The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from JIRA defect tickets to drive its logic, which poses a risk of indirect prompt injection where malicious instructions in a ticket could manipulate agent behavior.
Ingestion points: Phase 1 and Phase 2 load external data from .SS_WF/{{$var[ticket_id]s}}_JIRA_OUTPUT_.json.
Boundary markers: The instructions do not define clear delimiters or specific instructions for the agent to ignore or isolate potential commands embedded within the ticket content.
Capability inventory: The skill has significant capabilities including modifying source code (Phase 4), updating local documentation files (CODE_GENERATION_SUMMARY.md and CODING_AGENT_LEARNINGS.md), and executing shell-based tests via the run-test-cases task.
Sanitization: There is no description of content sanitization, validation, or escaping of the ticket data before it is used for decomposition, root cause analysis, or code generation.
[EXTERNAL_DOWNLOADS]: The skill fetches design and content configuration from well-known services like Figma and Sitecore via the context-gathering task to inform the defect fix.
Evidence: Phase 2 utilizes context-gathering to refetch artifacts from Figma or Sitecore based on the issue category (e.g., UI, Sitecore) and evidence of source changes in the ticket.

master-defect-fix-orchestrator