The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill uses curl to download data samples and files from arbitrary external URLs identified during the source discovery process.
[COMMAND_EXECUTION]: The instructions guide the agent to use shell utilities including curl, file, head, wc, and python3 to probe the integrity and structure of potential data sources.
[PROMPT_INJECTION]: The skill exhibits a vulnerability to indirect prompt injection because it ingests untrusted data from external URLs and presents it to the agent without sufficient isolation.
Ingestion points: External content is downloaded to sample.out and s.csv from URLs provided in lead briefs or the source playbook.
Boundary markers: There are no instructions to wrap the downloaded content in delimiters or to explicitly warn the agent to ignore any embedded instructions.
Capability inventory: The environment allows for shell execution, file system interaction, and network access.
Sanitization: The skill does not perform any validation, filtering, or escaping of the downloaded data before the agent reads it.

find-sources