The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill uses a custom script to fetch content from conference websites (Call for Papers URLs).
Evidence: scripts/fetch_cfp.py performs GET requests using urllib.request.urlopen to URLs provided by the user.
Mitigation: The script implements rate limiting (1-second delay between requests), a response size limit (8MB), and requires a CONTACT_EMAIL for a polite User-Agent string. All network activity is associated with the primary, transparent purpose of the skill.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted text from external websites, creating a surface for indirect prompt injection.
Ingestion points: scripts/fetch_cfp.py reads data from arbitrary URLs.
Boundary markers: The skill relies on the agent extracting facts into a specific YAML structure rather than following instructions within the text.
Capability inventory: The skill can write to the local filesystem and propose gh CLI commands.
Sanitization: fetch_cfp.py uses HTMLParser to strip HTML tags and scripts, reducing the data to plain text before processing.
[COMMAND_EXECUTION]: The skill instructs the user to run local Python scripts and suggests a GitHub PR creation command.
Evidence: SKILL.md provides commands for scripts/init_profile.py, scripts/fetch_cfp.py, and scripts/validate_profile.py.
Mitigation: All scripts are part of the skill package and are executed locally. The final gh pr create command is presented to the user for explicit approval and manual execution, ensuring no automated external actions occur.

add-venue-profile