[ambient-context] Daily Ambient Context Optimizer - 2026-06-13

[github-actions]

Executive Summary

• 4 runs sampled across 4 distinct workflows (PR Code Quality Reviewer, Daily Testify Uber Super Expert, Daily Safe Output Integrator, PR Triage Agent)
• Median first-request size: 25,286 chars / ~20k tokens; P95: 30,318 chars
• Source note: event-logs.jsonl absent for all sampled runs; events.jsonl user.message data empty; first-request text sourced from aw-prompts/prompt.txt
• Highest-leverage finding: Daily workflow bodies carry 40–50 verbatim code-fence blocks (8,000–11,000 chars of bash examples and templates) that can be replaced with brief references. Combined with the shared system preamble (~6.8–7.5k chars constant overhead), two budget items account for 50–70% of first-request size across all sampled workflows.

Highest-Leverage Changes

1. Trim verbatim bash code blocks in daily workflow bodies — daily-testify-uber-super-expert.md and daily-safe-output-integrator.md embed 40 and 50 code fences respectively (~8,464 and ~11,129 chars). Replace with compact one-line references. Estimated savings: 4,000–6,000 chars per run. (high impact, safe immediately)
2. Slim down shared Serena + reporting imports — go-source-analysis.md pulls in shared/mcp/serena.md (2,561 chars) and shared/reporting.md (2,231 chars). Shorten by replacing verbose capability listings with a single-sentence activation note. Estimated savings: ~3,500 chars on any workflow importing go-source-analysis.md. (high impact, needs review)
3. Remove redundant Safe Output Usage section from workflow bodies — pr-code-quality-reviewer.md has a 746-char ### Safe Output Usage section duplicating the 2,607-char system-preamble block. Remove it. Estimated savings: ~750 chars. (medium impact, safe immediately)
4. Collapse overly-granular heading structures — daily-testify-uber-super-expert.md has 63 headings + 166 list items; pr-triage-agent.md has 46 headings + 214 list items. Merge fine-grained sub-step headings into terse prose. Estimated savings: ~1,800 chars. (medium impact, safe immediately)
5. Remove repeated YAML frontmatter examples in Daily Safe Output Integrator — Phase 2 embeds 14 copies of test-workflow frontmatter blocks, driving a 2,592-char section. Replace with a find/grep instruction. Estimated savings: ~1,200 chars. (medium impact, safe immediately)
6. Add tools.github.mode: gh-proxy to PR Code Quality Reviewer — The only sampled workflow without gh-proxy; the other 3 all have it. (low impact, safe immediately)

Key Metrics

Metric	Value
Sampled runs	4
Distinct workflows	4
Median chars	25,286
P95 chars	30,318
Largest sampled request	Daily Safe Output Integrator (30,318 chars / 22,241 tokens)
Highest-AIC run	PR Code Quality Reviewer (522.3 AIC, 16,660 first-req tokens)
Shared system preamble (all runs)	~6,800–7,500 chars
safe-outputs system block (all runs)	2,607 chars (constant)

Per-Run First-Request Metrics

Run ID	Workflow	Chars	Tokens	Char/Tok	AIC	Headings	Lists	Fences	Dup-Line Ratio	Conclusion
27475470287	PR Code Quality Reviewer	12,364	16,660	0.742	522.3	15	64	6	0.000	success
27475076364	Daily Testify Uber Super Expert	27,185	21,017	1.293	354.8	63	166	40	0.036	success
27476058710	Daily Safe Output Integrator	30,318	22,241	1.363	289.0	68	151	50	0.184	failure
27475627987	PR Triage Agent	23,387	20,103	1.163	266.9	46	214	10	0.010	success

Source: All first-request texts from aw-prompts/prompt.txt (canonical event-logs.jsonl absent; events.jsonl user.message empty in all 4 sampled runs).

