The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates an indirect prompt injection surface due to its core function of ingesting untrusted project data to generate downstream agent instructions.
Ingestion points: The skill ingests external specifications, designs, and proposals during the analysis phases described in SKILL.md and the multi-engine logic in reference/tri-engine-failure.md.
Boundary markers: The skill employs structured FMEA templates and requires that output LLM Fix Prompts be enclosed in fenced code blocks to delineate generated instructions from prose.
Capability inventory: The skill produces LLM Fix Prompt blocks containing actionable verbs (e.g., ADD-GUARDRAIL, ADD-MONITOR) that are consumed by downstream agents with code-writing or monitoring capabilities, such as Builder and Beacon.
Sanitization: There is no explicit evidence in the instructions of input validation or sanitization for the external data being analyzed.
[DYNAMIC_EXECUTION]: The skill generates dynamic instruction sets (LLM Fix Prompts) intended for execution by other agents in the production pipeline. This creates an instruction-chaining risk where potentially malicious content in the initial analyzed assets could be translated into high-impact operational commands for agents with higher privilege or broader system access.

omen