The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-generated content from external market APIs.
Ingestion points: SKILL.md retrieves market comments from https://gamma-api.polymarket.com/comments.
Boundary markers: Comments are formatted as markdown list items (e.g., - [user]: "[comment excerpt]"), but there are no explicit instructions to the agent to treat this content as untrusted or to ignore instructions embedded within the text.
Capability inventory: The skill uses curl and WebFetch for network requests and ./notify for outputting findings.
Sanitization: No sanitization, escaping, or filtering is applied to the comment body before inclusion in the final report.
[COMMAND_EXECUTION]: The skill includes shell templates that interpolate user-supplied input directly into command strings.
Evidence: The ${var} parameter is parsed into SINGLE, which is subsequently used in curl commands in SKILL.md (e.g., curl -s "...slug=$SLUG...").
Risk: If the agent executes these shell templates without its own sanitization layers, a maliciously crafted ${var} input could potentially lead to command injection.
[EXTERNAL_DOWNLOADS]: The skill fetches data from several external API endpoints for market monitoring.
Sources: It communicates with gamma-api.polymarket.com, clob.polymarket.com, and api.elections.kalshi.com.
Implementation: Data is retrieved using curl or the WebFetch tool and processed using jq. All endpoints are identified as public, non-authenticated services.

monitor-polymarket