Skip to content

gate-evaluate prompt

The gate-evaluate prompt drives parallel quality gate evaluation for a slice. It runs before task execution begins, dispatching independent gate checks as subagents to catch issues early.

Gate evaluation is triggered automatically during auto-mode after a slice is planned but before its tasks execute. It’s part of the auto-mode pipeline — you don’t invoke it directly.

  1. Receives the slice plan and a list of quality gates (Q3–Q8) that apply to the slice
  2. Dispatches all gates in parallel using the subagent tool — each gate runs as an independent evaluation agent
  3. Collects results — each subagent calls gsd_save_gate_result with a verdict (pass, flag, or omitted)
  4. Reports the batch outcome — which gates passed, which flagged concerns, and which were omitted as not applicable
VerdictMeaning
passNo concerns found
flagConcerns identified — recorded but doesn’t block execution
omittedGate not applicable to this slice (e.g. no auth surface for Q3)

Gates returning omitted is expected for simple slices that don’t touch security, requirements, or other gate-specific surfaces.

VariableDescription
workingDirectoryAbsolute path to the working directory
milestoneId / milestoneTitleActive milestone context
sliceId / sliceTitleSlice being evaluated
slicePlanContentFull slice plan content for gate agents to analyse
gateCountNumber of gates to evaluate
gateListFormatted list of gate IDs and descriptions
subagentPromptsFull prompts for each gate subagent