The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to run various shell commands and pipelines using tools such as find, rg, xargs, npm, npx, pytest, jq, awk, and madge. These are used to identify test files, generate dependency graphs, and execute test suites. While standard for development workflows, these commands grant the agent direct execution capabilities in the local shell.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted project files, specifically spec-product.md and spec-tech.md, to determine the product type and testing scope. The lack of explicit boundary markers or sanitization logic when reading these files could allow adversarial instructions embedded in the metadata to influence the agent's behavior.
Ingestion points: Reads spec-product.md and spec-tech.md to extract product_type, appetite, and scope.
Boundary markers: None implemented for the ingested markdown/frontmatter content.
Capability inventory: Full subprocess execution for test runners and system utilities (find, rg, npm, pytest, cargo, go).
Sanitization: No filtering or escaping is applied to the extracted metadata before it is used to define the testing strategy.
[DYNAMIC_EXECUTION]: The skill dynamically assembles shell commands at runtime, such as constructing the testPathPattern for npm test by piping the results of a file search into a string. This reliance on the local file system state for command construction is a standard testing pattern but creates a dependency on the integrity of the project's file structure.

stelow-product-testing-ai-code