The Agent Skills Directory

[SAFE]: Analysis of the skill's documentation, instructions, and prompt templates reveals no security threats or malicious patterns.- [INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted source code during exploration. This attack surface is mitigated by explicit instructions in the subagent prompts (references/explorer-prompt.md and references/explainer-prompt.md) that label evidence as untrusted data and restrict tool usage to read-only operations.- [DATA_EXFILTRATION]: The skill defines its scope as read-only exploration with no external side effects. There are no network-capable tools, credential-accessing instructions, or exfiltration patterns present in the analyzed files.- [COMMAND_EXECUTION]: Exploration tasks are performed using restricted, read-only primitives (Read, Grep, Glob) within a platform-managed subagent lifecycle. No arbitrary shell execution or system modification patterns were detected.

how