ClawHub

flow test

v0.0.1

Designs agent-evaluated flow tests for browser tasks, LLM outputs, and tool workflows. Invoke when exact asserts are brittle and semantic success matters mor...

⭐ 1· 98·0 current·0 all-time
by@qipengguo
MIT-0
Download zip
LicenseMIT-0 · Free to use, modify, and redistribute. No attribution required.
Security Scan
VirusTotalVirusTotal
Benign
View report →
OpenClawOpenClaw
Benign
high confidence
✓
Purpose & Capability
The name/description match the SKILL.md: it designs semantic/flow tests and asks for evidence and rubrics. There are no unexpected env vars, binaries, or installs required that would be unrelated to test design.
ℹ
Instruction Scope
The instructions are scoped to designing tests: splitting deterministic vs semantic checks, specifying evidence to collect, and defining rubrics. They do ask the executor to capture evidence (URLs, page titles, extracted items), which is appropriate for the stated purpose. Note: that evidence could include sensitive or private content if the agent runs against authenticated/private targets — the skill itself does not instruct reading local files or credentials.
✓
Install Mechanism
No install spec and no code files — instruction-only, so nothing is written to disk or downloaded by the skill itself.
✓
Credentials
The skill declares no required environment variables, credentials, or config paths. There is no disproportionate request for secrets or external access in the manifest or instructions.
✓
Persistence & Privilege
always is false and the skill is user-invocable; it does not request permanent presence or modifications to other skills or global agent settings in its instructions.
Assessment
This skill appears coherent and safe as an instruction-only test designer, but keep in mind: the tests it designs will capture and log evidence (URLs, page content, extracted items), which can include sensitive or private data if run against authenticated services or user content. Only run these tests against public or authorized targets, avoid supplying unrelated credentials, and review collected evidence before storing or sharing it. If you plan to have the agent run tests autonomously against production systems, add safeguards (rate limits, access controls, and human review for 'needs_review' cases).

Like a lobster shell, security has layers — review code before you run it.

latestvk97b82phpm2hgef3vzhnc54e1s83hddz

License

MIT-0
Free to use, modify, and redistribute. No attribution required.
Termshttps://spdx.org/licenses/MIT-0.html

Comments