The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains strong instructions that command the agent to prioritize external data over its own internal evaluation. Specific directives include stating that the "ticket's Actual vs Expected is ground truth" and that "Dev notes... outrank the analysis plan, the code-generation document and your own judgement." This effectively instructs the agent to disregard its own reasoning if it conflicts with provided external inputs.
[INDIRECT_PROMPT_INJECTION]: The skill has a significant attack surface for indirect injection as it ingests various external documents and uses them to influence code generation and command execution.
Ingestion points: The agent is directed to read and process data from files such as .SS_WF/Agent/Analysis/{{$var[parent_ticket_id]s}}_ANALYSIS_PLAN.md, figma-output/, and .SS_WF/Agent/CODE/{{$var[parent_ticket_id]s}}_CODE_GENERATION.md, as well as "Intake blocks" from previous phases.
Boundary markers: There are no instructions to treat content from these files as untrusted data or to use specific delimiters to prevent embedded instructions in these files from being executed by the LLM.
Capability inventory: The agent has the capability to read any file in the repository, write and modify source code, and execute code via the run-test-cases skill.
Sanitization: The instructions do not mention any sanitization, filtering, or validation of the data retrieved from the analysis plans or figma outputs before it is used to determine code changes.

defect-investigate-and-fix