The Agent Skills Directory

[PROMPT_INJECTION]: The skill fetches untrusted data from external sources (Reddit and Hacker News) and instructs the agent to process it. This creates a surface for indirect prompt injection attacks where malicious content on those platforms could influence agent behavior. Ingestion points: scripts/fetch.js fetches data from hn.algolia.com and www.reddit.com. Boundary markers: None present in SKILL.md. Capability inventory: Agent executes bash scripts/evidence-log.sh using fetched data. Sanitization: None performed on the fetched content to filter instructions.
[COMMAND_EXECUTION]: The instruction to run bash scripts/evidence-log.sh <source> '<quote>' '<url>' relies on single-quoting values from external sources. If the fetched data contains single quotes, it could lead to command argument injection.
[COMMAND_EXECUTION]: The scripts/fetch.js script includes a --fixture argument that uses fs.readFileSync to read local files. This capability allows for local file access if the agent is induced to use this parameter with a sensitive path.
[EXTERNAL_DOWNLOADS]: The skill retrieves data from hn.algolia.com and www.reddit.com to identify trends. These are established services and the network calls are consistent with the skill's purpose.

trends