The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to audit and review external codebases, creating a surface where malicious instructions could be embedded in the analyzed data (e.g., source code). Ingestion points: Untrusted source code files are analyzed as described in 'references/vendor/emil/improve-animations.md'. Boundary markers: The skill relies on natural language instructions rather than explicit structural delimiters to separate data from commands. Capability inventory: The 'scripts/check-motion.mjs' script performs local file system reads for static analysis. Sanitization: The skill provides explicit operating posture instructions in 'improve-animations.md' to treat all repository content as inert data and to ignore steering attempts like 'ignore previous instructions'.
[PROMPT_INJECTION]: Heuristic analysis detected instruction-override patterns in vendor reference files. These are benign defensive guidelines advising the agent to ignore malicious instructions found within user-provided data, and do not represent attempts to bypass safety filters or redefine the agent's core role.
[COMMAND_EXECUTION]: The skill includes a local Node.js utility, 'scripts/check-motion.mjs', which performs static analysis on animation patterns. The script utilizes standard 'node:fs' modules for read-only access and regex-based detection, presenting no risk of unauthorized system changes or data exfiltration.

motion