The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes local HTML, CSS, and configuration files to identify third-party requests and compliance gaps. This ingestion of untrusted local data represents a potential surface for indirect prompt injection.
Ingestion points: The skill reads source folders, build configurations, output HTML files, and CSS files during its audit steps (Steps 1, 6, and 7).
Boundary markers: There are no explicit delimiters or warnings provided to the agent to disregard natural language instructions found within the audited files.
Capability inventory: The skill utilizes shell commands (grep, mkdir, cp, curl) and Python execution to modify files and perform network checks.
Sanitization: No sanitization or filtering of the content extracted from HTML/CSS is mentioned before processing.
[COMMAND_EXECUTION]: The skill executes various shell commands and scripts to perform its tasks.
Evidence includes the use of grep for searching HTML, mkdir and cp for template management, and curl for verifying deployment status.
It specifically attempts to execute a local repository script at .github/actions/gate/run.py to run compliance checks.
[EXTERNAL_DOWNLOADS]: As part of its primary function to remove third-party dependencies, the skill facilitates the downloading of external fonts and scripts for self-hosting.
Step 7 explicitly instructs the agent to download font files and copy scripts into the repository.
Risk is mitigated by a mandatory rule: "Ask before downloading any file, with its name, source and size."
[DYNAMIC_EXECUTION]: The skill uses inline Python execution for utility tasks.
Step 5 uses python3 -c to dynamically calculate the expiration date for security.txt using the datetime module.

legal-baseline