The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes shell commands for workflow orchestration and data processing.
It instructs the agent to use standard utilities like mkdir for directory creation, tee and cat for file management, and jq for parsing JSON search results (found in 01-plan.md and 02-gather.md).
It relies extensively on the vendor's anycap CLI tool to perform search, crawl, and publishing operations across all reference files.
[EXTERNAL_DOWNLOADS]: The skill conducts network operations to fetch content and assets from remote sources.
Instructions in 02-gather.md describe using curl to download images from arbitrary remote URLs discovered during the research process.
The anycap search and anycap crawl commands are used to retrieve information from various external web domains.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it is designed to ingest and process untrusted data from the internet.
Ingestion points: Untrusted data enters the agent context through anycap search, anycap crawl, and anycap actions video-read as specified in 02-gather.md.
Boundary markers: There are no specific instructions or delimiters provided to help the agent distinguish between research content and operational instructions, increasing the risk of the agent following malicious commands embedded in search results or transcripts.
Capability inventory: The skill possesses broad capabilities, including writing to the local file system, performing network requests, generating images, and publishing reports to the web via anycap page deploy.
Sanitization: The instructions do not prescribe any sanitization or validation of external content before it is interpolated into reports or used for synthesis.

anycap-deepresearch