The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes content from game design documents (GDDs) which could contain adversarial instructions designed to influence the agent's behavior during the scan.
Ingestion points: The skill reads from design/registry/entities.yaml and performs Grep operations on all files within design/gdd/*.md.
Boundary markers: The instructions do not define clear delimiters (e.g., XML tags or triple quotes) or provide system instructions to the agent to treat the document content as data rather than instructions.
Capability inventory: The skill possesses powerful capabilities including Bash execution and the ability to Write/Edit project files, which increases the potential impact of a successful injection.
Sanitization: There is no evidence of sanitization or filtering applied to the text extracted from GDDs before the agent processes it for conflict detection.
[COMMAND_EXECUTION]: The skill utilizes the Bash tool to execute shell commands.
Evidence: Phase 2 uses git log --name-only combined with grep and sort to identify files modified since the last review. The command is scoped to specific directories and does not incorporate unvalidated user input, limiting the risk of arbitrary command injection.

consistency-check