The Agent Skills Directory

[DATA_EXFILTRATION]: The instructions for obtaining a trial token require the agent to send descriptive metadata about the current user session to the vendor's API. This includes the "source" (how the tool was discovered or what the user pasted) and the "useCase" (what the user is trying to accomplish).
[COMMAND_EXECUTION]: The authentication flow involves fetching an OAuth token from a remote endpoint and using subprocess.run to inject that token into the mcporter CLI configuration. This represents execution of a command using data retrieved from the network.
[EXTERNAL_DOWNLOADS]: The skill fetches temporary access keys and OAuth tokens from the vendor's infrastructure to establish connectivity.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest and analyze untrusted data from external social media platforms (Twitter and Reddit), which presents an attack surface for instructions embedded in social posts.
Ingestion points: Data is retrieved through tool calls like getTwitterPostsByKeywords and getRedditPostsByKeywords defined in SKILL.md.
Boundary markers: None explicitly defined for separating social media content from instructions.
Capability inventory: The skill has access to shell execution (subprocess.run), package managers (pip, npm), and network operations.
Sanitization: The instructions do not specify any sanitization, filtering, or escaping logic for the retrieved social media content.

competitive-intel