The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill fetches and reads external specifications, issue descriptions, and comments to draft project tickets, which creates an attack surface where malicious instructions hidden in external data could influence the agent's behavior. Ingestion points: user-provided URLs and issue tracker content. Boundary markers: none defined to isolate fetched content. Capability inventory: reads local codebase, fetches remote URLs, writes local scratch files, and publishes to external trackers (GitHub/Linear). Sanitization: no explicit validation or filtering of ingested data.
[DATA_EXFILTRATION]: The skill performs network read operations to gather context from arbitrary user-supplied URLs. While used for ingestion, these requests to non-whitelisted domains represent a network interaction capability that can be used to reach remote servers.

to-tickets