The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local development tools and file management commands as part of its core functionality.
Evidence includes instructions to run lint, typecheck, build, and test based on detected project tooling in Stage 1.
Stage 4 uses shell commands like ls to verify the archival process and implies directory moves (mv).
[INDIRECT_PROMPT_INJECTION]: The skill processes content from a repository that may contain embedded instructions, creating a potential attack surface.
Ingestion points: The agent reads and analyzes git diff data and raw command outputs from automated testing tools in SKILL.md (Stage 1 and Stage 2).
Boundary markers: No explicit delimiters or instructions are provided to the agent or subagent to ignore instructions that might be contained within the diffs or tool logs.
Capability inventory: The skill has the ability to execute shell commands and invoke subagents to process the ingested data.
Sanitization: The skill lacks specific instructions for sanitizing or escaping the content read from the file system before analysis.

beyond-code-verify