The Agent Skills Directory

[PROMPT_INJECTION] (LOW): Indirect Prompt Injection risk due to external data ingestion.
Ingestion points: The skill fetches message content, thread history, and search results from Slack via several API endpoints (conversations.history, conversations.replies, search.messages) as seen in SKILL.md.
Boundary markers: Absent. The skill does not implement delimiters or provide instructions to the agent to treat Slack content as untrusted data.
Capability inventory: The skill utilizes shell execution (curl, sed, jq) and file system output via mkoutput.
Sanitization: Absent. There is no filtering or sanitization of message text before it enters the agent's context. Malicious Slack messages could contain instructions aimed at manipulating the agent's behavior.
[DATA_EXFILTRATION] (LOW): Network communication to a non-whitelisted domain.
The skill uses curl to send data to and receive data from slack.com. While this is the intended purpose of the skill, the domain is not included in the predefined trusted whitelist.
[COMMAND_EXECUTION] (SAFE): Use of shell commands for API interaction.
The skill uses curl, sed, and jq for parsing URLs and interacting with APIs. Variables derived from user input (such as queries and URLs) are handled with double quotes or safely encoded via jq -sRr @uri, minimizing the risk of direct command injection.

slack