crabbox
Skill88 starsChanged 43 days agoMITView on GitHub
Run this repository's GitHub-equivalent CI checks on Crabbox, rerun only failed lanes, reuse a clean lease while fixing failures, and verify native platform gates before pushing. Use whenever validation, CI reproduction, remote compute, clean-environment testing, or a pre-push confidence check is requested in this repository.
31 other projects use a version of this.
Adopted by
Most recently taken up by
- openclaw/agent-skillstheir own version, first seen Nov 14, 2026
- openclaw/crabboxtheir own version, first seen Nov 3, 2026
- OpenClawAndroid/openclaw-android-assistanttheir own version, first seen Oct 17, 2026
- bighamx/openclaw-android-node-apktheir own version, first seen Oct 9, 2026
- molt-bot/clawdbottheir own version, first seen Oct 8, 2026
- openclaw/waclitheir own version, first seen Oct 8, 2026
Used with
Projects with it also use openclaw-debugging, openclaw-testing and 4 more.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
clawsoulsSkillopenclaw-debugging
Debug OpenClaw model, provider, tool-surface, code-mode, streaming, and live/Crabbox behavior by choosing the right logs, probes, and proof path before changing code.
Also in 4 projects that use this
clawsoulsSkillopenclaw-testing
Choose, run, rerun, or debug OpenClaw tests, CI checks, Docker E2E lanes, release validation, and the cheapest safe verification path.
Also in 4 projects that use this
swyxioSkillautoreview
Perform a structured final code review of a local diff, commit, branch, or pull request. Use only when the user explicitly asks for autoreview, a second review, or a final review before commit, merge, or release, or when repository instructions require one. Do not trigger automatically for ordinary non-trivial edits.
Also in 9 projects that use this
mj-devingSkilldiscrawl
Discord guild archiving and search via discrawl CLI. USE WHEN Discord search, Discord history, Discord messages, Discord members, Discord channels, search Discord, Discord archive, Discord mentions, Discord server, Discord guild, discrawl.
Also in 3 projects that use this
siegertsSkillagent-transcript
Find and export agent chat session transcripts. Use when the user asks to export, save, find, or share a session transcript from any coding agent (Claude Code, Codex, Kiro, Cursor, Gemini CLI, etc.), wants to create an image of their conversation, or mentions agentpng.
Also in 3 projects that use this
spitoglouSkilltechnical-documentation
Create and improve technical documentation including architecture designs, PRDs, API docs, code explanations, and visual diagrams. Use when asked to document systems, explain codebases, create design documents, write PRDs, generate Mermaid/PlantUML diagrams, review architecture, write user stories, or create pull request descriptions. Triggers include "document this", "create a design doc", "explain this code", "write a PRD", "create a diagram", "architecture overview", "user story for", "PR description".
Also in 4 projects that use this
Other versions31 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 31
- openclaw/openclaw★ 395ksame name
- moltbot/clawdbot★ 394ksame name
- steipete/warelay★ 394ksame name
- steipete/clawdis★ 394ksame name
- moltbot/moltbot★ 394ksame name
- clawdbot/clawdbot★ 394ksame name
- molt-bot/clawdbot★ 391ksame name
- steipete/gogcli★ 8.6ksame name
- openclaw/gogcli★ 8.6ksame name
- steipete/Peekaboo★ 5.5ksame name
21 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository4 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- batch-grill-meSkill
- improveSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
14 rules. 0 are common, 10 are distinctive, and 4 are very specific to this project.
0
Common
Most of the 31 edited versions we read keep these.
10
Distinctive
This author's own choices. Worth a look.
4
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 9
Use the repository entrypoints.
- Distinctive: line 15
Run these before provisioning capacity:
- Distinctive: line 36
Run all Linux PR jobs with up to twelve boxes in parallel:
- Distinctive: line 50
Do not route a product or test failure to Hetzner as though it were a capacity failure.
- Distinctive: line 53
Run one lane while fixing it:
- Distinctive: line 59
When the Windows umbrella lane has already identified one failing package, use the configured focused package job before paying to rerun …
- Distinctive: line 99
After a grouped run, rerun only its failures:
- Distinctive: line 130
Run the entire platform set only when both providers are ready:
- Very specific: line 136
Never claim the full GitHub matrix is preflighted when only pr-linux ran.
- Very specific: line 137
When a change can affect Windows process handling, paths, packaging, or native dependencies, run pr-unit-windows.
- Distinctive: line 163
Read the first authoritative product/test failure in that named job.
- Distinctive: line 165
If sync or installed state is suspect, stop the lease and reproduce on a fresh box.
- Very specific: line 169
When it passes, use retry to clear the remaining failed set.
- Distinctive: line 170
Before pushing, run the relevant group and report exactly which native jobs were or were not executed.
“Common” is measured: kept by at least 6 in 10 of the 31 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Crabbox CI
Use the repository entrypoints. .crabbox.yaml owns lease routing and named
jobs; script/cbx-ci.ts owns grouping, bounded parallelism, result persistence,
and failure-only retries; platform scripts own the commands run on each box.
Start safely
Run these before provisioning capacity:
./script/cbx doctor
./script/cbx-ci.ts list
./script/cbx-ci.ts dry-run pr-linuxscript/cbx loads the repository's configured credentials and downloads and
checksum-verifies the tested Crabbox 0.40 CLI into .crabbox/bin; a global
Crabbox upgrade therefore cannot change CI behavior. Set CRABBOX_BIN only
when deliberately validating a candidate CLI release. Do not call bare
crabbox for this repository unless you have deliberately supplied the same
environment.
