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.
npx skills add larsboes/Axon --skill crystallizegit clone --depth 1 https://github.com/larsboes/AxonWrote 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.
[](https://agentmods.dev/skills/larsboes/axon/crystallize)<a href="https://agentmods.dev/skills/larsboes/axon/crystallize"><img src="https://agentmods.dev/badge/skills/larsboes/axon/crystallize/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/larsboes/axon/crystallize"><img src="https://agentmods.dev/badge/skills/larsboes/axon/crystallize.svg" alt="Reviewed on agentmods" width="80" height="20"></a>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.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00177 | $0.02186 |
| Opus 5 | $0.00088 | $0.01093 |
| Sonnet 5 | $0.00035 | $0.00437 |
| Haiku 4.5 | $0.00018 | $0.00219 |
Grade A, and why
crystallize 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 today.
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.
How it starts
The opening of the file, as written. The whole thing — 182 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Crystallize
Someone hands you a mess and wants a decision out of it. The mess is usually right about what matters and wrong about what is true, because it was written from memory.
Your job is a trade. They supply judgment. You supply measurement. Neither of you can do the other's half, and most bad versions of this task fail because one side tried.
The one rule
Never ask a question you can answer by measuring, and never assert a fact you have not measured.
Both halves fail constantly and they fail in opposite directions. Asking the user what they could have been told wastes the scarce resource — their attention. Asserting from memory is worse: it reads as authoritative and it is wrong at a rate that will surprise you.
This is not a warning about carelessness. In a long session working on a real system, claims made
from plausible inference were wrong roughly a third of the time. See
references/grounding.md for the catalogue of how — it is worth reading once, because the failure
modes are specific and repeat.
The loop
- Read the target, then read the ground truth. The document is a claim about a system. Go look at the system. Count things.
- Ask 3–4 questions in one round. Batched, concrete, with a recommendation. Never one at a time — that turns a conversation into an interrogation.
- Write the answers in immediately, dated, with the reasoning. A decision you did not record is a decision you will re-litigate.
- Report what the measurement contradicted. This is where most of the value is.
- Repeat until the open-question list is empty, then hand over what is work rather than undecided.
Start by measuring, not by reading
Before the first question, get numbers. A document says "most of my notes are a mess"; the measurement says "42% of one folder is under 40 words and the other folder is at 6%". The second sentence starts a real conversation and the first one starts a vague one.
scripts/census.py does the generic version for a markdown corpus — per-folder counts, stub
rates, frontmatter key usage, orphan detection, dialect drift. Run it before you form an opinion:
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.
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.
- today Changed · +14 lines f2827ad2d852
- 9d ago First seen · 168 lines · 177 tokens per session scan A e4ef5ad78104
crystallize is a skill published in the GitHub repository larsboes/Axon (1 stars, last pushed yesterday), licensed MIT. It adds 177 tokens to every session and 2,186 once invoked, about $0.0009 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 2026-09-07.
Other skills, from other repositories
orbit-notion
Open Orbit briefing skill — selected by the Orbit pipeline when Notion is the user's only connected connector, or when the user explicitly scopes their daily digest to Notion. Pulls the past 24 hours of document edits, comments, mentions, and database row changes from the user's authenticated Notion connection and…
webgl-halftone-drift
A self-contained WebGL2 hero: a flowing field screened through a rotated halftone dot grid into a duotone print aesthetic; move the cursor to bend the drift.
webgl-holographic-foil
A self-contained WebGL2 hero: thin-film interference over a crushed-foil surface whose palette shifts with the viewing angle; move the cursor to tilt the film.
html-ppt-taste-brutalist
16:9 HTML deck in tactical-telemetry / CRT-terminal taste. Deactivated-CRT charcoal slides, white-phosphor monospace, hazard-red accent, scanline overlay, ASCII syntax, density over decoration. Distilled from Leonxlnx/taste-skill brutalist-skill (Tactical Telemetry mode).
frame-light-leak-cinema
Film light leaks, grain, 16:9 letterbox, and large serif type for cinematic openings or chapter cards.
card-xiaohongshu
Xiaohongshu-style knowledge cards, arranged as a swipeable multi-card carousel.