[ambient-context] Daily Ambient Context Optimizer - 2026-10-18 #69991

Executive Summary

• Runs sampled: 4 (reduced data: downloaded logs contain no api-proxy-logs/event*.jsonl or copilot-session-state events, so first-request text could not be extracted; first-call input_tokens from token_usage.jsonl is used as a proxy; no Python text analysis was possible).
• Workflows: 3 (Daily Cache Strategy Analyzer, Daily Go Test Parallelizer, Smoke Copilot Routed Steering x2).
• First-call input tokens: 18125 / 13175 / 3 (routing classifier). Median ≈ 13.2k, p95 ≈ 18.1k.
• Conclusion: dominant cost is working-set rebuild across turns (WSRF 22-84), not first-request size. Cache Strategy Analyzer re-sends a 16 KB prompt plus growing context over 53 calls (5.67M cumulative input tokens, peak 188k).

Highest-Leverage Changes

1. daily-cache-strategy-analyzer.md (16 KB): move Phase 2/3 detection and data gathering into deterministic steps: so the agent gets compact JSON findings; inline templates (~lines 320-450) are already on-demand skills. Impact: high. Manual review. Keep >=60% of current size.
2. Smoke Copilot Routed Steering: WSRF 80-84 over 92-96 invocations with peak only ~28k. Bound turns / scenario to stop resending context (~2.3M tokens/run). Impact: high. Manual review.
3. daily-go-test-parallelizer.md (4.3 KB): move file selection and safety grep into steps: so the main agent starts with one file (WSRF 22.8, 35 calls, peak 63k). Impact: medium. Safe to try.
4. Summarise large logs/audit tool outputs via file paths rather than inline results; caching works (cache_read equals prior input), so waste is tool-output accumulation. Impact: medium.

CI-Validation Checklist for Implementing Agents

• 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 — do not re-attempt changes to any file 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

• Items 1 and 3: move data gathering/detection to steps:; use agentic-workflows MCP tools rather than raw gh aw prose.

Skills

• Item 1: keep templates as on-demand skills; avoid duplicating them in Important Guidelines.

Agents

• Items 2 and 4: bound turns for smoke steering; keep sub-agent batches small as in go-test-parallelizer.

References

• §39154568172
• §39152981591
• §39154765378

Generated by 🌫️ Daily Ambient Context Optimizer · copilot · auto · 25.5 AIC · ⌖ 0.764 AIC · ⊞ 12.6K · ◷

• expires on Oct 25, 2026, 11:50 AM UTC-08:00