[ambient-context] Daily Ambient Context Optimizer - 2026-07-19

[github-actions]

Executive Summary

• 4 runs sampled across 4 distinct workflows (Daily Cache Strategy Analyzer, PR Sous Chef, Agent Persona Explorer, Changeset Generator)
• Median first-request size: 22,047 chars | P95: 24,031 chars
• All 4 runs lacked API proxy event logs; prompt.txt used as request source
• Highest-leverage finding: daily-cache-strategy-analyzer.md has 37 headings and 16.2% duplicate lines — high structural fragmentation inflating every run. It also lacks gh-proxy and cli-proxy. pr-sous-chef.md is the largest workflow source (28,658 chars) with a 5,715-char inline nudge catalog that embeds full shell examples and is loaded on every run.

Highest-Leverage Changes

1. [workflow-md] Consolidate fragmented phases in daily-cache-strategy-analyzer.md — 37 headings / 16.2% dup-line ratio. Merge parallel sub-sections (2.1/2.2/2.3 into one detection block) and deduplicate repeated table templates. Est. ~2,500 char saving per run. (high)
2. [workflow-md] Enable gh-proxy + cli-proxy in daily-cache-strategy-analyzer.md — both are absent, meaning every gh CLI call goes through the slower unauthenticated path. Add tools: { github: { mode: gh-proxy }, cli-proxy: true }. (high, safe immediately)
3. [workflow-md] Trim the ## Required nudges section in pr-sous-chef.md — at 5,715 chars, this is the largest single section across all samples. Shell examples with multi-checkout notes (~800 chars) and step-by-step push instructions are repeated verbatim from the system preamble. Summarize to bullet references; keep the make fmt + add_comment essentials only. Est. ~2,000 char saving. (high)
4. [workflow-md] Reduce ## Token efficiency rules in pr-sous-chef.md — 2,262 chars of 10-point instructions, several of which restate safeoutputs guidance already in the system preamble (rules 10–11 duplicate the safe-output terminal guarantee). Remove redundant rules; keep only workflow-specific ones (rules 1–9). Est. ~600 char saving. (medium)
5. [workflow-md] Collapse parallel detection sub-sections in agent-persona-explorer.md — 25 headings with 9.2% dup-line ratio. Phase 2/3 sub-sections follow the same template. Merge into a single structured block. Est. ~1,000 char saving. (medium)
6. [workflow-md] Add gh-proxy to changeset.md — cli-proxy: true is set but tools.github.mode: gh-proxy is absent, leaving GitHub API calls unoptimized. (low, safe immediately)

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	22,047
P95 chars	24,031
Largest sampled request	24,031 (PR Sous Chef, run 29693331475)
Merged optimizer PRs (7d)	0
Closed optimizer PRs (7d)	0
Optimizer PR close-rate (7d)	N/A (<3 settled PRs)

Per-Run First-Request Metrics

Run	Workflow	Chars	Lines	Headings	Dup-line%	Input tokens
29699070953	Daily Cache Strategy Analyzer	22,047	491	37	16.2%	760,393
29693331475	PR Sous Chef	24,031	282	11	4.5%	94,700
29693386293	Agent Persona Explorer	20,226	361	25	9.2%	68,770
29701390965	Changeset Generator	18,781	320	18	9.3%	43,495

Note: request_source = prompt.txt for all runs — no API proxy event logs were present in the downloaded artifacts.

Repeated Ambient Context Signals

Cross-run repeated fragments (all 4 runs):

• Immutable security policy block (~4 lines, ~400 chars) — system-injected, not reducible
• Temporary files instruction — system-injected

Per-run internal duplication:

• daily-cache-strategy-analyzer.md: table row templates (| Workflow | Miss Streak | ... |) appear twice; for run_dir in /tmp/gh-aw/aw-mcp/logs/run-*/ shell pattern repeated twice; parallel sub-sections 2.1/2.2/2.3 share identical structural templates (~300 chars each)
• agent-persona-explorer.md: Phase 2 and Phase 3 sub-sections follow the same 5-item action pattern duplicated verbatim
• pr-sous-chef.md: Safe-output push instructions (~800 chars) repeat content from the system <safe-output-tools> preamble already injected at runtime

Safe-outputs block size: ~3,466 chars injected per run (14–19% of total prompt). This is system-managed and not reducible from workflow markdown.

Deterministic Analysis Output

Script: /tmp/gh-aw/ambient-context/analyze_requests.py

Key findings:

• Average heading count: 22.75 (high; ideal ≤ 12 for focused workflows)
• Average dup-line ratio: 9.9% across the sample
• No inline agents (## agent:) or imported skill refs (SKILL.md) found in any sampled prompt — skill loading is either absent or already compiled out
• daily-cache-strategy-analyzer.md has the highest fragmentation (37 headings) with the highest dup ratio (16.2%) despite having 760K input tokens — the token cost is dominated by tool-output accumulation across turns, not just the initial prompt
• pr-sous-chef.md has the largest workflow source file (28,658 chars) but relatively clean dup ratio (4.5%)
• Code fences: 4–13 per prompt; daily-cache-strategy-analyzer.md leads with 13 (each fence adds parse overhead)

Recommendations by Category

Workflow Markdown

1. daily-cache-strategy-analyzer.md — consolidate detection phases (high impact, needs review): Merge sections 2.1, 2.2, 2.3 into a single ## Phase 2: Detect Cache Issues block. Remove duplicate table templates. Target: ≤20 headings. Est. saving ~2,500 chars.
2. daily-cache-strategy-analyzer.md — add gh-proxy and cli-proxy (high impact, safe immediately): The workflow uses agenticworkflows logs but lacks proxy settings. Add under tools:: github: { mode: gh-proxy } and cli-proxy: true.
3. pr-sous-chef.md — trim ## Required nudges section (high impact, needs review): Condense the 5,715-char section by removing inline shell examples that duplicate system preamble guidance. Keep step headings and essential constraints only.
4. pr-sous-chef.md — remove redundant safe-output rules from ## Token efficiency rules (medium impact, safe immediately): Rules 10–11 restate the system-level terminal guarantee already injected at runtime. Remove them and renumber the section.
5. agent-persona-explorer.md — deduplicate phase templates (medium impact, needs review): Phase 2 and Phase 3 use the same 5-step action template. Consolidate into a shared pattern reference.
6. changeset.md — add tools.github.mode: gh-proxy (low impact, safe immediately): cli-proxy: true is set; add github: { mode: gh-proxy } to complete proxy coverage.

Skills

No skill imports were detected in any sampled prompt. Skill loading appears to be either absent or already well-scoped. No recommendations at this time.

Agents

No inline agents (## agent:) were found in any sampled prompt. No inline agent recommendations.

References

• §29699070953 — Daily Cache Strategy Analyzer (760,393 input tokens)
• §29693331475 — PR Sous Chef (94,700 input tokens)
• §29693386293 — Agent Persona Explorer (68,770 input tokens)

Generated by 🌫️ Daily Ambient Context Optimizer · 125.2 AIC · ⌖ 14.9 AIC · ⊞ 7.8K · ◷

• expires on Jul 26, 2026, 12:19 PM UTC-08:00