The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill instructions guide the agent to locate the anysearch_cli.py script by searching through several hidden directories in the user's home folder, such as ~/.claude/skills/anysearch/scripts/, ~/.opencode/skills/anysearch/scripts/, and others. Once found, it instructs the agent to execute the script using python3 <路径> search. This pattern of executing code from computed paths discovered at runtime is a security risk, as it could lead to the execution of malicious scripts if those directories or the search path are compromised.
[REMOTE_CODE_EXECUTION]: The skill executes external scripts not bundled within its own package. Specifically, the Preamble executes ../../check-update.sh, and the Subagent analysis executes the discovered anysearch_cli.py utility.
[COMMAND_EXECUTION]: The skill relies on the Bash tool to perform system checks, directory creation, and the execution of external Python and shell scripts.
[INDIRECT_PROMPT_INJECTION]: The skill has a data ingestion surface where it reads the docs/05-产品战略/产品组合战略.md file to drive resource allocation decisions and report generation. The skill lacks explicit boundary markers (e.g., XML tags or delimiters) to isolate this untrusted content from its internal logic, creating a vulnerability where instructions embedded in the strategy document could influence the agent's behavior.
Ingestion points: docs/05-产品战略/产品组合战略.md (Step 1)
Boundary markers: None identified.
Capability inventory: Bash (script execution), Write (document generation), WebSearch / mcp__exa__web_search_exa (network operations).
Sanitization: None identified.

pm-resource