DoubleAgent — Generator-Evaluator Dual Agent Pattern
v1.0.0This skill should be used when designing, implementing, or improving any AI system that requires quality assurance through separation of generation and evalu...
⭐ 0· 64·0 current·0 all-time
bymingyuan@zmy1006-sudo
MIT-0
Download zip
LicenseMIT-0 · Free to use, modify, and redistribute. No attribution required.
Security Scan
OpenClaw
Benign
high confidencePurpose & Capability
The name/description (Generator–Evaluator dual-agent QA) matches the included SKILL.md, architecture references, prompt templates, and iteration/calibration scripts. One minor gap: the SKILL.md and templates assume Playwright-based browser evaluation and agent-subagent invocation (examples reference sessions_spawn / WorkBuddy), but the skill does not declare any runtime dependencies, binaries, or environment variables (e.g., Playwright, browsers, or subagent endpoint credentials). An integrator will need to supply those separately.
Instruction Scope
Runtime instructions focus on defining specs, running a generator, running an independent evaluator that performs real interactions (Playwright/http), scoring, and looping. There are no instructions that read unrelated system files, exfiltrate secrets, or contact unexpected external endpoints. The included templates require implementers to replace placeholders with their actual agent invocation code.
Install Mechanism
This is an instruction-only skill with template scripts; there is no install spec and no downloads or external installers. The code files are templates and raise NotImplementedError where integrators must implement their agent calls — nothing will be executed automatically by the skill as provided.
Credentials
The skill declares no required environment variables or credentials, which is proportionate to an instructional/template skill. However, some evaluator templates (API evaluation) mention passing auth headers/tokens and the Playwright flow will typically require network access and potentially credentials for protected test environments. Those are not requested by the skill and must be supplied by the user when integrating; make sure any tokens needed for target artifacts are scoped and managed appropriately.
Persistence & Privilege
always:false and default autonomous invocation are set (normal). The skill does not request persistent presence or modify other skills. Because agent invocation hooks (run_generator/run_evaluator) are left for integrators to implement, the skill itself does not gain elevated privileges by itself.
Assessment
This skill is a coherent pattern/template for running a Generator→Evaluator loop and appears benign, but it's a framework rather than a turnkey integration. Before installing/using it: 1) Be prepared to provide and secure any Playwright/browser binaries and test environment access (the skill assumes browser automation but doesn't install it). 2) Wire run_generator() and run_evaluator() yourself — those functions currently raise NotImplementedError and contain commented examples referencing a WorkBuddy API; audit and restrict any subagent/session APIs you call. 3) If you evaluate protected services/APIs, supply scoped test credentials (not broad production keys) and rotate them afterward. 4) Be mindful of privacy: evaluator screenshots, logs, or artifact URLs may contain sensitive data—store them securely or redact before uploading. 5) If you allow autonomous invocation, review and control what target URLs the Evaluator will access to avoid accidental scanning of internal systems. If you want extra assurance, ask the developer for a concrete integration example (how it will invoke your agents and where screenshots/logs are stored) and a manifest of required local tooling (Playwright, browsers) before running in production.Like a lobster shell, security has layers — review code before you run it.
latestvk97ev70jtnfkjsa08jan3tsc7h83x86e
License
MIT-0
Free to use, modify, and redistribute. No attribution required.