Dry-run prints the exact lease and command plan and spends no cloud capacity. Inspect unexpected provider, target, architecture, server type, or command changes before a real run.
Run CI
Run all Linux PR jobs with up to twelve boxes in parallel:
./script/cbx-ci.ts run pr-linuxLinux named jobs use AWS by default. The runner cap is twelve concurrent jobs and is independent of provider capacity. Use the explicit Hetzner fallback only when AWS provisioning is unavailable:
./script/cbx-ci.ts run pr-linux-hetznerDo not route a product or test failure to Hetzner as though it were a capacity failure. A provider change is infrastructure triage, not a product fix.
Run one lane while fixing it:
./script/cbx-ci.ts run pr-unit-linuxWhen the Windows umbrella lane has already identified one failing package, use the configured focused package job before paying to rerun every package:
./script/cbx-ci.ts run focus-harness-windows
./script/cbx-ci.ts run focus-server-core-windowsFocused jobs are diagnostic subsets, not substitutes for their blocking PR
lanes. After the package passes, rerun pr-unit-windows to prove the complete
matrix and catch failures that the earlier umbrella run never reached.
Reuse an existing clean lease for a fast edit/test loop. Supplying a lease forces sequential execution so two jobs never mutate the same worktree at once:
./script/cbx list --provider aws --all --refresh
./script/cbx-ci.ts run --id <linux-slug> pr-linuxThe lease target must match every selected job. Use a Linux lease for
pr-linux and a Windows lease for Windows jobs; never send a mixed-platform
pr group through one --id. One reused Linux lease runs its lanes
sequentially. Parallel Linux lanes require multiple isolated AWS leases.
The repository sets sync.delete: false. This is required for warm native
Windows leases because Node and native addons can keep installed files open;
recursive sync deletion would fail before the test command starts. Normal
run performs an overlay sync. Add --no-sync only for repeated commands
after a successful sync. Use ./script/cbx run --id <slug> --sync-only when
only a sync is needed. sync.gitSeed: true and sync.baseRef: dev seed the
canonical public branch before the dirty-tree overlay when the transport
supports seeding. Native Windows manifest sync can omit .git, so
script/cbx-prepare-windows.ps1 independently hydrates the same public dev
commit before dependency installation; policy checks therefore use a real Git
index rather than a synthesized manifest. Overlay sync deliberately retains
remote files that were deleted locally: after deleting or renaming a source or
generated file, use a new lease or --fresh-sync before treating the result as
authoritative.
After a grouped run, rerun only its failures:
./script/cbx-ci.ts retryThe last result is stored at .crabbox/ci/last-run.json, which is ignored by
Git. Do not hand-edit the file or turn a failed result green without rerunning
the named job.
Native platform contract
Linux does not prove Windows or macOS behavior.
pr-unit-windowsuses AWS native-Windows boxes and executesscript/cbx-ci-windows.ps1. AWS credentials and quota must be configured in trusted user state, not committed to the repository.pr-e2e-desktop-macosusesprovider: sshand tag leases withclaxedo-macos. Configurestatic.host,static.user,static.port, andstatic.workRootin the trusted Crabbox user config (or theirCRABBOX_STATIC_*environment equivalents); never commit the host or key. The orchestrator serializes macOS jobs so they cannot mutate one durable host worktree concurrently.
Preview both native jobs with:
./script/cbx-ci.ts dry-run pr-nativeRun the entire platform set only when both providers are ready:
./script/cbx-ci.ts run prNever claim the full GitHub matrix is preflighted when only pr-linux ran.
When a change can affect Windows process handling, paths, packaging, or native
dependencies, run pr-unit-windows. When it can affect Electron packaging,
file:// rendering, desktop startup, embedded execution, or Host Connector
lifecycle, run pr-e2e-desktop-macos.
Historical triage order
Recent GitHub history makes these the expensive lanes. Use this order when several jobs fail, while still fixing the first authoritative failure inside each lane:
pr-unit-windows: native filesystem semantics (EPERM/fsync, path and state-fileENOENT), embedded-module resolution, and deadline kills have historically made this the largest source of failed runner time.pr-unit-linux: this umbrella lane can fail late in documentation guards, product closures, isolated production builds, or the full unit matrix.- Packaged macOS lifecycle: distinguish an app stuck on
loading.html, a signed-cloud dependency failure, and teardown/process-lifecycle failure. - Stateful browser E2E: inspect traces and the authoritative server log for missing UI state or a dead local server; do not treat a longer locator timeout as a diagnosis.
- Tier-real: preserve harness settlement, connection ownership, and multi-process state evidence when reproducing.
Failure workflow
- Read the first authoritative product/test failure in that named job. Do not repair it with fallback or synthesized events.
- If sync or installed state is suspect, stop the lease and reproduce on a fresh box. Do not debug product behavior on a corrupted reused box.
- Fix the canonical producer or contract locally.
- Rerun the one failed job, preferably on the same known-clean lease.
- When it passes, use
retryto clear the remaining failed set. - Before pushing, run the relevant group and report exactly which native jobs were or were not executed.
Crabbox job runs created without --id use stop: auto, so their leases are
released on completion. A caller-supplied lease remains available for the
interactive loop; stop it explicitly when finished:
The failure digest can still print a rerun command containing the released slug. That command is usable only when the run started from a caller-supplied lease; after an auto-stop named job, provision or supply a new lease instead.
./script/cbx stop <slug>Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

