The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on the execution of shell commands to perform its functions, including the gemini CLI for analysis and standard Unix utilities like ps, pkill, kill, and lsof for process monitoring and termination.
[METADATA_POISONING]: The skill documentation makes highly suspicious claims regarding 'Gemini 3 Pro' models and 'January 2025' knowledge cutoffs. It cites performance metrics (e.g., 76.2% on SWE-bench) that appear fabricated or speculative compared to verified industry benchmarks for currently available models. This misinformation may lead users to over-rely on the tool's automated capabilities.
[INDIRECT_PROMPT_INJECTION]: The skill creates a vulnerability surface by processing large amounts of external data (entire codebases) while recommending the use of --approval-mode yolo. This configuration allows the tool to perform automated actions, such as code edits, without manual review, which could be exploited by malicious instructions embedded within the files being analyzed.
Ingestion points: Full codebase directories and large documentation sets (>200k tokens) processed via the --include-directories flag, as described in SKILL.md.
Boundary markers: No delimiters or specific 'ignore embedded instructions' warnings are specified in the provided command templates or instructions to the agent.
Capability inventory: The skill facilitates file system interaction and automated code modification through the gemini CLI tools when used in 'yolo' mode.
Sanitization: The skill does not mention any sanitization, validation, or filtering of the external file content before it is processed by the model.

gemini