The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local system commands through the Bash tool and a provided shell script (scripts/probe_environments.sh). It specifically runs gemini --version, cursor --version, and gh copilot explain "test" to verify tool availability and authentication status.
[PROMPT_INJECTION]: The skill establishes a surface for indirect prompt injection by capturing output from external CLI tools and persisting it in context/memory/environment.md, which serves as a source of truth for downstream skills like os-architect and os-evolution-planner.
Ingestion points: Standard output and error streams from the gemini, gh, and cursor commands as defined in the Probe Commands table in SKILL.md and the check_cmd function in scripts/probe_environments.sh.
Boundary markers: Absent. The captured output is written directly into Markdown tables and strategy sections in context/memory/environment.md without protective delimiters or instructions to downstream agents to ignore embedded instructions within these values.
Capability inventory: The skill possesses Bash, Read, and Write capabilities; it triggers subprocesses to evaluate the environment which could be manipulated if an attacker-controlled binary were placed in the PATH matching the expected tool names.
Sanitization: The skill uses head -n 3 or head -1 to limit the volume of data captured, which restricts the payload size but does not perform any content validation, escaping, or filtering of the resulting strings before they are incorporated into the prompt context for other skills.

os-environment-probe