agent-optimize

agent-optimize — the free-form loop you own

The one algorithm with no deterministic subprocess and no per-iteration optimizer: you — the agent that ran intake — are the optimizer, the scheduler and the stopping rule. cap-evolve run (with orchestration_mode: agent) does check → baseline, prints a handoff, and returns. From there the search is yours, bounded by core's invariants and the free-text stop_condition. Drive the existing primitives so the run dir and dashboard stay populated as in a deterministic run (unattended: host.py/--agent-driver). A forensic run (#665) followed this loop in PROSE, not BEHAVIOR — 1 candidate/round forever, full-val paid regardless of scope, two mergeable accepts never merged — so every step below names the script that decides for you.

Shell variables used below

R="<run_dir from the agent-mode handoff>"      # e.g. .capevolve/run_20250101_120000
P="<project dir>"                              # the dir holding capevolve.yaml + adapters/
S="${CAPEVOLVE_SKILLS_DIR:?set CAPEVOLVE_SKILLS_DIR to the skills/ dir}"
A="$S/algorithms/agent-optimize/scripts"       # this skill's helpers
mkdir -p "$R/work"                             # working copies live here (RunDir does NOT create it)

Every script imports _bootstrap itself and prints JSON on stdout.

Phase 0 — understand before you optimize

Once, before any edit, and ask the user any blocking question here so the loop then runs unattended. Read PROJECT.md, capevolve.yaml, the adapter, every file under capability_path, ./guidance/<cap>/SKILL.md, optimizer/INSTRUCTIONS.md if present; understand the task, run_target output, score() reward, per-task feedback, val/test size, num_trials, gate_mode/gate_k_se, objectives/model_routing if set (sections below), edit surface.

Then let spend.py parse the free-text stop_condition rather than restating it: it prints constraints.predicates, every concrete check it extracts. If constraints.ambiguous is non-empty, ASK THE USER before the loop starts — a vague clause is reported, never guessed at.