The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute a custom CLI tool (apm) and a Python script (scripts/validate_apm_package.py) to deploy packages using the Bash tool.
[PROMPT_INJECTION]: Indirect Prompt Injection Surface found in file SKILL.md:
Ingestion points: Files are ingested from external packages via the apm install command.
Boundary markers: A validation script is used, but there are no explicit delimiters or instructions to ignore embedded commands in the materialized content.
Capability inventory: The skill allows use of Bash, Read, and Glob tools and executes shell and Python commands across multiple script contexts.
Sanitization: Relies on scripts/validate_apm_package.py for package validation, but does not specify content sanitization or filtering of the package instructions.
[COMMAND_EXECUTION]: The skill permits the agent to write files to sensitive directories, most notably .github/, which houses CI/CD workflow configurations. Modification of these files can lead to persistence or unauthorized code execution in automated environments.
[COMMAND_EXECUTION]: The skill facilitates the dynamic loading of artifacts into agent runtime directories (e.g., .agents/, .claude/). This behavior allows the introduction of new skills or configurations into the agent's operational context at runtime, which is a form of dynamic execution.

install-apm-package