The Agent Skills Directory

[NO_CODE]: The skill is comprised entirely of Markdown instructions, templates, and YAML metadata. No source code (e.g., Python, JavaScript) or shell scripts were found within the provided files.
[SAFE]: No network operations, external downloads, or sensitive file access patterns were detected. All described procedures are internal to the agent's reasoning process.
[SAFE]: There are no indicators of prompt injection, obfuscation, persistence, or privilege escalation. The instructions focus exclusively on the minimax regret decision criterion.
[INDIRECT_PROMPT_INJECTION]: The skill has a minimal and safe attack surface for indirect prompt injection. \n
Ingestion points: The skill processes user-provided options, states, and payoffs as described in SKILL.md and references/TEMPLATE.md. \n
Boundary markers: No explicit delimiters are specified for input data, though the output follows a strict matrix format. \n
Capability inventory: The skill lacks any dangerous tools (networking, file writes, or shell execution) across all files. \n
Sanitization: No explicit sanitization is described. \n
Result: The assessment is safe as the skill possesses no capabilities that could be leveraged for malicious activity via injected data.

think-minimax-regret