The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to discover and execute build and startup commands (e.g., Docker Compose, Makefiles, Procfiles, and CI workflows) derived directly from the target repository's documentation and source code to establish a test environment.
[INDIRECT_PROMPT_INJECTION]: The skill ingests instructions and configuration from repository files such as AUTOQA.md, README.md, and CLAUDE.md to determine its behavior and test matrix. A malicious repository could supply deceptive instructions to manipulate the agent's actions or test outcomes.
Ingestion points: Multiple files within the target repository, prioritized in references/discovery.md (e.g., AUTOQA.md, README.md, Dockerfile, Procfile).
Boundary markers: The skill generates a plan matrix for review before execution, which provides a logical separation, but it lacks formal prompt delimiters when processing the content of ingested repository files.
Capability inventory: The agent is empowered to execute shell commands, perform network operations (curl), interact with browser tooling, and write persistent report files.
Sanitization: No explicit content sanitization or validation routines are specified for the instructions extracted from the repository documentation.
[DYNAMIC_EXECUTION]: The skill dynamically assembles a test matrix of executable shell commands and API calls based on its analysis of the repository's code diffs and feature catalog, which are then executed sequentially in the target environment.

autoqa