new-agent-evolution

A skill for defining an agent-evolution experiment: a structured brief describing an objective, success checks, scope, and limits. It writes the brief only after the user agrees.

In plain words
What is it for?
Use it to prepare an optimization brief, identify test or build checks, and define candidate changes for later agent-evolution runs.
Why use it?
It turns a vague improvement idea into a measurable plan before any files are changed, reducing unclear goals and unapproved work.

Skill for Claude CodeCodex

Part of the agent-evolutions plugin — 3 skills shipped together

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 skills/fmind/agent-evolutions/new-agent-evolution
Any agent
npx skills add fmind/agent-evolutions --skill new-agent-evolution
Clone the repo
git clone --depth 1 https://github.com/fmind/agent-evolutions

Made for: Claude Code, Codex.

Or install agent-evolutions, the plugin that ships this one along with the rest of its 3 skills.

Per session 74 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 1,003 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 $0.00074 $0.01003
Opus 5 $0.00037 $0.00502
Sonnet 5 $0.00015 $0.00201
Haiku 4.5 $0.00007 $0.00100

Measured 2d ago against content hash 70c66db1e018, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

new-agent-evolution 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 2d ago.

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.

skills/new-agent-evolution/SKILL.md · 48 lines

How it starts

The opening of the file, as written. The whole thing — 48 lines — stays where its author put it; the contents beside it link to each section on GitHub.

new-agent-evolution

Capture a verifiable optimization brief for <title>. Writes nothing until the user explicitly agrees.

0. Pre-flight

Resolve the argument. If it matches an existing .agents/evolutions/<id>-*/ (pure integer id or slug fragment), stop with "Evolution <id> already exists. Run /run-agent-evolution <id> (or /apply-agent-evolution <id> if the winner is set)." This skill only seeds new evolutions.

Read-only setup: AGENTS.md, README.md, the user ask, any spec the ask points at. Identify project test/build/lint commands — candidate gates. Confirm the cwd is a git repo (git rev-parse --is-inside-work-tree) — variants default to git worktrees.

1. Dialog first

First-turn output is chat-only: clarifying questions and a draft brief inline. Don't create files until the user explicitly agrees ("looks good", "ship it", edits incorporated).

Five elements. Propose concrete answers; ask only when irreducible (cap at 5 questions).

  • Objective. One sentence — "Optimize X for Y." Reject vague answers; insist on a measurable Y.
  • Gates. Hard pass/fail shell commands (exit 0 = pass). 2–4 typical (pnpm typecheck, pnpm test, cargo build, custom verifiers). A variant failing any gate is excluded from ranking.
  • Rubric. Numeric ranking dimensions (≥ 1, typically 1–3). Each: direction (minimize | maximize), cmd, optional extract (regex with one capture group, or wall_clock_seconds). Composite is the rank-normalized weighted mean across axes, recomputed on read.
  • Scope. kind: worktree (default — git worktree add HEAD) or files (copy listed paths). For worktree, list include informationally so apply knows what to diff. For files, include is required. exclude blocks paths apply must not touch.
  • Budget. max_variants (total), parallel (concurrent per generation; sets the learning cadence). Optional max_minutes, plateau_generations (early stop on no-improvement). Default parallel = ceil(max_variants / 5) so the loop runs ~5 generations.

Read the full file on GitHub · 48 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. 2d ago First seen · 48 lines · 74 tokens per session scan A 70c66db1e018

Subscribe to this mod's changes

new-agent-evolution is a skill published in the GitHub repository fmind/agent-evolutions (1 stars, last pushed 3mo ago), licensed MIT. It adds 74 tokens to every session and 1,003 once invoked, about $0.0004 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.