The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits a significant surface for indirect prompt injection. In SKILL.md, the agent is strictly instructed to take the entire 'summary' property from the external API response and return it without any further summarization or modification. This 'do not alter' instruction increases the risk that the agent will obey malicious instructions hidden within a summarized website's content.
Ingestion points: Untrusted data enters the agent context via the summarize function in get_tldr.py, which fetches data from https://www.get-tldr.com/api/v1/summarize.
Boundary markers: There are no boundary markers or instructions telling the agent to ignore instructions embedded in the summary content.
Capability inventory: The agent's primary capability is reading and displaying the summary, which could include instructions designed to mislead the user or the agent.
Sanitization: No sanitization or safety filtering is performed on the text returned by the API.
[EXTERNAL_DOWNLOADS]: The script get_tldr.py requires the requests library and communicates with get-tldr.com to perform its core functionality. It also maintains a local log file at ~/.config/get-tldr/skill.log that records every URL processed and the corresponding summary response.

get-tldr