The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests untrusted data from an external source (Linear issue titles and descriptions) and incorporates it into the logic of subagents.
Ingestion points: The skill fetches issue data via mcp__linear-server__list_issues and mcp__linear-server__get_issue (found in SKILL.md).
Boundary markers: While prompts for subagents (e.g., in references/prompts/llm-fallback-bucket.md) use structured section headers like ### Issue, they lack robust delimiters (such as random nonces) to isolate untrusted content from the agent's instructions.
Capability inventory: The skill has the capability to write to the local file system (reports) and modify external platform state by updating issue descriptions and posting comments on Linear.
Sanitization: There is no evidence of sanitization or character escaping of the Linear data before it is interpolated into prompts or written back to the platform.
[COMMAND_EXECUTION]: The skill performs file system operations, such as directory creation using mkdir -p and writing audit report files as described in the artifact generation section of SKILL.md. These actions are within the expected scope of a product vision auditing tool.

prioritize