The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external websites, creating a potential surface for indirect prompt injection attacks. 1. Ingestion points: The scripts/fetch_page.py script retrieves content from public URLs. 2. Boundary markers: The skill contains explicit instructions in SKILL.md and references/skill-contract.md to treat external data as untrusted evidence and to ignore any instructions embedded within that data. 3. Capability inventory: The skill utilizes network access for data retrieval via scripts/fetch_page.py and provides scripts for local file manipulation and reporting. 4. Sanitization: The instructions emphasize evidence labeling and prohibit the silent upgrade of uncertainty.
[EXTERNAL_DOWNLOADS]: The scripts/generate_pdf_report.py script suggests the installation of the reportlab library if PDF generation is required. This is an optional dependency on a well-known Python library and does not pose a security risk.
[DATA_EXFILTRATION]: The scripts/brand_scanner.py script allows for reading and scanning arbitrary local files provided as arguments, which could be misused to access sensitive information if the agent is instructed to scan paths outside the intended project scope.
[DYNAMIC_EXECUTION]: The scripts/citability_scorer.py script uses __import__ to dynamically load the sys module for reading from standard input. This is a common pattern for CLI tools and is considered safe in this context as it targets a standard library module.

geo-report