evaluate

An evaluation step that runs an agent candidate on several tasks and reports its average score, uncertainty, and how many tasks were actually measured.

In plain words
What is it for?
Use it to compare candidate agents, inspect their scores, and check whether an apparent improvement is reliable.
Why use it?
A single score can hide inconsistent results or failed runs. This gives the score’s variability and its true measurement coverage.

Skill for Claude CodeCodex

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/skillberry-ai/cap-evolve/evaluate
Any agent
npx skills add skillberry-ai/cap-evolve --skill evaluate
Clone the repo
git clone --depth 1 https://github.com/skillberry-ai/cap-evolve

Made for: Claude Code, Codex.

Per session 84 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 2,254 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 $0.00084 $0.02254
Opus 5 $0.00042 $0.01127
Sonnet 5 $0.00017 $0.00451
Haiku 4.5 $0.00008 $0.00225

Measured 2d ago against content hash 3366d08a36e8, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

evaluate 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.

The scan reads SKILL.md. This mod also ships 4 executable files (scripts/_bootstrap.py, scripts/abstract.py, scripts/check.py, …), listed below but not scanned — reading those needs a real analyzer, not pattern matching.

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/phases/evaluate/SKILL.md · 137 lines

How it starts

The opening of the file, as written. The whole thing — 137 lines — stays where its author put it; the contents beside it link to each section on GitHub.

evaluate — honest, multi-trial scoring

Turns a candidate into a score you can trust. A reward number is only as honest as the variance around it and as the denominator under it: agents are stochastic, and infrastructure fails. evaluate produces a point estimate, its uncertainty, and the count of tasks that actually produced a measurement. The math lives in cap_evolve.stats; this skill drives the adapter and aggregates.

What it produces

A SplitResult (core/cap_evolve/loop.py:59-115):

  • reward — the mean, over the tasks that were scored, of each task's mean over its valid trials (harness.py:405-414, loop.py:127-131). Not the mean over every task in the split — see the next section.
  • n_tasks / n_scored (and coverage = n_scored/n_tasks, loop.py:79-84) — the honest denominator. Read these on every result, never reward alone.
  • stderr — the combined SE of that reported mean: between-task variance (do different tasks agree?) folded with within-task trial variance (is the agent consistent on a fixed task?), stats.combined_stderr. This is what the report prints and what the gate's significant mode consumes — not what the default gate reads; see "What the gate actually consumes".
  • pass_k — when trials > 1, the estimated probability that all k i.i.d. trials pass (reliability). Also pass_at_k — at least one of k passes (capability). Opposite questions; see references/concepts.md.
  • per-task scores + feedback, and the rollout files diagnose reads: <run-dir>/rollouts/<split>/<task>__<tag>__t<k>.json (harness.py:334).

A crashed rollout is missing data, not a zero

The single largest honesty mechanism in the eval path. Two ways a trial produces no measurement:

  • the runner errored (rollout.error set) — the target never ran;
  • the rollout succeeded and the scorer could not grade it (crashed grading harness, missing report file). There is no rollout.error, so adapters must flag it by setting Score.raw["errored"] (harness.py:308-323). An adapter that doesn't is how a scorer outage becomes a real 0.0.

Read the full file on GitHub · 137 lines

Files

What ships with it

6 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.

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 · 137 lines · 84 tokens per session scan A 3366d08a36e8

Subscribe to this mod's changes

evaluate is a skill published in the GitHub repository skillberry-ai/cap-evolve (47 stars, last pushed 2d ago), licensed Apache-2.0. It adds 84 tokens to every session and 2,254 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 2026-08-30.