Proprioception

Self-spatial awareness for AI agents. Gives your bot a real-time sixth sense of where it is relative to the user's goal, its own confidence boundaries, conve...

What to check before installing: - Review the omitted file (text-utils.js) and any other truncated/omitted files. Those utilities implement tokenization, similarity, and pattern counting — if they also include network calls, file writes, or telemetry hooks, that would change the assessment. - Double-check there are no hidden network calls or use of process.env in the full codebase. The reference doc mentions a "Data Flywheel" (collecting aggregate telemetry). That is not present in the visible scripts, but if telemetry/upload code exists elsewhere it would be unexpected for a skill that claims “zero external API calls.” - Confirm how your agent runtime executes skill code: is execution sandboxed, are filesystem and network accesses restricted? Running this skill in a restricted test environment first is prudent. - If you plan to use this in production, ask the author for provenance (who published it) and for an explicit statement about data handling (no telemetry, no uploads, no logging outside the agent). The skill is coherent with its stated purpose, but the missing file and the 'data flywheel' language are reasons to verify there's no undeclared telemetry or side effects.

Type: OpenClaw Skill Name: proprioception Version: 1.0.0 The skill's core logic, implemented in the Node.js scripts, is benign, performing local text analysis without external network calls or sensitive file access. However, the `SKILL.md` file contains an instruction for the OpenClaw agent to execute a shell command (`node "$(dirname "$SKILL_PATH")/scripts/proprioception-engine.js" --root-intent "$ROOT_INTENT" --current-response "$CURRENT_RESPONSE" ...`). This command directly interpolates agent-provided variables (`$ROOT_INTENT`, `$CURRENT_RESPONSE`, `$PRIOR_SIGNALS_JSON`) into the shell. If the OpenClaw agent platform does not properly sanitize or escape these variables before execution, a malicious user could craft inputs to achieve shell injection, leading to Remote Code Execution. This constitutes a significant vulnerability, classifying the skill as suspicious.