design-decision-gate: reduce max-turns 20→12, add explicit turn budget and stopping criteria

[Copilot]

The Design Decision Gate was configured with max-turns: 20 but averaged only 7.1 turns — a 5× variance (4–20) that signals over-investigation on complex PRs with no hard stopping signal.

Changes

• max-turns: 20 → max-turns: 12 — tightens the engine-enforced ceiling while preserving headroom for complex ADR generation scenarios

• Turn Budget Allocation table — maps each step to a per-step turn cap (≤7 total), with turns 8–12 reserved for fallback, unexpected data gaps, or complex ADR generation:

  Step	Max turns
  Read pre-fetched summary + gate check	1
  Fetch PR details (only if needed)	1
  ADR search (branch + PR body)	1
  Fetch linked issue ADR (only if referenced)	1
  Generate draft ADR or verify alignment	2
  Commit draft + post comment	1
  Total	≤ 7

• Mandatory Stopping Criteria — six named exit conditions, each with a specific safe output to emit immediately:

  • Noop (neither condition met)
  • ADR found, no divergence → approved comment
  • ADR found, divergence → divergence comment
  • No ADR, decision inferable → push draft + blocking comment
  • No ADR, decision not inferable → comment with missing context
  • Turn 11 reached → emit best available output on turn 12 (hard engine limit), no new investigation

[github-actions]

Hey @Copilot 👋 — great direction on adding a turn budget guardrail to the Design Decision Gate! Keeping agent runs bounded is an important reliability improvement.

A few things to address before this is ready for review:

• No changes yet — the diff is currently empty. Once commits land, make sure the workflow files include the updated max-turns, prompt body changes, turn allocation table, and stopping criteria.
• Description missing motivation — the PR body is a checklist of tasks, but doesn't explain why this guardrail is needed. A short paragraph summarising the motivation (e.g. prevent runaway agent loops, improve cost predictability) would help reviewers.
• No tests — once the changes land, make sure make recompile and make agent-finish pass.

If you'd like a hand completing this, you can assign this prompt to your coding agent:

Complete the changes described in PR #29168 for the Design Decision Gate workflow:
1. Lower max-turns from 20 to 8 in the workflow frontmatter.
2. Update the Hard Turn Budget text in the prompt body to reflect 8 turns.
3. Add an explicit per-step turn allocation table to the prompt.
4. Add an explicit stopping criteria / decision tree section.
5. Run make recompile and make agent-finish to validate.
6. Update the PR body with a clear description of why this guardrail is needed.

Generated by Contribution Check · ● 1.2M · ◷

[pelikhan]

@copilot keep turn at 12

[Copilot]

@copilot keep turn at 12

Done in f484b1f — max-turns is now 12. The prompt's hard budget, reserve-turn note, and hard-stop trigger (turn 11 → emit on 12) are all updated to match.

Warning

Firewall rules blocked me from connecting to one or more addresses (expand for details)

I tried to connect to the following addresses, but was blocked by firewall rules:

• https://api.github.com/graphql
  • Triggering command: /usr/bin/gh /usr/bin/gh api graphql -f query=query($owner: String!, $name: String!) { repository(owner: $owner, name: $name) { hasDiscussionsEnabled } } -f owner=github -f name=gh-aw (http block)
  • Triggering command: /usr/bin/gh /usr/bin/gh api graphql -f query=query($owner: String!, $name: String!) { repository(owner: $owner, name: $name) { hasDiscussionsEnabled } } -f owner=github -f name=gh-aw _j9me15veeKU (http block)
• https://api.github.com/repos/actions/checkout/git/ref/tags/v6
  • Triggering command: /usr/bin/gh gh api /repos/actions/checkout/git/ref/tags/v6 --jq [.object.sha, .object.type] | @tsv (http block)
• https://api.github.com/repos/actions/setup-node/git/ref/tags/v6
  • Triggering command: /usr/bin/gh gh api /repos/actions/setup-node/git/ref/tags/v6 --jq [.object.sha, .object.type] | @tsv (http block)
• https://api.github.com/repos/astral-sh/setup-uv/git/ref/tags/eac588ad8def6316056a12d4907a9d4d84ff7a3b
  • Triggering command: /usr/bin/gh gh api /repos/astral-sh/setup-uv/git/ref/tags/eac588ad8def6316056a12d4907a9d4d84ff7a3b --jq [.object.sha, .object.type] | @tsv (http block)
• https://api.github.com/repos/github/gh-aw
  • Triggering command: /usr/bin/gh gh api /repos/github/gh-aw --jq .default_branch (http block)

If you need me to access, download, or install something from one of these locations, you can either:

• Configure Actions setup steps to set up my environment, which run before the firewall is enabled
• Add the appropriate URLs or hosts to the custom allowlist in this repository's Copilot coding agent settings (admins only)

[Copilot]

Pull request overview

Reduces the Design Decision Gate agent’s maximum investigation depth and adds explicit guidance for how to spend turns and when to stop, to prevent over-investigation on complex PRs.

Changes:

• Lowered max-turns from 20 to 12 in the gate’s engine configuration and the generated lock workflow.
• Added a per-step “Turn Budget Allocation” table to guide investigation sequencing.
• Added explicit “Mandatory Stopping Criteria” to enforce early exits and a hard stop near the turn limit.

Show a summary per file

File	Description
.github/workflows/design-decision-gate.md	Updates engine turn limit and adds turn budgeting + mandatory stopping criteria guidance for the agent.
.github/workflows/design-decision-gate.lock.yml	Regenerates the compiled workflow to reflect the new --max-turns 12 and related metadata.

Copilot's findings

Tip

Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

• Files reviewed: 2/2 changed files
• Comments generated: 3