The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external data (specifications and tickets) to guide agent behavior and implementation tasks, creating a potential vector for instructions hidden within data.
Ingestion points: Input specifications and ticket descriptions mentioned throughout SKILL.md that guide the agent's implementation strategy.
Boundary markers: The instructions lack defined delimiters or explicit safety prompts to ignore embedded commands within the input specifications.
Capability inventory: The skill orchestrates subagents with the ability to create and modify files, manage git branches and worktrees, and perform code merges.
Sanitization: The workflow does not describe any validation or sanitization steps for the text contained in the specifications or tickets before they are processed by the agent.

implement-spec