edits aren't clobbered on every restartAlso appends a one-liner reference to
AGENTS.md(when it exists and doesn't already mention CLAWBOX.md), so the agent's normal session-start bootstrap loads our guide without us having to overwrite AGENTS.md.Test plan
bash -nsyntax clean## ClawBox integrationsection pointing at CLAWBOX.mdSummary by CodeRabbit
Release Notes
Documentation
Chores