The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from Jira tickets (acceptance criteria, comments), API responses, and database records. This data is interpolated into prompts for sub-agents to generate test plans, scenarios, and reports. While the documentation mentions delimiters as a mitigation, there is no strict enforcement, creating a surface for potential injection.
Ingestion points: Processes story.md, acceptance-criteria.md, and comments.md (materialized from Jira), API response bodies from authenticated curl requests, and database query results.
Boundary markers: Mentions the use of delimiters and 'ignore embedded instructions' warnings as remediation guidance in the documentation, but these are not rigidly implemented in the prompt generation logic.
Capability inventory: The skill possesses significant capabilities, including executing shell commands via bun run and curl, writing to the local filesystem (PBI and session folders), and performing write operations to external issue trackers (Jira) and Test Management Systems (Xray).
Sanitization: Relies on template-based reporting and ADF conversion for Jira publishing, but does not implement comprehensive sanitization or filtering of external NL content before processing.
[COMMAND_EXECUTION]: The skill executes local shell commands using bun run for project-specific scripts (e.g., jira:sync-issues, api:login) and uses osascript via a local JavaScript utility for clipboard interaction. These are part of the intended QA automation framework.
[EXTERNAL_DOWNLOADS]: Performs network requests using curl to project-specific environment URLs (defined in .agents/project.yaml) for reachability checks and API testing. These operations are scoped to the project's infrastructure.

sprint-testing