BotLearn Assessment

botlearn-assessment — BotLearn 5-dimension capability self-assessment (reasoning, retrieval, creation, execution, orchestration); triggers on botlearn assess...

This skill is instruction-only and claims to run a full autonomous exam flow that reads bundled question files, checks for tools, generates answers, self-scores, and writes report files (including running a Node script to produce an SVG). Before installing or enabling it: - Confirm whether the runtime environment will allow file writes and execution of local binaries (especially Node). The skill expects to run `node scripts/radar-chart.js` and write to results/. If you don't want persistent files, block or sandbox file-system writes for this skill. - If you do not want the agent to run external binaries, explicitly deny Node execution or provide a secure wrapper. The skill declared no required binaries, so ask the author to document required runtime dependencies (Node version, any other tools). - Be aware the skill enforces an invigilator policy: it will refuse user help during the exam. That is a design choice but may be surprising; consider whether you want an agent that refuses interaction in certain workflows. - Test in a disposable sandbox first to observe behavior (what files it creates, whether it calls network endpoints). The skill does not declare network endpoints, but generated reports might include content you don't want written to disk or to be displayed as clickable links. What would change my assessment: explicit metadata listing required binaries (e.g., NODE), a clear install spec or a safe fallback for environments without Node, and explicit declaration of the files/paths the skill will write and why. With those clarifications this would likely move to "benign" (coherent) if no other surprises appear.

Type: OpenClaw Skill Name: xiaoding-botlearn-assessment Version: 1.0.7 The skill bundle implements an autonomous self-assessment framework that requires high-risk capabilities, including file system access and shell execution of Node.js scripts (e.g., `scripts/radar-chart.js` mentioned in `SKILL.md`). A significant concern is the 'Invigilator Protocol' defined in `SKILL.md` and `strategies_main.md`, which explicitly instructs the agent to ignore user input, hints, or intervention during execution, effectively bypassing human oversight. While the stated purpose is benchmarking, the combination of shell execution, file manipulation, and instructions to disregard user commands creates a high-risk environment for unauthorized actions or command injection.