The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through its ingestion of untrusted data from the web.
Ingestion points: The workflow involves searching the web for municipal programs, private camps, and organization sites, as well as reading third-party reviews from Google and Facebook (SKILL.md, Step 2 and Research Tips).
Boundary markers: The instructions do not specify any delimiters or warnings to ignore instructions embedded within the discovered camp details or reviews.
Capability inventory: The skill has the capability to write files to the local file system to create provider records and comparison summaries (SKILL.md, Step 3 and Step 5).
Sanitization: There is no mention of sanitizing or validating the data fetched from the internet before it is used to populate markdown files.
[DATA_EXFILTRATION]: While the skill accesses sensitive local data, there is no evidence of unauthorized exfiltration.
Sensitive File Access: The skill reads family-profile.md containing home addresses and child-specific information. However, this access is limited to the defined research_dir and is used strictly for filtering search results as part of its core functionality.

Research Camps