The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes subprocess.run to execute git commands (e.g., git rev-parse) for identifying the stable repository root name. This usage is restricted to metadata resolution and does not involve unsafe shell interpolation.
[REMOTE_CODE_EXECUTION]: The skill employs dynamic loading via importlib.util to access configuration data from associated build skills (e.g., build-maven, build-npm). This is a modular architecture pattern used to coordinate between trusted internal skills and is restricted to local bundle paths.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it ingests untrusted project content (READMEs, doc comments, source code) to generate architectural summaries and 'responsibility' descriptions. This is a low-risk finding inherent to the skill's primary function of documentation analysis.
[DATA_EXFILTRATION]: The skill crawls the project filesystem to build a comprehensive file inventory and internal dependency graph. All processed data is stored locally in the .plan/project-architecture/ directory for subsequent project tasks; no network operations or external exfiltration vectors were identified.

manage-architecture