The Agent Skills Directory

[COMMAND_EXECUTION]: The skill explicitly instructs the AI agent to invoke local shell commands, specifically codex exec and agy -p. These tools are described as having agentic capabilities, including file system modification (workspace-write).
[PROMPT_INJECTION]: The skill creates an indirect prompt injection surface by ingesting and processing untrusted external data.
Ingestion points: Processes content from issues, PR descriptions, diffs, and READMEs as described in SKILL.md.
Boundary markers: The skill recommends using stdin or temporary files with quoted here-doc delimiters to isolate untrusted input from the shell command.
Capability inventory: The skill leverages the Bash tool to execute external CLI commands and acknowledges the ability to edit files and run commands via the agy tool.
Sanitization: It provides clear instructions to avoid string interpolation and shell expansion, although compliance depends on the LLM following these behavioral guidelines correctly.
[COMMAND_EXECUTION]: Deceptive metadata and tool references are present. The skill mentions a 'Google Antigravity CLI' and models like 'Nano Banana Pro' and 'GPT-OSS'. 'Antigravity' is a well-known Python easter egg rather than an official Google CLI for Gemini. These fabricated or non-standard tool names may mislead users into installing or executing untrusted binaries that masquerade as legitimate services.

dispatch