Add sub_agent_strategy A/B experiment to smoke-gemini workflow

[Copilot]

This updates smoke-gemini to run as an explicit A/B experiment comparing the current single-agent smoke flow vs a sub-agent decomposition strategy. The goal is to measure token-efficiency impact while preserving reliability guardrails.

• Experiment frontmatter

  • Added experiments.sub_agent_strategy in rich-object form with:
    • variants: single_agent, sub_agents
    • primary metric: effective_tokens
    • secondary metrics: run_duration_seconds, success_rate
    • guardrail: success_rate >= 0.95
    • sampling/weights/analysis metadata (min_samples, weight, start_date, analysis_type, tags)

• Prompt branching by variant

  • Replaced the single static requirements block with explicit conditional branches:
    • single_agent: keep baseline sequential execution in the main agent
    • sub_agents: instruct launching 5 parallel background task sub-agents (one per smoke check), then aggregate via read_agent
  • Kept shared output/reporting behavior outside variant blocks.

• Compiled workflow artifact

  • Regenerated smoke-gemini.lock.yml from updated source to include experiment selection/runtime wiring and branch-specific prompt content.

experiments:
  sub_agent_strategy:
    variants: [single_agent, sub_agents]
    metric: effective_tokens
    secondary_metrics: [run_duration_seconds, success_rate]
    guardrail_metrics:
      - name: success_rate
        threshold: ">=0.95"

[Copilot]

Pull request overview

Adds an explicit A/B experiment (sub_agent_strategy) to the smoke-gemini agentic workflow to compare the baseline single-agent smoke flow against a sub-agent decomposition strategy, with metrics/guardrails wired through the compiled lock workflow.

Changes:

• Added experiments.sub_agent_strategy frontmatter (variants/metrics/guardrails/weights/analysis metadata).
• Introduced variant-conditional prompt branches for single_agent vs sub_agents.
• Regenerated smoke-gemini.lock.yml to include experiment selection, state restore/upload/push, and updated runtime wiring.

Show a summary per file

File	Description
.github/workflows/smoke-gemini.md	Defines the experiment and branches the smoke instructions by variant.
.github/workflows/smoke-gemini.lock.yml	Compiled workflow updates to pick variants, persist experiment state, and render the variant-specific prompt.

Copilot's findings

Tip

Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

• Files reviewed: 2/2 changed files
• Comments generated: 3