Repeated Ambient Context Signals

• System preamble (~6.8–7.5k chars per run) is identical across all 4 runs: security policy, safe-outputs guardrail block (2,607 chars), MCP CLI docs (393 chars), github-context (354 chars), gh-cli instructions (421 chars).
• Reporting guidelines (2,231 chars) from shared/reporting.md appear in 3 of 4 sampled runs.
• Serena MCP docs (2,561 chars) from shared/mcp/serena.md appear in workflows importing shared/go-source-analysis.md.
• Daily Safe Output Integrator duplicate_line_ratio = 0.184 — highest of all sampled runs; driven by 14 repeated YAML frontmatter blocks in Phase 2 and repeated Python loop patterns across Phase 2 and Phase 5.
• PR Code Quality Reviewer "Safe Output Usage" section (746 chars) duplicates system-preamble safe-outputs guidance already injected at runtime.

Deterministic Analysis Output

Script: analyze_requests.py (stdlib only), output: request-analysis.json + request-analysis.md

Most structurally dense: Daily Safe Output Integrator — 68 headings, 50 code fences, 0.184 dup-line ratio
Highest char-to-token ratio: Daily Safe Output Integrator (1.363) — prose-heavy vs PR Code Quality Reviewer (0.742) which has denser structured XML tags in system section
Largest single section in every run: (preamble) / system section (5,339–7,498 chars) — confirms constant system overhead dominates across all workflows
Paragraph-level duplication: dup_para_ratio = 0.0 in all 4 runs; line-level duplication is internal to Daily Safe Output Integrator only
Sub-agent optimizer: 2/2 runs completed; 6 total opportunities returned, deduplicated into the 6 recommendations above

Recommendations by Category

Workflow Markdown

1. daily-testify-uber-super-expert.md — Replace 40 code fences (8,464 chars total) with compact references; remove or shorten the verbatim markdown issue template (~1,000 chars) embedded in the prompt. High impact, safe immediately.
2. daily-safe-output-integrator.md — Replace 14 repeated test-workflow frontmatter examples in Phase 2 with a single find/grep instruction. Shorten Phase 5 Go compiler test examples (~2,710 chars). Medium impact, safe immediately.
3. pr-code-quality-reviewer.md — Remove the ### Safe Output Usage section (746 chars) that duplicates the runtime-injected safe-outputs block. Add tools.github.mode: gh-proxy to enable the MCP GitHub tool path (only sampled workflow missing it). Medium + low impact, both safe immediately.
4. pr-triage-agent.md — Merge Phase 1–6 sub-step bullet lists (214 list items) into terse imperative prose; remove minute-by-minute time budget annotations. Medium impact, safe immediately.

Skills

5. shared/mcp/serena.md (2,561 chars) and shared/reporting.md (2,231 chars)** — Imported transitively via shared/go-source-analysis.md, these add ~4,700 chars to every workflow that uses Serena. Slim each down to a short activation note with a capability reference link. Consider making them lazy/optional imports. High impact, needs manual review before merge.

Agents

6. No ## agent: inline-agent blocks found in any sampled run — no inline-agent bloat detected today. If the grumpy-coder sub-agent in PR Code Quality Reviewer is not consistently triggered, consider making its dispatch conditional to avoid sub-agent startup overhead on PRs that don't need it. Low impact, needs review.

References

• §27475470287 — PR Code Quality Reviewer (522 AIC)
• §27475076364 — Daily Testify Uber Super Expert (355 AIC)
• §27476058710 — Daily Safe Output Integrator (289 AIC, failure)

Generated by 🌫️ Daily Ambient Context Optimizer · 1.4K AIC · ⌖ 29.5 AIC · ⊞ 21.9K · ◷

• expires on Jun 20, 2026, 12:23 PM UTC-08:00

[github-actions]

This issue is being closed as outdated. A newer issue has been created: #39279

View newer issue

---

This action was performed automatically by the Daily Ambient Context Optimizer workflow.