The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides explicit instructions for the agent to execute shell commands using curl and bash to interact with an external search API at https://deepapi.co/v1/search/web. Additionally, it encourages the use of bash to explore local paths of cloned GitHub repositories.
[DATA_EXPOSURE_AND_EXFILTRATION]: The instructions direct the agent to load sensitive environment configuration data, including API keys, from the local file path ~/.deepapi/env using shell sourcing.
[EXTERNAL_DOWNLOADS]: The skill references a dependency on the pi-web-access package, which is not a standard, well-known library or a package from a trusted vendor.
[INDIRECT_PROMPT_INJECTION]: The skill implements tools that ingest and process a wide variety of untrusted data from external sources, which could contain malicious instructions designed to influence the agent's behavior.
Ingestion points: Untrusted content is fetched via web_search and fetch_content from arbitrary web pages, PDF documents, YouTube transcripts, and GitHub repositories.
Boundary markers: The skill lacks any defined delimiters or instructions to the agent to ignore or isolate potential instructions embedded in the external data.
Capability inventory: The agent possesses capabilities to execute shell commands (bash, curl), explore the local file system (read), and perform network operations.
Sanitization: There are no specified sanitization, validation, or filtering processes for the external content before it is processed by the agent.

pi-web-search