Dont read the code ego
Run source-blind black-box product checks on macOS with Ego Lite from a validated release-candidate request, then produce auditable evidence and failure routing without opening the source repository. Use when testing or inspecting a deployed web product as a simulated user, especially as the optional Mac-side companion to the Linux-capable dont-read-the-code four-role pipeline.From its SKILL.md
npx -y skills add YYshen07/dont-read-the-code-egoAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
2 things to look at
- 24 days oldThe repository was created 24 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
- 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
SKILL.md
2.9 KB, 573 tokens by cl100k_base, as published. Nobody here has run it
不要看代码!Ego 检查
Test the released behavior, not the implementation. This Skill requires macOS,
Ego Lite, its ego-browser Skill, and Node.js. It does not require SwarmForge,
tmux, Babashka, or a source checkout.
Read before acting
Read:
references/simulated-user.mdreferences/safety-policy.mdreferences/failure-routing.md
Then read the installed ego-browser Skill in full before invoking its CLI.
Prepare a clean-room run
Check the Mac runtime:
scripts/check-prereqs.sh
Use a directory that does not contain the product repository. It should contain only:
ego-check-request.json- The selected persona file.
- Explicitly approved upload fixtures.
- An empty evidence output directory.
Validate the request:
node scripts/validate-request.mjs ego-check-request.json
Reject a request whose deterministic quality result is not passed, whose
starting URL leaves the origin allowlist, or whose account kind is not test or
fixture.
Use assets/persona.example.json when no project-specific persona exists. Do not
invent implementation details to make the task easier.
Execute with Ego Lite
- Create one Ego Task Space named with the request and run IDs.
- Open only the request's starting URL.
- Act through screenshots, semantic snapshots, visible text, accessible names, keyboard, mouse, and ordinary form controls.
- Stay inside the request's allowed origins.
- Default-deny purchases, publishing, messaging, durable deletion, account changes, legal consent, and unlisted uploads.
- Treat user ownership, login, CAPTCHA, or manual approval as a hard handoff.
- Verify observable postconditions; a click is not evidence.
- Capture evidence before completing the Task Space.
Never use raw CDP, page or locator evaluation, browser/server fetch, source files, tests, selectors, internal APIs, Git history, or another agent's transcript.
Produce the result
Create run-result.json from assets/run-result.example.json. Preserve the
request ID, source commits, and quality-report digest.
Validate:
node scripts/validate-bundle.mjs ego-check-request.json run-result.json
Return the validated result and evidence to the core pipeline or human operator. Do not modify product code.
passedroutes tospecifieror release owner.product-defectroutes tocoder.specification-gaproutes tospecifier.blockedroutes to the operator for environment, fixture, authentication, or safety resolution.
What ships with it: 16 files
44.5 KB alongside SKILL.md, 4 of them executable
agents/
- openai.yaml228 B
assets/
references/
- failure-routing.md1.4 KB
- safety-policy.md1.9 KB
- simulated-user.md1.5 KB
scripts/
- check-prereqs.shruns516 B
- validate-bundle.mjsruns3.0 KB
- validate-request.mjsruns5.5 KB
- validate-run.mjsruns12.5 KB
- .gitignore37 B
- LICENSE1.0 KB
- README.md1.6 KB