devil-advocate
v1.0.2通过正反方辩论和自我反驳，深入分析问题，提升回答的逻辑严密性和全面性，确保输出质量更高。
⭐ 0· 141·0 current·0 all-time
bySlava Chan@uynewnas
MIT-0
Download zip
LicenseMIT-0 · Free to use, modify, and redistribute. No attribution required.
Security Scan
OpenClaw
Benign
high confidencePurpose & Capability
Name and description promise an internal 'devil's advocate' debating process. The skill is instruction-only, requires no binaries, env vars, or config paths — all proportional and expected for a purely cognitive-enhancement skill.
Instruction Scope
SKILL.md and CLAUDE.md mandate a 3-stage internal process (Pro, Con, Integrate) and label the internal thinking as 'Must Execute'. The instructions do not ask to read files, access secrets, or call external endpoints, but the explicit 'Must Execute' internal chain-of-thought could conflict with platform policies or runtime safeguards that forbid exposing model internal reasoning. Verify the runtime preserves internalness (doesn't log or output chain-of-thought) and that the agent's responses reflect the process without revealing private internal deliberation.
Install Mechanism
No install spec; INSTALL.md only gives manual copy/uninstall steps and a local dashboard URL. No downloads, no extracted archives, and no scripts that would run during installation — low risk.
Credentials
The skill declares no required environment variables, credentials, or config paths. The instructions do not reference any secrets or unrelated environment data — requested access is minimal and proportional to the purpose.
Persistence & Privilege
Flags show always:false and default model invocation allowed (normal). The skill does not request persistent system-wide changes, nor does it instruct modifying other skills or agent settings. Manual installation writes files into the user's skills directory, which is expected for an OpenClaw skill.
Assessment
This skill appears coherent and low-risk: it only contains instructions for internal self-reflection and asks for no secrets or installs. Before installing: 1) Inspect the files yourself (they are present in the package) since the source/homepage is unknown; 2) Test with non-sensitive queries to confirm the agent does not print or log the internal 'Pro/Con' deliberations (the skill says the thinking should be internal); 3) If you operate in a sensitive environment, ensure agent runtime enforces that internal chain-of-thought is not exposed in logs or to external telemetry; 4) Keep the provided uninstall steps handy and prefer installing only from authors/sources you trust.Like a lobster shell, security has layers — review code before you run it.
latestvk9778j11t3fxm7pvw902bpkh4983d0bs
License
MIT-0
Free to use, modify, and redistribute. No attribution required.