The Agent Skills Directory

[PROMPT_INJECTION]: The skill defines a command template that incorporates untrusted user input into a shell execution environment, creating a surface for indirect prompt injection or command injection. \n- Ingestion points: User-defined goal strings are interpolated into the <user goal> placeholder within the command no-mistakes axi run --skip=pr,ci --intent "<user goal>" in SKILL.md. \n- Boundary markers: Although the placeholder is wrapped in double quotes, there are no instructions to sanitize or escape shell-sensitive characters (such as backticks, semicolons, or command substitutions) that may be present in the user input. \n- Capability inventory: The skill assumes the agent has the capability to execute the no-mistakes CLI tool within the host shell environment. \n- Sanitization: No validation or sanitization logic is provided to ensure the user goal does not contain malicious sequences.

ptlam-skills-git