The Agent Skills Directory

[METADATA_POISONING]: The skill's YAML frontmatter lists 'Alireza Rezvani' as the author, which contradicts the provided author context of 'iwanhe'. This inconsistency in metadata can be deceptive regarding the provenance and maintenance responsibility of the skill.\n- [INDIRECT_PROMPT_INJECTION]: The protocol establishes an inter-agent communication channel ([INVOKE:role|question]) that creates a vulnerability surface where malicious instructions could be passed from one agent to another. • Ingestion points: Data is ingested through the question parameter in [INVOKE] and the response body in [RESPONSE] tags as defined in SKILL.md. • Boundary markers: The skill uses structured tags as delimiters, which helps distinguish messages, but lacks explicit directives to sanitize or ignore instructions embedded within the inter-agent data payload. • Capability inventory: The protocol targets 'C-suite' agents who typically have access to critical business data and high-level analytical capabilities. • Sanitization: The protocol includes defensive architectural rules such as a maximum chain depth of 2, circular call detection, and isolation periods (Phase 2 analysis) to prevent the propagation of errors or malicious prompts.\n- [NO_CODE]: The skill consists entirely of Markdown-based protocol definitions and instructions; no executable scripts, binaries, or external dependencies are included in the package.

agent-protocol