Add A/B experiment wiring for smoke-pi sub-agent decomposition

[Copilot]

smoke-pi currently runs all five Pi smoke checks in a single agent turn. This change introduces an A/B experiment on sub_agent_decomposition so the workflow can compare the current single-agent path against a parallel sub-agent implementation using the built-in experiments runtime.

• Experiment metadata

  • Adds experiments.sub_agent_decomposition to /.github/workflows/smoke-pi.md
  • Configures:
    • variants: single_agent, parallel_sub_agents
    • primary metric: effective_token_count
    • secondary and guardrail metrics
    • sample size, weights, start date, analysis type, and tags
  • Leaves the issue field as a placeholder comment until a real tracking issue number is available

• Prompt branching

  • Replaces the single static test-requirements block with variant-specific Handlebars branches
  • parallel_sub_agents instructs the workflow to launch five background task agents, one per smoke check, then aggregate results via read_agent
  • else preserves the existing sequential single-turn behavior as the baseline

• Compiled workflow regeneration

  • Recompiles smoke-pi so /.github/workflows/smoke-pi.lock.yml includes:
    • experiment spec serialization
    • runtime variant selection wiring
    • prompt interpolation inputs for sub_agent_decomposition

Example of the new prompt split:

{{#if experiments.sub_agent_decomposition == "parallel_sub_agents"}}
Launch five parallel `task` agents using mode: "background" to execute each smoke test independently.
{{else}}
Execute the following tests sequentially in a single turn:
{{/if}}

[Copilot]

Pull request overview

Introduces an A/B experiment (sub_agent_decomposition) for the smoke-pi workflow to compare the existing single-agent smoke run against a parallel background sub-agent approach, with the compiled lock workflow regenerated to include experiments runtime wiring.

Changes:

• Added experiments.sub_agent_decomposition metadata to smoke-pi workflow frontmatter (variants, metrics, weights, analysis).
• Branched the “Test Requirements” prompt to run either sequential tests (baseline) or five parallel background task agents with aggregation via read_agent.
• Regenerated smoke-pi.lock.yml to serialize the experiment spec, select variants at runtime, persist experiment state, and pass variant inputs into prompt interpolation.

Show a summary per file

File	Description
.github/workflows/smoke-pi.md	Adds experiment spec and Handlebars branching for single-agent vs parallel sub-agent smoke execution.
.github/workflows/smoke-pi.lock.yml	Recompiled workflow to include experiment selection, state restore/push, artifact wiring, and prompt env var injection.

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