The Agent Skills Directory

[METADATA_POISONING]: The skill references non-existent and deceptive model versions, specifically "Claude Opus 4.6" and "gpt-5.3-codex". This misleading metadata may cause users to misjudge the skill's capabilities or potentially install unverified third-party tools masquerading as these official models.
[INDIRECT_PROMPT_INJECTION]: The skill demonstrates a high attack surface for indirect prompt injection as it processes untrusted document content across multiple sub-agents.
Ingestion points: In Step 2, the skill uses cat <〃ぁいるぱす> to read the content of local files.
Boundary markers: The skill lacks delimiters (such as XML tags or triple backticks) or instructions to the sub-agents to disregard embedded commands within the processed data.
Capability inventory: The skill has the capability to execute shell commands through the gemini and codex CLI tools.
Sanitization: No sanitization, escaping, or validation is performed on the file content before it is passed to the sub-agents.
[DYNAMIC_EXECUTION]: The skill utilizes shell command substitution $(cat <〃ぁいるぱす>) to dynamically build the command string for the codex tool. If the content of the file being checked contains specific shell metacharacters (e.g., mismatched quotes followed by a semicolon), it could result in arbitrary command execution on the user's system.
[COMMAND_EXECUTION]: The instructions rely on shell commands and pipes (cat, gemini, codex, echo, |) to process data and interact with external AI services.

quad-fact-check