The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes multiple untrusted data sources including git history (git log), internal changelogs, backlog tasks, and design documents. While the goal is translation to player-facing notes, there are no explicit instructions to ignore or strip embedded prompt injection attempts within these data sources.
Ingestion points: Reads from production/releases/[version]/gamedev:changelog.md, docs/CHANGELOG.md, backlog/tasks/, design/balance/, and git log output.
Boundary markers: None present. The skill interpolates data directly into the categorization and translation phase.
Capability inventory: Uses Read, Glob, Grep, Write, and Bash to gather and save data.
Sanitization: Missing instructions to escape or filter markdown or text that could contain instructions targeted at the LLM during the 'Categorize and Translate' phase.

patch-notes