The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill uses the Node.js vm module via vm.runInNewContext in scripts/patterns/pattern-utils.js and scripts/patterns/ob-variant-pass.js to evaluate parts of the input JavaScript code. This is used to resolve obfuscated string tables and decoder functions. Although the tool employs a limited sandbox and execution timeouts, the vm module is not a secure boundary, and evaluating untrusted code from input files poses a potential risk of sandbox escape.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted JavaScript files provided by the user, which presents a vulnerability surface for indirect attacks where a malicious file could target the Babel parser or the vm execution environment.
Ingestion points: Untrusted content enters the system via the inputPath file argument in run-pipeline.js.
Boundary markers: There are no explicit boundary markers or instructions to ignore embedded malicious content within the processed files.
Capability inventory: The skill has capabilities to read/write local files and execute commands via spawnSync (to run its internal scripts) and vm.runInNewContext.
Sanitization: The skill relies on Babel's parsing and a restricted Node.js vm context with a 1000ms timeout for safety.
[METADATA_POISONING]: The SKILL.md file contains instructions referencing an absolute local file path (C:\Users\25198\.codex\skills\.system\skill-creator\scripts\quick_validate.py) intended for validation by the author. This indicates that the documentation includes environment-specific artifacts that were not sanitized for distribution.

ast-deobfuscation