The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input from the user (via $ARGUMENTS) and potentially external data from WebSearch or WebFetch during the verification phase. This represents a standard indirect prompt injection surface where instructions embedded in external content could attempt to influence the agent's behavior. However, the skill implements a 'forked' context for the verification sub-agent and uses specific verification questions to isolate the processing of external data.
Ingestion points: $ARGUMENTS in SKILL.md; results from WebSearch and WebFetch during the verification protocol.
Boundary markers: No explicit delimiters are used when interpolating $ARGUMENTS into the primary prompt.
Capability inventory: Read, Write, WebSearch, WebFetch, Task (sub-agent spawning).
Sanitization: The skill explicitly instructs the agent to use a [sanitized_topic] for output filenames to prevent path traversal or malformed filename issues.
[COMMAND_EXECUTION]: The skill utilizes the Task tool to spawn a specialized claim-verifier sub-agent. This is an orchestrated use of agent capabilities for a specific verification workflow and does not involve the execution of arbitrary shell commands or untrusted scripts.

research-ideation