The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from PR diffs, branch names, and issue bodies/checklists to determine 'Acceptance criteria' for testing. This represents an attack surface where malicious instructions in a PR or issue could potentially influence the agent's behavior during QA.
Ingestion points: SKILL.md references reading PR numbers, branch names, and issue checklists.
Boundary markers: The instructions do not define specific delimiters or 'ignore' instructions for the ingested external data.
Capability inventory: The skill utilizes git for worktree management, gh for GitHub interactions, and browser automation tools.
Sanitization: No explicit sanitization or validation of the ingested external content is described.
[COMMAND_EXECUTION]: The skill uses git worktree add to create isolated environments for testing and gh pr comment to post results. These are standard project management operations but involve shell command execution based on PR and branch identifiers.
[DATA_EXPOSURE]: The skill accesses project worktrees and references environment variables. It follows a secure practice by instructing the agent to use environment variable names rather than hardcoded values for test logins.

agentic-qa