The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements an autonomous execution model where the agent is instructed to read and execute instructions from external markdown files (e.g., tests/docs/02-CLI-TEST-CASES.md). This creates a significant surface for indirect prompt injection if those files are modified by an untrusted party. Ingestion points: Markdown files in tests/docs/. Boundary markers: Absent; instructions emphasize strict adherence to external content. Capability inventory: Shell command execution, file system access, and network operations to perform test steps. Sanitization: Absent; the agent is directed to follow documented steps precisely.
[COMMAND_EXECUTION]: The skill's core functionality involves the agent executing various shell commands for setup and testing, including git clone, pnpm install, docker exec, and ccpm install.
[PROMPT_INJECTION]: The 'Master QA Prompt' uses forceful directives ('CRITICAL INSTRUCTIONS', 'MANDATORY RULES') and role-play to ensure the agent strictly follows the test plan and overrides its default operating behavior.
[DATA_EXPOSURE]: Documentation and templates include examples that access local configuration directories (e.g., ~/.claude/skills/) and suggest testing for path traversal vulnerabilities.

qa-expert