static-tool-verify
Skill0 starsChanged 9 months agoMITView on GitHub
Browser-verify a self-contained single-page tool before shipping — start it via .claude/launch.json, exercise the core mechanic, assert zero console errors, screenshot, and catch the content/logic bugs (seasonal incoherence, misleading severity labels, dead interactions) that pass a syntax check but fail a human. Use when "verify the tool", "check the page before deploy", "browser-test this POC", or as Phase 3 of poc-tool-forge. Never trust an agent's self-report of "done".
In this repository47 other files it gives its agents.
- agent-teachingSkill
- cf-pages-trespies-deploySkill
- compression-ritualSkill
- context-ingestionSkill
- decision-propagationSkill
- documentation-auditSkill
- era-architectureSkill
- file-managementSkill
- frontend-from-backendSkill
- handoff-protocolSkill
- health-auditSkill
- house-poc-builderSkill
- implementation-promptSkill
- iterative-scoutingSkill
- mcp-cloudflare-builderSkill
- mcp-server-builderSkill
- memory-gardenSkill
- multi-surface-strategySkill
- parallel-tracksSkill
- patient-learning-protocolSkill
- planning-with-filesSkill
- poc-tool-forgeSkill
- pre-implementation-checklistSkill
- process-extractionSkill
- product-positioningSkill
- project-explorationSkill
- release-specificationSkill
- repo-context-syncSkill
- repo-statusSkill
- research-modesSkill
- research-synthesisSkill
- retrospectiveSkill
- seed-extractionSkill
- seed-librarySkill
- seed-to-skill-converterSkill
- semantic-clustersSkill
- skill-audit-upgradeSkill
- skill-creationSkill
- skill-maintenanceSkill
- spec-constellation-to-prompt-suiteSkill
- status-templateSkill
- status-writingSkill
- strategic-scoutSkill
- strategic-to-tactical-workflowSkill
- web-research-externalSkill
- workflow-routerSkill
- workspace-navigationSkill
The file
We split this skill into the rules it gives an agent
5 rules. About 0 are common, 3 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 20
If real-data: spot-check that the inlined DATA matches the source snapshot (a couple of values), to confirm nothing was paraphrased or …
- Very specific: line 24
Use the preview MCP (preview_*), driven off .claude/launch.json:
- Distinctive: line 36
Look for output that is *valid* but *wrong*:
- Distinctive: line 41
Dead interactions: do "see also" / regenerate / toggles actually do something?
- Distinctive: line 43
When you find one, fix the source (root cause, per the debugging protocol), reload, re-assert — then move on.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
OpenClaw Integration: This skill is invoked by the Dojo Genesis plugin via
/dojo run static-tool-verify. The agent receives project context automatically via thebefore_agent_starthook. Usedojo_get_contextfor full state,dojo_save_artifactto persist outputs, anddojo_update_stateto record phase transitions and decisions.
static-tool-verify — prove it works before it ships
The discipline that caught real bugs this session. An agent reporting "verified, HTTP 200, zero
errors" is not verification — confirm it yourself. Part of poc-tool-forge; applies to anything
house-poc-builder produces (and any static single-page tool).
Step 1 — Structural / syntax (cheap, do first)
- File exists + non-trivial size; balanced
<script>/<style>tags. - Extract the inline JS and
node --checkit (catches the syntax errors that would white-screen the page). - If real-data: spot-check that the inlined
DATAmatches the source snapshot (a couple of values), to confirm nothing was paraphrased or invented.
Step 2 — Real browser pass (the part that matters)
Use the preview MCP (preview_*), driven off .claude/launch.json:
- Register a launch entry if missing:
{"name":"<slug>","runtimeExecutable":"python3", "runtimeArgs":["-m","http.server","<port>","--directory","trespies-poc/<slug>"],"port":<port>}. preview_startthe slug → get the serverId.- Exercise the core mechanic, don't just load the landing: set the input / click the real
generate button via
preview_eval(example chips often only prefill — trigger generation explicitly), then assert the artifact rendered (body length jumps, expected sections + SVG element counts present). preview_console_logs level:error→ must be empty.preview_screenshot→ eyeball the actual render (layout, the hero element, the disclaimer banner).
Step 3 — Catch what a syntax check can't (this is the point)
Look for output that is valid but wrong:
- Coherence: does generated vocab contradict the inputs? (A "Summer" almanac that cites a "March" driver — vocab pool wasn't season-filtered.)
- Honesty of framing: does a label overstate reality? (A public-health page labeling a quiet summer "HIGH" because severity was banded on cross-state rank instead of absolute %-of-peak.)
- Dead interactions: do "see also" / regenerate / toggles actually do something?
- Guardrail present: disclaimer banner + footer actually render in the artifact, not just the source. When you find one, fix the source (root cause, per the debugging protocol), reload, re-assert — then move on.
Step 4 — Hand off proof
Screenshot(s) of the working mechanic are the deliverable to the operator. Then proceed to
cf-pages-trespies-deploy.
Notes
- Foreground
sleepis blocked; to wait on something, use a background poll or apreview_evalreload with a short settle, then re-query. - Kill stray
python3 -m http.serveryou started manually; launch.json-managed servers are fine to leave.
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.

