local-seek
Pass
Audited by Gen Agent Trust Hub on Nov 16, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONOBFUSCATION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses
subprocess.runto invoke system binaries includingrg(ripgrep),fd,git, andmdfindfrom withinscripts/seek.py. These operations are required to perform efficient local filesystem searches and to retrieve file metadata.\n- [INDIRECT_PROMPT_INJECTION]: The skill retrieves and processes local file content, which could potentially contain malicious instructions designed to influence the agent.\n - Ingestion points: Data is read from the local filesystem and returned to the agent context by
scripts/seek.py.\n - Boundary markers: Output results are prefixed with
path:line:metadata, providing basic structural context, although the search content is not wrapped in specialized security delimiters.\n - Capability inventory: The skill includes scripts that execute various system commands for searching and git operations.\n
- Sanitization: The
truncatefunction inscripts/seek.pylimits output snippets to 120 characters, which reduces the total volume of untrusted content ingested.\n- [OBFUSCATION]: Thescripts/seek_rank.pyfile contains Unicode escape sequences for Chinese keywords within its ranking dictionary. This is a documented functional implementation detail intended to prevent the search tool from matching its own source code during execution and is not considered malicious.
Audit Metadata