pm-resource
Warn
Audited by Gen Agent Trust Hub on Aug 3, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONREMOTE_CODE_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The Preamble section instructs the agent to execute a bash script
check-update.shlocated in the skill's own directory to check for updates upon initialization. - [COMMAND_EXECUTION]: The skill logic (Subagent Parallel Analysis) directs the agent to locate an external script named
anysearch_cli.pyby scanning several hidden platform directories in the user's home folder, including~/.claude/,~/.cursor/, and~/.anysearch/. Once found, it executes the script viapython3with dynamic arguments. This pattern of searching and executing code from arbitrary hidden system paths increases the attack surface. - [REMOTE_CODE_EXECUTION]: While the script execution is local, the instructions to search and run an unverified Python utility from system-wide hidden directories constitute dynamic code execution from computed paths.
- [DATA_EXPOSURE]: The skill is designed to read and process potentially sensitive business documents (e.g.,
产品组合战略.md) to perform resource planning, which is its primary stated purpose. - [EXTERNAL_DOWNLOADS]: The skill utilizes web search tools (
mcp__exa__web_search_exa,WebSearch) to ingest industry benchmark data, which introduces a surface for indirect prompt injection.
Audit Metadata