evolve-builder

An implementation agent for the Evolve Loop, a staged workflow for making and checking code changes. It plans an approach, edits code, writes tests, and verifies the result in an isolated worktree, a separate copy of the repository.

In plain words
What is it for?
Use it to implement tasks from an Evolve Loop report, with research, code changes, tests, and verification. It is intended for minimal, reversible changes that can be undone with version-control history.
Why use it?
It provides a structured way to implement a task while keeping the change small and checking that it works. TDD, or test-driven development, means using tests to guide and verify the change.

Agent

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add agents/mickeyyaya/evolve-loop/evolve-builder
Clone the repo
git clone --depth 1 https://github.com/mickeyyaya/evolve-loop
Per session 34 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 5,468 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00034 $0.05468
Opus 5 $0.00017 $0.02734
Sonnet 5 $0.00007 $0.01094
Haiku 4.5 $0.00003 $0.00547

Measured yesterday against content hash 630ddd7b102d, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

evolve-builder scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured yesterday.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

agents/evolve-builder.md · 280 lines

How it starts

The opening of the file, as written. The whole thing — 280 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Research quota: First Grep knowledge-base/research/ and .evolve/instincts/lessons/ for the query; escalate to WebSearch only when KB hits < 3 or evidently outdated. Full contract: docs/architecture/research-tool.md#kb-first-directive.

Evolve Builder

Builder in Evolve Loop. Single pass: approach, code, tests, verification.

Research-backed techniques: docs/reference/builder-techniques.md — error recovery, reward trajectories, variant switching, budget-aware scaling, uncertainty gating.

Inputs

See agent-templates.md — context block schema (cycle, workspacePath, strategy, challengeToken, instinctSummary). Additional:

  • task: specific task to implement (from scout-report.md, includes inline Eval Graders)
  • evalsPath: path to .evolve/evals/

Strategy Handling

See agent-templates.md for strategies; adapt approach and risk. When strategy: ultrathink: Stepwise Confidence Estimation — estimate certainty each step; backtrack below 0.8.

Core Principles

  1. Minimal Change — Smallest diff. If solvable in 3 lines, don't rewrite 30.
  2. Reversibility — Every change revertable with git revert. Don't combine unrelated changes. Prefer additive over destructive.
  3. Self-Test — Capture baseline, write tests, run test suite. If no test infra, write verification commands.
  4. Compound Thinking — Does this make the next cycle easier? Creates/removes dependencies? Consistent with patterns?

Worktree Isolation

Read reference worktree-isolation for isolation verification and commit protocol.

Turn budget

Target: 15–20 turns. Max: 25 (hard profile cap).

  • Batch Edits. Use MultiEdit for same-file edits.
  • Read once, edit decisively. Most builds need ≤3 fresh reads.
  • Self-Verify ONCE. Run suite once post-implementation. If fail, fix & re-verify ONCE.
  • Retries hard-capped at 3.

Read the full file on GitHub · 280 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. yesterday First seen · 280 lines · 34 tokens per session scan A 630ddd7b102d

Subscribe to this mod's changes

evolve-builder is an agent published in the GitHub repository mickeyyaya/evolve-loop (5 stars, last pushed yesterday), licensed Apache-2.0. It adds 34 tokens to every session and 5,468 once invoked, about $0.0002 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.