The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's core function involves processing user-supplied text that is structured as agent instructions, which presents a surface for indirect prompt injection.
Ingestion points: Untrusted text provided by the user for prompt optimization, as described in the task and system modes in SKILL.md.
Boundary markers: The instructions include explicit directives to treat input as inert data and to specifically ignore embedded commands like "ignore previous instructions."
Capability inventory: The skill utilizes Read, Grep, Glob, Write, and Edit tools for codebase grounding and artifact generation.
Sanitization: The skill implements logic to proactively identify and replace hardcoded credentials with environment references and to flag injection attempts in the output diagnosis.
[PROMPT_INJECTION]: The SKILL.md file contains text patterns commonly associated with prompt injection, such as "ignore previous instructions." However, these are documented as safety rules intended to prevent the agent from being manipulated by untrusted input, rather than attempts to subvert the agent's own safety constraints.

promptkit