[ambient-context] Daily Ambient Context Optimizer - 2026-10-19 #70357

Executive Summary

• Sampled 4 runs / 4 workflows (Issue Monster, Copilot Agent PR Analysis, Auto-Triage Issues, Design Decision Gate), chosen by highest AIC among runs with usable first-request text.
• First request (system + user prompt from aw_session.jsonl): median 30,391 chars, p95 36,177 chars. System prompt is a stable ~11–15K chars; the rest is workflow-specific (15–25K chars).
• Conclusion: the biggest avoidable weight is inline procedural text and output templates in a few workflow bodies, especially Issue Monster (one 14.9K-char section).
• Data limits: logs hit its 60s timeout (partial download); audit timed out, so no WSRF. API proxy request payloads were not captured, so session prompt.system/prompt.user events were used.

Highest-Leverage Changes

1. Issue Monster: move the "Review Pre-Searched and Prioritized Issue List" section (14,880 chars, ~60% of user prompt) into a pre-step or on-demand ## skill: block. Impact: high. Needs manual review.
2. Copilot Agent PR Analysis: move the ~3.2K-char inline report template into a skill; drop prose duplicating shared/copilot-pr-analysis-base.md. Impact: medium. Safe.
3. Design Decision Gate: shrink "Mandatory Efficiency Rules" (1.5K) and move the ADR draft template ("No ADR Found", 1.6K) into a skill. Impact: medium. Needs manual review.
4. Auto-Triage Issues: scheduled and workflow_dispatch sections (~2.8K together) are largely parallel; merge into one procedure plus a short per-mode delta. Impact: low-medium. Safe.
5. Auto-Triage Issues: 6 raw gh aw mentions in source; rewrite as explicit agentic-workflows MCP tool calls (logs, audit, status). Impact: low. Safe.

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

Metric	Value
Sampled runs	4
Distinct workflows	4
Median chars	30,391
P95 chars	36,177
Largest sampled request	36,177 (Issue Monster)
Merged optimizer PRs (7d)	0
Closed optimizer PRs (7d)	0 (1 earlier: #68002; no blocked files)
Optimizer PR close-rate (7d)	n/a (<3 settled PRs; auto_pause=false)
WSRF (audited runs)	n/a (audit timed out)

<b>Per-Run First-Request Metrics</b>

Run	Workflow	Chars	System	User	Headings	List items	Dup-line ratio	AIC
39302459462	Issue Monster	36,177	11,268	24,907	15	93	0.004	4.25
39302791923	Copilot Agent PR Analysis	30,472	15,048	15,422	36	120	0.004	17.62
39302952246	Auto-Triage Issues	28,669	12,512	16,155	31	112	0.017	20.43
39303780333	Design Decision Gate	30,310	13,825	16,483	43	140	0.004	9.54

<b>Repeated Ambient Context Signals</b>

• Duplicate line/paragraph ratios are near zero: the bloat is unique prose, not copy-paste.
• Every request carries ~30–35 lines mentioning tools and safe outputs, mostly system-prompt boilerplate shared across all runs.
• No inline agents, skills or linters, and no SKILL.md references, appear in any sampled request.
• Largest sections: Issue Monster issue-list review (14,880); Copilot Agent PR Analysis report template (3,203); Auto-Triage scheduled/dispatch (1,469/1,291); Design Decision Gate ADR-draft (1,567) and efficiency rules (1,482).

<b>Deterministic Analysis Output</b>

/tmp/gh-aw/ambient-context/analyze_requests.py produced request-analysis.json and request-analysis.md. Request sizes are 28.7K–36.2K chars. Where proxy token usage was readable, Auto-Triage billed 19,520 input tokens for 28,669 chars and Issue Monster 13,099 input tokens for 36,177 chars. A prompt.txt cross-check was not available in the downloaded artifacts.

Recommendations by Category

Workflow Markdown

• Issue Monster (issue-monster.md, 41.6K chars): extract issue-list review instructions into a deterministic steps: output (compact prioritized JSON) and keep only decision rules inline. Keep the cut under 40% of the file. Needs manual review.
• Copilot Agent PR Analysis: load the report template on demand. Safe.
• Auto-Triage Issues: replace raw gh aw command text with agentic-workflows MCP tool calls. cli-proxy and GitHub toolsets are already configured on all four workflows, except copilot-agent-analysis.md, which uses github: mode: local with no explicit toolsets; add them. Safe.

Skills

• Create on-demand skill blocks for the report and ADR templates (Copilot Agent PR Analysis, Design Decision Gate). No SKILL.md references appear in these requests, so the saving is the full template size per run.

Agents

• No inline agents were found in the sampled requests, so no agent-pruning recommendation. The ~11–15K-char system prompt is fixed overhead and needs a platform-level change.

References

• §39302459462
• §39302791923
• §39302952246

Generated by 🌫️ Daily Ambient Context Optimizer · copilot · auto · 33.3 AIC · ⌖ 0.843 AIC · ⊞ 12.6K · ◷

• expires on Oct 26, 2026, 11:56 AM UTC-08:00