The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill presents an indirect prompt injection vulnerability surface when scanning local files.
Ingestion points: The workflows in characterization-test-generator/SKILL.md and tdd-refactor-guard/SKILL.md use Glob, Grep, and Read to import source files (**/*.swift) directly into the agent's context window.
Boundary markers: The prompt instructions do not mandate XML tags, delimiters, or clear guidance telling the model to ignore natural language instructions embedded within source code or comments.
Capability inventory: The skill allows powerful capabilities via the Bash tool, enabling it to execute xcodebuild test and execute local shell utilities like coverage-gate.sh or crap-report.sh.
Sanitization: No pre-processing sanitization or escaping mechanisms are specified before passing file strings to the model, meaning malicious code comments could influence agent behavior.

testing