Add strategy system to autoloop; ship AlphaEvolve as the first playbook

[Copilot]

Programs can now opt into specialized per-iteration playbooks that supersede the default analyze/propose/accept/reject loop. Ships AlphaEvolve (MAP-Elites + islands + four operators) as the first concrete strategy, suited to evolving a self-contained artifact toward a scalar fitness.

Opt-in mechanism (agent-prompt only)

.github/workflows/autoloop.md gets a ### Strategy discovery step at the top of ## Iteration Loop. The agent reads program.md's ## Evolution Strategy section; if it points to a strategy file, the agent follows that file literally (state read, branching, CI gating still apply). If absent, behaviour is unchanged — no regression for coverage-style programs. No scheduler changes.

Reusable template — .autoloop/strategies/alphaevolve/

• strategy.md — runtime playbook with <CUSTOMIZE: …> markers: 8-step loop (load → pick operator → pick parents → apply → implement → evaluate → update population → fold to default), operator weights with deterministic rut-breaking overrides (3-reject → forced exploration, 5-reject → forced migration), MAP-Elites feature dimensions, population schema with strikethrough-eviction.
• CUSTOMIZE.md — creator-time guide (when AlphaEvolve fits, copy steps, marker-by-marker guidance, worked example). Not copied into programs.
• prompts/mutation.md, prompts/crossover.md — operator framing templates.

Proof-of-concept — .autoloop/programs/tsb-perf-evolve/

Targets Series.sortValues in src/core/series.ts. Fitness = tsb_mean_ms / pandas_mean_ms (lower is better; <1.0 means tsb wins).

• program.md with the resolved ## Evolution Strategy pointer block and explicit validity invariants (tests pass, signature unchanged, strict TS, NaN semantics preserved).
• code/{config.yaml, benchmark.ts, benchmark.py, README.md} — fixed evaluator: 100k float Series with 5% NaN, 5 warmup + 50 measured iterations on each side, JSON output.
• strategy/alphaevolve.md + strategy/prompts/{mutation,crossover}.md — fully customized: 5 islands (comparison-pairs, indirect typed-array, packed typed-array, radix/non-comparison, hybrid); feature dimensions = storage × algorithm-class; vocabulary and domain knowledge (Bun JIT monomorphization, typed-array allocation costs, NaN partitioning, pandas/NumPy baseline) written in. grep -R "<CUSTOMIZE" .autoloop/programs/tsb-perf-evolve/ is empty.

Pointer block authors copy into program.md

## Evolution Strategy

This program uses the **AlphaEvolve** strategy. On every iteration, read `strategy/alphaevolve.md` and follow it literally — it supersedes the generic analyze/accept/reject steps in the default autoloop loop.

Out of scope

Strategy-picker UI/CLI, additional strategies beyond AlphaEvolve, scheduler-level strategy awareness — all explicitly deferred per the issue.

Warning

Firewall rules blocked me from connecting to one or more addresses (expand for details)

I tried to connect to the following addresses, but was blocked by firewall rules:

• https://api.github.com/repos/githubnext/tsessebe/issues/181
  • Triggering command: /usr/bin/curl curl -s REDACTED (http block)
  • Triggering command: /usr/bin/curl curl -sL -H Accept: application/vnd.github.raw+json REDACTED (http block)

If you need me to access, download, or install something from one of these locations, you can either:

• Configure Actions setup steps to set up my environment, which run before the firewall is enabled
• Add the appropriate URLs or hosts to the custom allowlist in this repository's Copilot coding agent settings (admins only)