[q] optimize: reduce token usage in high-consumption workflows

[github-actions]

Q Workflow Optimization Report

Issues Found (from live data)

Contribution Check

• Run analyzed: 25686395186
• Token usage: 663K tokens / 6.7M effective tokens / 24 turns
• Issues identified:
  • No timeout-minutes — could run unboundedly if subagent dispatch stalls
  • No max-continuations cap — orchestrator + 3 subagent calls = 24 turns, no hard ceiling

Matt Pocock Skills Reviewer

• Run analyzed: 25686664431
• Token usage: 456K tokens / 4.5M effective tokens / 10 turns
• Issues identified:
  • No max-continuations cap for Copilot engine
  • Agent fetches full gh pr diff inline — large PRs produce multi-thousand-line diffs fed directly into the context window

Test Quality Sentinel

• Run analyzed: 25686664489
• Token usage: 285K tokens / 2.8M effective tokens / 7 turns
• Issues identified:
  • max-continuations: 40 is 5-6× the observed run length (7 turns) — allows significant cost overrun on adversarial or complex PRs

Changes Made

test-quality-sentinel.md

• Reduced max-continuations from 40 → 15 (2× the observed maximum of 7 turns, still provides comfortable headroom)

contribution-check.md

• Added max-continuations: 60 (covers 3× subagent runs of ~20 turns each)
• Added timeout-minutes: 30 (hard wall to prevent runaway on retry loops)

mattpocock-skills-reviewer.md

• Changed engine: copilot → engine: {id: copilot, max-continuations: 15}
• Added pre-agent step Pre-fetch PR diff (truncated) that caps the diff at 8 000 lines before the agent starts — prevents huge diffs from inflating the context window on every turn
• Updated Step 1 in the prompt to read the pre-fetched file instead of re-running gh pr diff

Expected Improvements

• Test Quality Sentinel: worst-case token spend reduced by ~63% (40 → 15 continuation cap)
• Matt Pocock Reviewer: large-PR scenarios bounded; diff no longer re-fetched live inside the agent loop
• Contribution Check: runaway protection added; cost capped even if subagent retry loops occur

Validation

All modified workflows compiled successfully:

• ✅ test-quality-sentinel
• ✅ contribution-check
• ✅ mattpocock-skills-reviewer

Note: .lock.yml files will be regenerated automatically after merge.

References

• Log analysis: run IDs 25686395186, 25686664431, 25686664489

• Triggered by: /q optimize token usage on issue [Incident] Incident: github-pages deployment failure (run 25685733509) #31552

• Fixes [Incident] Incident: github-pages deployment failure (run 25685733509) #31552

🎩 Equipped by Q · ● 26.1M · ◷

• expires on May 13, 2026, 6:18 PM UTC

---

Note

This was originally intended as a pull request, but the git push operation failed.

Workflow Run: View run details and download bundle artifact

The bundle file is available in the agent artifact in the workflow run linked above.

To create a pull request with the changes:

# Download the artifact from the workflow run
gh run download 25688146365 -n agent -D /tmp/agent-25688146365

# Fetch the bundle into a local branch
git fetch /tmp/agent-25688146365/aw-q-optimize-token-usage.bundle refs/heads/q/optimize-token-usage:refs/heads/q/optimize-token-usage-20fef7476b3af908
git checkout q/optimize-token-usage-20fef7476b3af908

# Push the branch to origin
git push origin q/optimize-token-usage-20fef7476b3af908

# Create the pull request
gh pr create --title '[q] optimize: reduce token usage in high-consumption workflows' --base main --head q/optimize-token-usage-20fef7476b3af908 --repo github/gh-aw