Counter Evidence Hunter
v2.0.0LLM通用反证搜索技能。围绕当前主线判断，主动寻找反例、冲突证据、翻转条件和替代路径支撑，减少单线叙事偏差。在已有主线判断后、高风险结论输出前、风险分析前使用。触发条件：需要降低幻觉和单线偏差、需要补充替代叙事证据、高风险决策前的纠偏。
⭐ 0· 102·1 current·1 all-time
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 (counter‑evidence search) aligns with the SKILL.md and reference files: all materials describe generating queries, executing searches, classifying evidence, and producing flip conditions. No unrelated credentials, binaries, or installs are requested.
Instruction Scope
The runtime instructions explicitly require the agent to "execute searches" for each generated query and to assess source quality. This is coherent with the skill's purpose, but the SKILL.md does not declare or constrain which search/browsing tool or endpoints to use — so the actual network calls will depend on the agent's toolchain. The instructions do not ask the agent to read arbitrary local files or secrets.
Install Mechanism
No install spec and no code files — the skill is instruction-only, which minimizes on-disk risk. Nothing is downloaded or installed by the skill itself.
Credentials
The skill requires no environment variables, credentials, or config paths. The lack of requested secrets is proportionate to its described function (search + synthesis).
Persistence & Privilege
Flags are default (always:false, user-invocable:true, model invocation allowed). The skill does not request permanent presence or system-wide configuration changes; nothing suggests elevated privileges.
Assessment
This skill appears internally consistent and low-risk because it's purely instructional and asks the agent to search and analyze evidence. Before enabling it broadly, consider: (1) confirm which search/browsing tool the agent will use (and that tool's network/endpoint policies) so you know where queries and results travel; (2) avoid passing secrets or sensitive documents as inputs — the skill will instruct the agent to search and could surface or transmit input text to external search tools; (3) prefer user-invocable use (not always-enabled) so searches only run when you request them; (4) review example outputs and run a few dry tests with non-sensitive claims to validate the quality and scope of returned sources; (5) if you need auditability, enable logging/monitoring of the agent's external searches and outputs. Overall the design is coherent, but its effectiveness depends on the agent's search/browsing capabilities and operator controls.Like a lobster shell, security has layers — review code before you run it.
analysisvk97bwm9rxp7s5b0qpcaanmqmg983mgx5counter-evidencevk97bwm9rxp7s5b0qpcaanmqmg983mgx5evidencevk97bwm9rxp7s5b0qpcaanmqmg983mgx5fact-checkvk97bwm9rxp7s5b0qpcaanmqmg983mgx5groundingvk97bwm9rxp7s5b0qpcaanmqmg983mgx5latestvk97bwm9rxp7s5b0qpcaanmqmg983mgx5searchvk97bwm9rxp7s5b0qpcaanmqmg983mgx5signal-intelligencevk97bwm9rxp7s5b0qpcaanmqmg983mgx5
License
MIT-0
Free to use, modify, and redistribute. No attribution required.