The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data, specifically source code and terminal outputs from testing tools, to determine the 'release readiness' of a project. This creates a surface where malicious instructions embedded in code comments or test reports could attempt to influence the orchestrator's decision-making process.
Ingestion points: Affected source files (Step 1), test stdout/stderr and artifact paths (Step 5, 6, 7).
Boundary markers: The skill does not define specific string delimiters for untrusted input, but it utilizes a multi-agent boundary where data is passed to specialized validators.
Capability inventory: The skill possesses the Bash, Write, Edit, Skill, and Agent tools across its various steps.
Sanitization: The skill mitigates this risk through 'Hard Rule 10', which mandates that the final verdict is re-computed from raw evidence by deterministic code (e.g., qa-recompute-gate.js) rather than relying on the LLM's summarized interpretation. It also uses a dedicated qa-evidence-validator agent to audit the evidence bundle.
[COMMAND_EXECUTION]: The skill uses the Bash tool to execute a local utility script, qa-sdk.sh, and to trigger specialized testing sub-agents. These operations are restricted to the orchestration of QA workflows and the persistence of test evidence within the project's .qa/ directory.

enterprise-qa-testing