The Agent Skills Directory

[SAFE]: The skill consists entirely of Markdown documentation, YAML configuration, and JSON data. No executable scripts or binary files are present.
[SAFE]: The documentation includes extensive security-focused 'traps' and 'anti-patterns' sections (e.g., references/patterns-scenarios-traps.md) which educate users on avoiding tacit collusion, sybil attacks, and miscalibrated agent self-reports.
[SAFE]: All external references are to legitimate academic sources (arXiv), official documentation, or the vendor's own repositories.
[INDIRECT_PROMPT_INJECTION]: The skill documents the attack surface of multi-agent synthesis but provides explicit, research-backed mitigations (Reasoning-Tree Audit, Per-Claim Credibility Scoring) to prevent malicious inputs from compromising the system. This is a safety feature rather than a vulnerability.
[DYNAMIC_CONTEXT_INJECTION]: The skill mentions the !command`` syntax in a purely descriptive and safe context related to workflow examples like git log, with no malicious payload execution.

foundations-game-theory