Use API proxy event logs as the ambient context source of truth

[Copilot]

The ambient context optimizer was analyzing prompt.txt instead of the request text actually sent to the model, which led to false inline-agent findings and missed real compiler leakage. This updates the workflow guidance to treat API proxy event logs as authoritative, while keeping prompt.txt only as a diagnostic cross-check.

• Request source selection

  • Prefer the canonical API proxy request log:
    • sandbox/firewall/logs/api-proxy-logs/event-logs.jsonl
  • Accept the legacy filename:
    • sandbox/firewall/logs/api-proxy-logs/events.jsonl
  • Preserve the matching sandbox/firewall-audit-logs/... fallback path.
  • Use Copilot session events.jsonl only as the fallback source for first user.message extraction.

• prompt.txt semantics

  • Stop treating prompt.txt as ambient-context source of truth.
  • Keep it only for compiler/debug cross-checking against the API proxy request payload.
  • Explicitly call out inline agents or inline linters present only in prompt.txt as likely compilation bugs rather than workflow bloat.

• Cross-validation metadata

  • Add guidance to capture:
    • request_input_tokens
    • prompt_chars
    • request_prompt_char_delta
  • Add per-request char/token ratio to make size analysis less dependent on raw character counts alone.

• Analysis rubric

  • Extend deterministic analysis to count inline linters (## linter:) alongside inline agents and skills.
  • Clarify that the workflow should optimize the actual first request sent to the DLLM, not the compiled prompt artifact.

Example of the updated source-order rule:

1. use the first DLLM request payload from
   sandbox/firewall/logs/api-proxy-logs/event-logs.jsonl
2. fall back to sandbox/agent/logs/copilot-session-state/<session-id>/events.jsonl
3. treat prompt.txt only as a cross-check artifact

[github-actions]

Hey @Copilot 👋 — thanks for tackling the ambient context analysis fix! Switching the primary request source from prompt.txt to the API proxy event logs in daily-ambient-context-optimizer.md is a meaningful correctness improvement, and the complementary removal of --artifacts usage in agentics-maintenance.yml keeps both files in sync.

A couple of things to address before this is ready for review:

• Complete the final checklist item — "Run targeted validation, secret scan, and final review checks" is still unchecked, and the PR is in draft/WIP state. Mark this done and promote to ready-for-review when validation passes.
• No test changes in the diff — The checklist marks "Add or update focused tests for the corrected behavior" as done, but no test files appear in the diff. If tests live outside this diff (e.g., a separate commit or separate PR), please link to them. If test coverage for the new log-path logic isn't needed, a brief note in the PR body explaining why would help reviewers.

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

Complete the remaining checklist item for PR #37655 in github/gh-aw:

1. Run the targeted validation checklist:
   - Confirm the API proxy event log path `sandbox/firewall/logs/api-proxy-logs/event-logs.jsonl` (and fallback `sandbox/firewall-audit-logs/...`) resolves correctly in test runs.
   - Run a secret scan over the changed files to confirm no credentials or tokens are embedded.
   - Review the updated eligibility rules in daily-ambient-context-optimizer.md to ensure the fallback chain (API proxy → events.jsonl → prompt.txt cross-check) is logically consistent.
2. If test coverage for the new log-path resolution logic is warranted, add or update the relevant test files and include them in this PR.
3. Once all checklist items are checked, remove the [WIP] prefix from the PR title and mark the draft as ready for review.

Generated by ✅ Contribution Check · 466.7 AIC · ⌖ 13.2 AIC · ⊞ 24.3K · ◷

[Copilot]

Pull request overview

This PR updates the “Daily Ambient Context Optimizer” workflow guidance to treat API proxy event logs as the authoritative source for the first DLLM request, and de-emphasizes prompt.txt to a secondary cross-check artifact to reduce false findings in ambient-context analysis.

Changes:

• Updates sampling/extraction rules to prioritize API proxy event logs and use prompt.txt only for compilation-debug cross-checking.
• Extends deterministic analysis guidance to include inline linter counting and request char/token ratio metadata.
• Adjusts the maintenance workflow’s gh aw logs invocation by removing the --artifacts usage filter.

Show a summary per file

File	Description
.github/workflows/daily-ambient-context-optimizer.md	Revises first-request source-of-truth and analysis rubric/metadata for ambient-context optimization.
.github/workflows/agentics-maintenance.yml	Changes the artifact download scope for the activity report logs step by removing --artifacts usage.

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