The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses shell commands to manage the test lifecycle, utilizing python3 for result tracking and git for repository state management and commit verification.
[DYNAMIC_EXECUTION]: The skill automates the generation and application of product code fixes by dispatching isolated 'fix subagents' using the root-cause skill. These subagents are instructed to modify the codebase to resolve detected test failures.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from the target application's UI and external test guides, creating a vulnerability surface where malicious content could influence the agent's behavior.
Ingestion points: Application UI (captured via kimi-webbridge), dogfood.json test guides, and dogfood-review.md reports.
Boundary markers: The instructions mandate strict evidence fields (saw for UI content, server for backend probes) and require 'quoted' observations, although no automated sanitization is described.
Capability inventory: Powerful browser automation (navigating, interaction), file system write access for committing code fixes, and the ability to spawn subagents.
Sanitization: Relies on natural language instructions to quote UI strings rather than programmatic escaping.

run-dogfood