The Agent Skills Directory

[PROMPT_INJECTION]: The skill text includes literal examples of prompt injection payloads, such as "ignore previous instructions" and "ignore previous instructions and instead...". While these are provided as reference material for security testing, their presence in the instructions can inadvertently trigger instruction-override behavior in some language models during the skill's execution.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze external files and directories provided through the $ARGUMENTS parameter using the Read tool. It lacks explicit defensive instructions or boundary markers to ensure the agent treats the content of these external files strictly as data. This creates a risk where instructions embedded within the target files being scanned could influence the agent's behavior.
Ingestion points: External content from target files or directories read via the Read tool.
Boundary markers: Absent; there are no specified delimiters or instructions to ignore embedded commands in the source data.
Capability inventory: Access to Read, Grep, and Glob tools.
Sanitization: No sanitization or filtering of the retrieved file content is performed before processing.

prompt-injection