The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local shell scripts (detect-mode.sh) and standard build tools (such as mvn, gradle, npx, and dotnet) to scan project code, detect stacks, and validate the compilation of generated tests. These operations are restricted to the context of the testing orchestration and include error handling to stop execution on failure.
[INDIRECT_PROMPT_INJECTION]: The skill represents a surface for indirect prompt injection because it ingests untrusted data from the project being analyzed (source code, existing tests, API schemas) and from external web search results used to research best practices.
Ingestion points: Project analysis phases in references/mode-install.md (Step S2) and references/mode-create.md (Step C2), plus web research tasks in references/mode-rules.md (Step L2).
Boundary markers: The skill uses highly structured prompts for its sub-agents, employing clear headers (GOAL, ROLE, SCOPE, CONTEXT, CONSUMER, DONE) to isolate instructions from data.
Capability inventory: The skill and its agents have permissions for Read, Write, Edit, Bash, Agent (delegation), and WebSearch tools.
Sanitization: Significant mitigation is provided through mandatory human-in-the-loop checkpoints (AskUserQuestion) at every critical decision point, including agent roster approval, scenario verification, and final test review.

e2e