crystallize-verify

crystallize-verify is a command for coding agents from axiomantic/spellbook. It costs 39 tokens per session (3,331 once invoked), scanned A, original, MIT.

A command that compares a shortened document with its original to check whether any required behavior or instruction was lost. It returns either PASS or FAIL with findings.

In plain words
What is it for?
Use it to verify condensed versions of instructions, specifications, or other documents before delivery and identify exactly what must be restored.
Why use it?
It catches important omissions that a simple wording comparison might miss. The review focuses on changed behavior, not just different phrasing.

Command

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/axiomantic/spellbook/crystallize-verify
Clone the repo
git clone --depth 1 https://github.com/axiomantic/spellbook

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 crystallize-verify

README.md
[![agentmods](https://agentmods.dev/badge/commands/axiomantic/spellbook/crystallize-verify.svg)](https://agentmods.dev/commands/axiomantic/spellbook/crystallize-verify)
Your own site
<a href="https://agentmods.dev/commands/axiomantic/spellbook/crystallize-verify"><img src="https://agentmods.dev/badge/commands/axiomantic/spellbook/crystallize-verify.svg" alt="Measured on agentmods" height="20"></a>
Per session 39 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 3,331 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 $0.00039 $0.03331
Opus 5 $0.00019 $0.01665
Sonnet 5 $0.00008 $0.00666
Haiku 4.5 $0.00004 $0.00333

Measured 5d ago against content hash 20c75f6dedf0, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

crystallize-verify 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 5d 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.

commands/crystallize-verify.md · 333 lines

How it starts

The opening of the file, as written. The whole thing — 333 lines — stays where its author put it; the contents beside it link to each section on GitHub.

MISSION

Perform structurally isolated adversarial review of a crystallized document. Identify behaviors the original instructs that the crystallized version does not.

Invariant Principles

  1. Adversarial posture: Ask "What behaviors does the original instruct that the crystallized version does not?" -- not "Is this a good crystallization?"
  2. Structural isolation: Analysis based ONLY on the two provided documents. Do not access files, skills, or external context.
  3. Behavior-level, not word-level: Phrasing differences are acceptable. Behavioral differences are findings.
  4. PASS threshold is strict: PASS requires ZERO CRITICAL or HIGH findings. A single unresolved CRITICAL or HIGH = FAIL.
  5. Findable and fixable: Every finding must cite original location and describe the specific restoration needed.

Input Contract

Receives exactly:

  1. ORIGINAL DOCUMENT -- full text before crystallization
  2. CRYSTALLIZED DOCUMENT -- full text of crystallized output

Protocol

Phase 1: Behavioral Inventory of Original

Read the original document. Extract every behavioral instruction.

A behavioral instruction is any statement that, if absent from the crystallized output, would cause an LLM executor to behave differently. INCLUDE: IF/THEN conditions, MUST/NEVER/FORBIDDEN rules, phase sequences with step counts, thresholds with numbers, and gate conditions. EXCLUDE: rationale text explaining WHY a rule exists, historical context, and examples that illustrate rather than specify behavior.

REQUIRED format for each section -- do not abbreviate or skip:

<analysis>
Section: "[section name]"
Instructed behaviors: [list]
- Action required: [yes/no]
- Condition: [if any]
- Threshold/quantity: [if any]
- Exception/edge case: [if any]
</analysis>

Read the full file on GitHub · 333 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. 5d ago First seen · 333 lines · 39 tokens per session scan A 20c75f6dedf0

Subscribe to this mod's changes

crystallize-verify is a command published in the GitHub repository axiomantic/spellbook (10 stars, last pushed yesterday), licensed MIT. It adds 39 tokens to every session and 3,331 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.