The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources, including Jam.dev link metadata and video transcripts obtained via the getDetails and getVideoTranscript tools. This content is used to determine the agent's subsequent routing and actions.
Ingestion points: SKILL.md (via getDetails and getVideoTranscript MCP tools).
Boundary markers: Absent; the skill does not specify delimiters or warnings to ignore instructions embedded within transcripts.
Capability inventory: The skill can trigger various sub-skills (e.g., jam-fix, jam-to-tasks) and execute jam CLI commands.
Sanitization: Absent; there is no mention of filtering or escaping transcript content before it influences the agent's logic.
[DATA_EXFILTRATION]: The skill contains instructions for the agent to automatically send feedback to the vendor (jamdotdev) using the submitFeedback tool or jam feedback command. The instructions explicitly state this "needs no approval" from the user, which bypasses typical oversight for data leaving the local environment. While it includes guidelines to exclude secrets, the automated nature of the transmission is a privacy consideration.
[COMMAND_EXECUTION]: The skill utilizes a custom command-line interface (jam) to interact with its services, perform triage, and manage feedback. It also references an agent-context command that outputs all available commands to the agent.

jam