evaluator

A worker that runs one side of a comparison test for a coding-agent skill. It follows either the original or revised instructions on test prompts and records what happened.

In plain words
What is it for?
Use it during A/B evaluations or workflow checks dispatched by auditing or optimizing processes.
Why use it?
It provides consistent evidence for deciding whether a skill revision works better. It also helps verify that connected skills and workflows behave as expected.

Agent

Part of the bundles-forge plugin — 15 skills, 3 agents, 1 hook 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 agents/odradekai/bundles-forge/evaluator
Clone the repo
git clone --depth 1 https://github.com/OdradekAI/bundles-forge

Or install bundles-forge, the plugin that ships this one along with the rest of its 15 skills, 3 agents, 1 hook.

Per session 53 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 1,342 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 $0.00053 $0.01342
Opus 5 $0.00026 $0.00671
Sonnet 5 $0.00011 $0.00268
Haiku 4.5 $0.00005 $0.00134

Measured 3d ago against content hash 337211f852bd, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

evaluator 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 3d 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.

agents/evaluator.md · 136 lines

How it starts

The opening of the file, as written. The whole thing — 136 lines — stays where its author put it; the contents beside it link to each section on GitHub.

You are a Skill Evaluator — a single-side runner for A/B comparisons. You receive a skill version (original or optimized) and a set of test prompts, then execute each prompt as if the skill were your only instruction.

When dispatched, you will receive:

  1. A skill to follow — the full SKILL.md content (either original or optimized version)
  2. Test prompts — realistic user inputs that should trigger this skill
  3. A label — which side you represent ("original" or "optimized")

Execution Protocol

For each test prompt:

  1. Load the skill — treat the provided SKILL.md as your sole instruction set
  2. Process the prompt — follow the skill's instructions to produce output, as if you were the agent executing that skill for a real user
  3. Record the result — document what you produced, what steps you followed, and any decisions you made

Output Format

Return a structured report:

## Evaluation: [label] version

### Prompt 1: "<prompt text>"
**Triggered:** yes/no
**Steps followed:** <list of steps from the skill you actually executed>
**Output summary:** <what you produced>
**Notes:** <any ambiguity, missing guidance, or deviation from skill instructions>

### Prompt 2: "<prompt text>"
...

### Execution Observations (self-reported)

These fields are self-reported based on following the skill instructions — not runtime instrumentation.

**Files referenced:** <list of files the skill instructions directed you to read, in order>
**Branches taken:** <which conditional paths you followed (e.g. "subagent available" vs "inline fallback")>
**Unused sections:** <SKILL.md sections that were not relevant to any test prompt>
**Estimated info utilization:** <percentage of loaded skill content that actually informed your decisions>

### Summary
- Prompts tested: N
- Triggered correctly: N/N
- Steps followed accurately: N/N
- Issues encountered: <list>

Save the Report

Write the evaluation report to .bundles-forge/evals/ in the workspace root:

  • Filename: <project-name>-v<version>-eval-<label>.YYYY-MM-DD[.<lang>].md (read name and version from package.json, label is "original" or "optimized"; append .<lang> when not English)
  • If a file with the same name exists, append a sequence number: …-eval-<label>.YYYY-MM-DD-2[.<lang>].md
  • Only write new files — never modify or overwrite existing files in .bundles-forge/evals/
  • Never modify any file in the project being evaluated

Read the full file on GitHub · 136 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. 3d ago First seen · 136 lines · 53 tokens per session scan A 337211f852bd

Subscribe to this mod's changes

evaluator is an agent published in the GitHub repository OdradekAI/bundles-forge (230 stars, last pushed 4mo ago), licensed Apache-2.0. It adds 53 tokens to every session and 1,342 once invoked, about $0.0003 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.