Skip to content

[deep-report] Add Design Decision Gate turn-budget guardrail — 4–20 turn variance observed #29149

@github-actions

Description

@github-actions

Description

Workflow execution logs (run §25117657405) flagged Design Decision Gate 🏗️ with medium-severity execution drift:

Workflow varied from 4 to 20 turns across runs (avg 7.1 turns) — suggests changing task shape or unstable prompts.

A 5× variance in turn count signals that the agent prompt lacks clear stopping criteria. In high-turn runs the agent likely over-investigates before reaching a decision, burning unnecessary tokens and time.

Expected Fix

  1. Read the current Design Decision Gate workflow prompt
  2. Add an explicit turn budget instruction (e.g., complete your analysis in at most 8 turns)
  3. Add explicit decision tree / stopping criteria: when to output a decision vs. when to ask for clarification
  4. Add a test run to verify the turn variance narrows (target: < 2× spread)

Expected Impact

  • More consistent execution time and cost
  • Reduced tail-case token usage on complex PRs
  • Improves predictability for downstream workflows that may chain on its output

Suggested Agent

Design Decision Gate itself in a self-improvement pass, or a general Claude code agent reviewing the prompt file.

Estimated Effort

Fast (< 30 min) — prompt-only change.

Data Source

DeepReport Intelligence Briefing — 2026-04-29 · Workflow execution logs · Observability insights from logs batch · Run §25117657405

Generated by DeepReport - Intelligence Gathering Agent · ● 654.3K ·

  • expires on May 1, 2026, 3:37 PM UTC

Metadata

Metadata

Type

No type
No fields configured for issues without a type.

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions