[ambient-context] Daily Ambient Context Optimizer - 2026-07-28

[github-actions]

Executive Summary

• Runs sampled: 4 (Daily Cache Strategy Analyzer, Agent Persona Explorer, Smoke Copilot, Linter Miner) — all completed, distinct workflows, breadth-first selection.
• Workflows covered: 4 distinct .lock.yml targets.
• Median first-request size: ~17,032 chars · P95: 19,331 chars.
• Conclusion: No blocked-file recommendations were needed (3 workflows from a recent closed optimizer PR were excluded per the dedup guard). First requests are moderate-sized but show recurring duplicate table/heading fragments from shared/reporting.md and shared/mcp/serena-go.md imports repeated across workflows, plus a very high token_usage outlier (Smoke Copilot, 1.24M tokens vs a 14.7KB first request) suggesting continuation-turn cost rather than first-request bloat.
• Note: prompt.txt was the only usable request source in this run's downloaded artifacts (no sandbox/firewall or copilot-session-state event logs were present); used per the fallback rule.

Highest-Leverage Changes

1. Deduplicate the ## Report Structure Guidelines / ## Workflow Run References blocks from shared/reporting.md — appears in 3+ sampled first requests; move common report boilerplate into a single condensed shared fragment.
2. Trim shared/mcp/serena-go.md "Serena Go Code Analysis" section — repeated verbatim in 2 sampled requests; consolidate with shared/mcp/serena.md to avoid double-loading similar tool guidance.
3. Reduce daily-cache-strategy-analyzer.md's duplicate table/example rows (duplicate_line_ratio 0.15, duplicate_paragraph_ratio 0.14 — highest of the sample) — collapse repeated | Workflow | Miss Streak | ... | example rows into one illustrative row.
4. Investigate Smoke Copilot's outsized token_usage (1.24M) vs its 14.7KB first request — likely a continuation/retry loop, not ambient context; recommend adding a max-continuations or turn-budget guardrail rather than trimming the prompt.
5. All 4 sampled workflows already have tools.github.mode: gh-proxy and tools.cli-proxy: true enabled — no proxy-enablement recommendation needed this cycle.

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	17,032
P95 chars	19,331
Largest sampled request	Daily Cache Strategy Analyzer (21,156 chars)
Merged optimizer PRs (7d)	0
Closed optimizer PRs (7d)	1
Optimizer PR close-rate (7d)	n/a (< 3 settled PRs; auto-pause not triggered)

Per-Run First-Request Metrics

Run	Workflow	Status	Chars	Headings	Code Fences	Details	Dup Line Ratio	Dup Para Ratio
§30389296712	Daily Cache Strategy Analyzer	success	21,156	33	22	6	0.155	0.137
§30376833278	Agent Persona Explorer	success	19,331	24	10	5	0.041	0.012
§30377662004	Smoke Copilot	failure	14,732	10	2	1	0.007	0.000
run-30384667352	Linter Miner	failure	14,216	14	2	0	0.056	0.131

Repeated Ambient Context Signals

• ## Report Structure Guidelines / ## Workflow Run References (from shared/reporting.md) recur across 3 sampled requests.
• ## Serena Go Code Analysis (from shared/mcp/serena-go.md) recurs across 2 sampled requests; a near-duplicate ## Serena Code Analysis exists in shared/mcp/serena.md.
• Daily Cache Strategy Analyzer's example table rows (| Workflow | Miss Streak | ... |) are duplicated verbatim as both header and filled-example, inflating duplicate-line/paragraph ratios well above the other 3 samples.
• No inline ## agent:, ## linter:, or ## skill: blocks were found in any sampled request — inline-agent bloat is not currently a driver for this sample.

Deterministic Analysis Output

Script: /tmp/gh-aw/ambient-context/analyze_requests.py (stdlib only) → request-analysis.json / .md.

• Aggregate: 4 sampled runs, 4 distinct workflows, median 17,032 chars, P95 19,331 chars.
• most_common_large_section_headings: Report Structure Guidelines (3x), Workflow Run References (3x), Important Guidelines (2x), Success Criteria (2x), Serena Go/Code Analysis (2x), Analysis Constraints (2x).
• Highest duplicate-line ratio: Daily Cache Strategy Analyzer (0.155) — driven by repeated example table rows.
• char_token_ratio could not be computed for any sample (no per-request input_tokens entry available in token_usage.jsonl for the first request specifically); token_usage totals were used instead for cost context.
• Audit enrichment was run for the single most expensive sampled run (Daily Cache Strategy Analyzer, agenticworkflows audit --run_id 30389296712), confirming no elevated GitHub API quota consumption tied to the request itself.

Recommendations by Category

Workflow Markdown

• daily-cache-strategy-analyzer.md — collapse duplicated example/header table rows in the "Phase 2: Detect Cache Issues" / "Phase 3: Cross-Reference" sections (evidence: dup_line_ratio 0.155, dup_para_ratio 0.137, both highest in sample). Expected impact: medium. Safe to apply immediately with the CI-validation checklist above.
• smoke-copilot.md — do not shrink the first request (already lean at 14.7KB); instead add/verify a continuation-turn or max-turns guardrail given token_usage of 1.24M against a small first request. Expected impact: high (cost), but needs manual review (behavioral, not textual, change).

Skills

• Consolidate shared/mcp/serena.md and shared/mcp/serena-go.md "Code Analysis" sections into one parameterized fragment referenced by both, reducing duplicate content pulled into every workflow that imports either. Evidence: identical heading pattern appears in 2/4 samples. Expected impact: medium. Needs manual review (shared file, used by many workflows).
• Trim shared/reporting.md's "Report Structure Guidelines" / "Workflow Run References" boilerplate to the minimum needed per consuming workflow, or move some of it to deterministic steps: output. Evidence: appears in 3/4 samples' longest-sections list. Expected impact: medium, safe to apply immediately with the checklist.

Agents

• No inline ## agent: blocks were observed in the sampled first requests (inline_agent_count = 0 for all 4), so no inline-agent-pruning recommendation is warranted this cycle. Continue monitoring in future sampling windows.

References

• §30389296712
• §30376833278
• §30377662004

Generated by 🌫️ Daily Ambient Context Optimizer · sonnet50 · 77 AIC · ⌖ 8.56 AIC · ⊞ 10.7K · ◷

• expires on Aug 4, 2026, 12:23 PM UTC-08:00