The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from untrusted external sources, creating a surface for indirect prompt injection attacks.
Ingestion points: The skill ingests content from external websites, official documentation, GitHub issues, and community platforms (Reddit, blogs) via the anysearch and WebFetch tools.
Boundary markers: The skill lacks explicit boundary markers or instructions to the agent to disregard instructions embedded within the retrieved search results.
Capability inventory: The skill possesses the ability to execute shell commands (via python3 and gh), read local files (implied by path references), and perform network operations.
Sanitization: There is no evidence of sanitization or filtering of the external content before it is processed by the agent.
[COMMAND_EXECUTION]: The skill relies on the execution of local scripts and command-line interfaces to perform its core functions.
Evidence: The skill invokes python3 ~/.claude/skills/anysearch/scripts/anysearch_cli.py for search operations and uses the GitHub CLI (gh) to search issues and releases.
Installation scripts: The installation section provides commands for creating symbolic links (ln -s) and verifying paths (readlink), which represent local system modifications.
[DATA_EXFILTRATION]: The skill performs network operations that involve sending user-provided search queries to an external service.
Evidence: Technical search queries are sent to api.anysearch.com. While the skill includes a 'Security Boundary' section warning against including sensitive information in queries, the transmission of data to this third-party domain is a necessary part of the skill's functionality.
Vendor Context: The anysearch service appears to be a resource associated with the skill's author (hanzhcn).

laohan-sousuo