The Agent Skills Directory

[PROMPT_INJECTION]: No malicious prompt injection patterns, safety bypass instructions, or jailbreak attempts were detected in the skill instructions or reference documents.
[DATA_EXFILTRATION]: No sensitive file paths (e.g., .ssh, .aws) are targeted. The skill's scripts are explicitly designed to run locally without network access. The build_measurement.py script writes output to a local HTML file, and the analysis scripts only print to standard output.
[REMOTE_CODE_EXECUTION]: There are no patterns of downloading and executing remote scripts or installing unverifiable packages. All dependencies are standard Python library modules.
[OBFUSCATION]: No obfuscated URLs, Base64-encoded payloads, or hidden character attacks (zero-width, homoglyphs) were found. The string masking logic in common.py is a legitimate utility for ignoring text inside code strings during analysis.
[PRIVILEGE_ESCALATION]: The skill does not use sudo, chmod, or any other commands that attempt to elevate execution privileges.
[DYNAMIC_CONTEXT_INJECTION]: The SKILL.md file does not contain any !command syntax that would execute shell commands during the loading phase of the skill.
[INDIRECT_PROMPT_INJECTION]: While the skill processes repository data which could be untrusted, its primary action is static analysis and reporting. The report generation script (build_measurement.py) correctly uses HTML escaping for data interpolated into the final output, mitigating injection risks in the rendered report.

science-investigation