[ambient-context] Daily Ambient Context Optimizer - 2026-10-09 #67271

Executive Summary

• Reduced-data report. The logs MCP call returned run metadata for 10 runs (the 60-run request timed out). The per-run artifacts under /tmp/gh-aw/aw-mcp/logs/ were unreadable in the agent sandbox (Permission denied). First-request text was therefore not extracted, and the Python char/duplication analysis was not run. All findings use run metadata, WSRF and workflow-source sizes only.
• 4 runs selected by AIC/WSRF: PR Code Quality Reviewer, Avenger, Design Decision Gate, PR Sous Chef. 4 distinct workflows. Request char medians and p95 are unavailable.
• Main conclusion: the cost signal is context re-sent across turns (WSRF 13.2 on PR Code Quality Reviewer, 4.9 on Avenger). Design Decision Gate carries a ~21 KB source prompt with a 5.6 KB imported agent.
• The audit MCP tool was run for Avenger only. Its WSRF comes from the logs metadata.
• Blocked per the closed-PR guard: pr-code-quality-reviewer.md, test-quality-sentinel.md. Both are excluded from the recommendations. The auto-pause rule is not triggered (close_rate is null).

Highest-Leverage Changes

1. Design Decision Gate: stop unconditionally importing adr-writer.agent.md (5.6 KB) and shared/reporting.md. Load the ADR writer only on the "no ADR" branch (workflow-md/agents, medium, needs manual review).
2. Design Decision Gate: shorten the "Turn Budget Allocation", "Mandatory Stopping/Efficiency Rules" blocks (lines ~174-225) that restate each other. Keep the existing ## skill: report templates (workflow-md, medium, safe).
3. Avenger (WSRF 4.9, 5 invocations, failed run): cut cumulative input by moving the CI-state check, merge and recompile prechecks into deterministic steps: that write a compact JSON. Then trim Steps 0-7 to branches the JSON selects (workflow-md, high, manual review).
4. PR Sous Chef (12 KB, 13 invocations, WSRF 12.3 on tiny counts): the queue is already pre-downloaded. Move the "Act on each eligible PR" section (lines ~192-242) into a skill block so noop runs stay small (workflow-md/skills, medium, safe).
5. Ponytail Reviewer (3 KB source but WSRF 5.3, 96K tokens, 6 invocations): the cost comes from imports. Check that pr-review-base plus the pinned external skill and the diff data fetch do not over-fetch (skills, low, manual review).

CI-Validation Checklist for Implementing Agents

Any agent implementing workflow-file recommendations must complete every item below before opening a PR:

• Run make recompile for every modified .github/workflows/*.md file — zero compilation errors required
• Confirm no target workflow file is reduced by more than 40% of its current character count
• Run make agent-report-progress before the final commit and confirm it passes
• Check the blocked_files list in /tmp/gh-aw/ambient-context/closed-pr-targets.json (written by Step 4) — do not re-attempt changes to any file that appears in a closed ambient-context optimization PR from the last 14 days
• Include a smoke-test result or dry-run output in the PR description for every changed workflow spec
• Document all generated .lock.yml changes in the PR body
• Do not submit as a draft PR

Key Metrics

Recommendations by Category

Workflow Markdown

See items 2-4 above (design-decision-gate, avenger, pr-sous-chef). Keep each file above 60% of its current size.

Skills

Item 4 (move the PR action section into a skill block) and item 5 (review ponytail skill and diff-fetch imports).

Agents

Item 1: import adr-writer.agent.md conditionally instead of always. Needs manual review for behavior.

References

• §37981206055
• §37980958799
• §37981205227

Generated by 🌫️ Daily Ambient Context Optimizer · copilot · auto · 22.3 AIC · ⌖ 0.748 AIC · ⊞ 12K · ◷

• expires on Oct 16, 2026, 11:53 AM UTC-08:00