The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from GitHub issue bodies and titles to drive automated code implementation and branch creation. This creates a surface where malicious instructions in an issue could influence the agent's behavior.
Ingestion points: Reads issue content via gh issue view during the 'Start' and 'Spec gate' phases.
Boundary markers: The skill defines specific 'briefs' for subagents and uses structured files in a .afkkit/ directory (e.g., assumptions.md, checks.md) to isolate decisions and criteria.
Capability inventory: Extensive use of Bash for shell commands, Edit and Write for file system changes, and Agent or SendMessage for subagent orchestration.
Sanitization: Not explicitly mentioned for issue text, but the skill relies on an independent 'Acceptance check' subagent to verify the builder's work and an escalation policy to halt on ambiguous instructions.
Mitigation: The skill strictly enforces a requirement for a ready label, which is intended to be applied by a human after a manual review process, providing a critical human-in-the-loop safety barrier.
[COMMAND_EXECUTION]: The skill uses the Bash tool to perform numerous local and remote operations, including repository management, build tool execution (e.g., pnpm), and interaction with the GitHub CLI (gh). This level of access is inherent to its primary purpose as an automated workflow runner.

afkkit