[ambient-context] Daily Ambient Context Optimizer - 2026-07-05

[github-actions]

Executive Summary

• 4 runs sampled across 4 distinct workflows from the last 24h
• Median first-request size: 17,643 chars | P95: 21,033 chars
• Largest request: Copilot Agent PR Analysis at 27,299 chars (AIC 116.9) driven by two full inline variant output templates (~3,816 chars) and duplicated jq examples
• Highest AIC run: Daily Code Metrics (245.4) adds 5,870 chars of shared viz/trend imports on every run
• All workflows already have gh-proxy and cli-proxy enabled ✓
• No blocked files; close-rate 0%; auto-pause not triggered

Highest-Leverage Changes

1. [HIGH] copilot-agent-analysis.md: Collapse A/B experiment variant templates — Both structured (~3,067 chars) and prose (~749 chars) output templates are inlined in the prompt, but only one variant runs per execution. Move inactive variant or replace both with a single compact output spec. Estimated saving: ~2,000–2,500 chars.
2. [HIGH] copilot-agent-analysis.md: Gate historical-rebuild instructions behind a conditional import — The cold-start rebuild instructions (§4.1, ~1,470 chars) are only needed when the cache is empty (rare). Use {{#runtime-import? shared/copilot-agent-analysis-history.md}} or an optional: import so they're absent on warm runs.
3. [MEDIUM] daily-code-metrics.md: Audit shared/python-dataviz.md import for inline duplication — This shared import adds 3,870 chars of matplotlib guidance. The workflow body already mentions matplotlib-specific requirements (5 occurrences). Review overlap; strip duplicated chart-type specs from whichever side is redundant.
4. [MEDIUM] copilot-agent-analysis.md: Remove jqschema/SKILL.md import — The jqschema skill (~992 chars) teaches schema discovery for unknown API shapes. Pre-fetched data is already at a known path with a *-schema.json file. This import adds no value; remove it.
5. [MEDIUM] copilot-agent-analysis.md: Deduplicate jq examples between workflow body and shared base — The workflow body has 3 jq examples and shared/copilot-pr-analysis-base.md adds 4 more against the same file path. Remove the redundant in-body examples and rely on the shared base (~300 chars saved).

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	17,643
P95 chars	21,033
Largest sampled request	Copilot Agent PR Analysis (27,299 chars)
Merged optimizer PRs (7d)	4
Closed optimizer PRs (7d)	0
Optimizer PR close-rate (7d)	0%

Per-Run First-Request Metrics

Run	Workflow	AIC	Chars	Lines	Dup%	Headings	Source
§28750739008	Copilot Agent PR Analysis	116.9	27,299	560	7.2%	45	prompt.txt
§28751029351	Daily Code Metrics	245.4	21,033	365	2.7%	30	prompt.txt
§28750490517	PR Code Quality Reviewer	88.7	14,253	221	3.1%	16	prompt.txt
§28750744185	PR Triage Agent	101.7	13,839	237	1.8%	21	prompt.txt

Note: API proxy event logs were not present in the MCP log artifacts; prompt.txt was used as the first-request source for all runs.

Repeated Ambient Context Signals

• Runtime-injected boilerplate (not author-controlled): security policy (4 lines), safe-outputs instructions, and temporary-files/env-limitations blocks appear identically in all 4 runs — these are system-injected and cannot be reduced by workflow authors
• shared/noop-reminder.md (411 chars) imported in 3/4 runs — expected and minimal
• shared/reporting.md (497 chars) imported in all 4 runs — small, no action needed
• copilot-agent-analysis.md: variant templates add ~3,816 chars for an experiment where only one branch executes; the "Important Brevity Guidelines" (360 chars) restates instructions already in the mission section
• copilot-agent-analysis.md: jq commands appear in body (3×) and shared/copilot-pr-analysis-base.md (4×) for the same data file

Deterministic Analysis Output

Script: /tmp/gh-aw/ambient-context/analyze_requests.py (stdlib only)

• copilot-agent-analysis: 45 headings, 14 code fences, 7.2% duplicate-line ratio (highest). Top section "Phase 6 discussion template" = 3,927 chars containing embedded markdown tables inside fenced code blocks.
• daily-code-metrics: 30 headings, 6 code fences. Shared imports contribute 5,870 chars (python-dataviz.md 3,870 + trends.md 2,000); "Required Charts" section = 1,217 chars.
• pr-code-quality-reviewer: 16 headings, grumpy-coder sub-agent = 1,111 chars (present and justified — haiku model). "Step 4: Write Review Comments" = 1,378 chars including a full example.
• pr-triage-agent: cleanest at 13,839 chars, 1.8% dup ratio.
• Cross-run: 10 fragments appear in all 4 runs — all are runtime-injected system context, not reducible by workflow authors.

Recommendations by Category

Workflow Markdown

1. copilot-agent-analysis.md — Collapse A/B variant templates (HIGH): Replace the two full inline output templates (structured + prose, ~3,816 chars total) with a single compact output spec. Since only one variant is active per run, the inactive template is pure waste. Consider a shared shared/copilot-agent-output-spec.md that conditionally renders the active variant only — or simply trim each template to the essential column/field list rather than full markdown examples. Safe to implement; does not change report quality for the active variant.

2. copilot-agent-analysis.md — Gate historical rebuild section (HIGH): Move §4.1 cold-start rebuild instructions (~1,470 chars) to an optional import {{#runtime-import? shared/copilot-agent-history-rebuild.md}}. The rebuild path is rare; most runs have data already in repo-memory. Needs manual review to confirm optional-import semantics don't affect the base-case flow.

3. copilot-agent-analysis.md — Remove body-level jq examples (MEDIUM): The 3 jq examples in the workflow body (copilot-prs.json queries) are already covered by shared/copilot-pr-analysis-base.md (4 examples, same file). Remove the in-body duplicates. Safe immediately.

Skills

4. copilot-agent-analysis.md — Drop jqschema/SKILL.md import (MEDIUM): Pre-fetched PR data is at a fixed path with a companion schema file; jqschema is designed for unknown API shapes. Import adds ~992 chars with no marginal value here. Safe to remove.

5. daily-code-metrics.md — Audit shared/python-dataviz.md (MEDIUM): At 3,870 chars this import is the largest shared fragment in any sampled daily workflow. Review overlap with inline matplotlib guidance in the workflow body; strip whichever side is more generic. Needs manual review to confirm no chart-type detail is lost.

Agents

No new sub-agent additions recommended. The existing grumpy-coder inline agent in pr-code-quality-reviewer.md (1,111 chars, claude-haiku-4.5) is appropriately scoped and justified.

References

• §28750739008 — Copilot Agent PR Analysis (largest request, 27,299 chars)
• §28751029351 — Daily Code Metrics (highest AIC, 245.4)
• §28750744185 — PR Triage Agent (cleanest baseline, 13,839 chars)

Generated by 🌫️ Daily Ambient Context Optimizer · 162.6 AIC · ⌖ 17 AIC · ⊞ 7.6K · ◷

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