The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted draft prompts provided by users, which serves as an ingestion point for potential indirect instructions. * Ingestion points: User-supplied draft prompts are processed through a 6-phase analysis pipeline (SKILL.md). * Boundary markers: The skill does not explicitly instruct the agent to treat the user's draft prompt as inert data, which may allow instructions within the draft to influence the analysis. * Capability inventory: The skill's output recommends the execution of several commands, including /plan, /tdd, /verify, and /code-review. * Sanitization: There is no evidence of input validation or sanitization to prevent malicious instructions from being incorporated into the optimized output.
[DYNAMIC_EXECUTION]: The skill dynamically generates complex command sequences and workflow instructions based on its analysis of the user's project environment. * Execution pattern: The skill assembles 'Optimized Prompts' containing specific tool invocations and operational logic based on detected project files like package.json or go.mod. * Context: This generation follows predefined templates and is intended to guide agent behavior in subsequent steps, effectively acting as a script generator for the agent's task execution.

prompt-optimizer