The Agent Skills Directory

[SAFE]: The skill provides a structured framework for inter-agent communication without introducing any executable code, shell commands, or network operations.
[PROMPT_INJECTION]: The instructions contain strict behavioral constraints (e.g., 'Loop Prevention', 'Isolation Rules') designed to prevent circular reasoning and groupthink in multi-agent environments. These represent healthy operational boundaries rather than malicious safety bypasses.
[INDIRECT_PROMPT_INJECTION]: The skill defines a syntax for ingesting data between agents (INVOKE/RESPONSE tags).
Ingestion points: Text within [INVOKE:role|question] and [RESPONSE:role] blocks in SKILL.md.
Boundary markers: Explicit bracketed tags and chain tracking [CHAIN:...] serve as effective delimiters.
Capability inventory: No risky capabilities (subprocess, network, file-write) are present in the skill.
Sanitization: The protocol requires explicit attribution and confidence scoring, which mitigates the impact of potentially poisoned data.
[DATA_EXPOSURE]: No hardcoded credentials, sensitive file paths, or data exfiltration mechanisms were found. The protocol handles business metrics (ARR, burn rate) within the agent context only.

agent-protocol