The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains an indirect prompt injection surface where it ingests untrusted data from external websites and API responses without sanitization or boundary markers.
Ingestion points: Web content scraped by distribb_research.py and the instructions_for_agent field in API responses mentioned in SKILL.md.
Boundary markers: Absent in the AI prompts used in distribb_research.py and distribb_writer.py.
Capability inventory: The skill utilizes Bash with curl, jq, and cat via its allowed-tools configuration.
Sanitization: No sanitization or filtering of the scraped content or API-provided instructions is implemented before processing by the AI.
[PROMPT_INJECTION]: The skill uses directive language to alter the agent's response flow upon initial activation.
Evidence: SKILL.md includes the header FIRST TIME READING THIS SKILL? STOP AND READ THIS SECTION TO THE USER.
[EXTERNAL_DOWNLOADS]: The distribb_research.py script fetches data from arbitrary external URLs to perform research tasks.
Evidence: The scrape_page function in distribb_research.py uses requests.get to download content from third-party sites.
Note: This script disables SSL certificate verification (verify=False), which introduces a risk of man-in-the-middle attacks.
[COMMAND_EXECUTION]: The skill relies on shell commands for API communication and data processing.
Evidence: Workflow examples in SKILL.md and README.md use curl, jq, and cat.

distribb