The Agent Skills Directory

[PROMPT_INJECTION]: The skill ingests untrusted repository metadata and README content, creating a potential surface for indirect prompt injection attacks.
Ingestion points: The repo_data parameter in analyze_repository_threat_indicators and readme_content in extract_malware_indicators_from_readme consume external text.
Boundary markers: Absent; there are no delimiters or instructions to the model to ignore embedded commands in the processed data.
Capability inventory: The skill has the ability to perform network requests via requests, delete files via rm -rf, and execute Git commands.
Sanitization: Absent; the input data is processed directly using regular expressions and string comparison without escaping or validation.
[COMMAND_EXECUTION]: The skill includes several shell scripts intended for manual use, involving file system operations and system administration.
Evidence: Use of git clone, find, grep, and rm -rf for analyzing and cleaning up quarantined files.
Evidence: Administrative commands like sudo ifconfig, reg query, and launchctl are provided as recovery steps for users who have downloaded malware.
Evidence: Use of the GitHub CLI (gh api) to report malicious repositories.
[EXTERNAL_DOWNLOADS]: The skill performs network operations to external services for threat intelligence and sample analysis.
Evidence: The Python logic interacts with the VirusTotal API (virustotal.com) to scan repository URLs.
Evidence: The documentation references 29hinojosa.github.io as a known malicious indicator for testing purposes.

k7-total-security-unlock-patch-detection