eval-behavior

eval-behavior is a command for Claude Code from Curt-Park/autology. It costs 21 tokens per session (1,067 once invoked), scanned A, original, MIT.

A command for testing an Autology skill's behavior against predefined cases, both with and without the skill enabled.

In plain words
What is it for?
Use it to run every case in a skill's evals.json file and inspect the resulting outputs.
Why use it?
It helps compare whether the skill changes an agent's results as intended, using isolated workspaces to keep runs separate.

Command for Claude Code

Part of the autology plugin — 6 skills, 2 commands, 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 commands/curt-park/autology/eval-behavior
Clone the repo
git clone --depth 1 https://github.com/Curt-Park/autology

Made for: Claude Code.

Or install autology, the plugin that ships this one along with the rest of its 6 skills, 2 commands, 1 hook.

Wrote this? Show the measurements

A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.

agentmods badge for eval-behavior

README.md
[![agentmods](https://agentmods.dev/badge/commands/curt-park/autology/eval-behavior.svg)](https://agentmods.dev/commands/curt-park/autology/eval-behavior)
Your own site
<a href="https://agentmods.dev/commands/curt-park/autology/eval-behavior"><img src="https://agentmods.dev/badge/commands/curt-park/autology/eval-behavior.svg" alt="Measured on agentmods" height="20"></a>
Per session 21 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 1,067 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 $0.00021 $0.01067
Opus 5 $0.00010 $0.00534
Sonnet 5 $0.00004 $0.00213
Haiku 4.5 $0.00002 $0.00107

Measured 4d ago against content hash c605283e21c9, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

eval-behavior 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 4d 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.

.claude/commands/eval-behavior.md · 109 lines

How it starts

The opening of the file, as written. The whole thing — 109 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Autology Behavioral Eval

Run behavioral evals for the skill named $ARGUMENTS.

Setup

Read these files before proceeding:

  • skills/$ARGUMENTS/SKILL.md — the skill instructions
  • skills/$ARGUMENTS/evals/evals.json — eval cases with prompts and assertions

Workspace: $ARGUMENTS-workspace/ (create if needed, overwrite previous runs)

Step 1: Run each eval case

For each case in evals.json, run it twice in the same turn using the Agent tool — once with the skill, once without. Launch all pairs in parallel.

Always use isolation: "worktree" on every agent. Each agent runs in its own isolated git worktree — a full copy of the repository at the current HEAD. This means:

  • docs/ in the worktree is a clean copy; agents can freely write to it without contaminating the real repo or each other's state
  • No manual sandbox setup needed
  • After the agent completes, the worktree path is returned in the result (use it to read output files)

⚠️ Absolute paths bypass isolation. Worktree isolation only works when agents use relative paths. If an agent writes to an absolute path (e.g. the real repo's docs/ directory), it lands outside the worktree and contaminates the shared repo. Every agent prompt must include the instruction below to prevent this.

Fixture files: If the eval requires a file to pre-exist (e.g., docs/api-gateway.md that the task says "assume exists"), commit it to skills/$ARGUMENTS/evals/fixtures/ in the repo. The agent prompt should instruct the agent to copy it to docs/ at the start: "Copy skills/$ARGUMENTS/evals/fixtures/api-gateway.md to docs/ before starting."

With-skill agent prompt:

Execute this task following the skill instructions below.

SKILL INSTRUCTIONS:
<paste full contents of skills/$ARGUMENTS/SKILL.md here>

TASK: <eval prompt>

The project knowledge base is at docs/. Save:
- Your complete response → $ARGUMENTS-workspace/<eval-name>/with_skill/response.txt
- A step-by-step log of every tool you used → $ARGUMENTS-workspace/<eval-name>/with_skill/transcript.md

Create directories as needed.

IMPORTANT: Use only relative paths for all file operations (docs/, $ARGUMENTS-workspace/, etc.).
Do NOT write to any absolute path — absolute paths bypass your isolated worktree and contaminate the shared repo.

Read the full file on GitHub · 109 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. 4d ago First seen · 109 lines · 0 tokens per session scan A c605283e21c9

Subscribe to this mod's changes

eval-behavior is a command published in the GitHub repository Curt-Park/autology (24 stars, last pushed 5mo ago), licensed MIT. It adds 21 tokens to every session and 1,067 once invoked, about $0.0001 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.