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 zwolf25/tokenminning --skill visual-diffgit clone --depth 1 https://github.com/zwolf25/tokenminningWrote 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/zwolf25/tokenminning/visual-diff)<a href="https://agentmods.dev/skills/zwolf25/tokenminning/visual-diff"><img src="https://agentmods.dev/badge/skills/zwolf25/tokenminning/visual-diff/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/zwolf25/tokenminning/visual-diff"><img src="https://agentmods.dev/badge/skills/zwolf25/tokenminning/visual-diff.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 | $0.00000 | $0.01043 |
| Opus 5 | $0.00000 | $0.00522 |
| Sonnet 5 | $0.00000 | $0.00209 |
| Haiku 4.5 | $0.00000 | $0.00104 |
Grade A, and why
visual-diff 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.
How it starts
The opening of the file, as written. The whole thing — 67 lines — stays where its author put it; the contents beside it link to each section on GitHub.
visual-diff
Pixel-diff pre-filter for screenshot validation. Compares a new screenshot against a cached baseline of the same artifact: if nothing meaningfully changed, you never view the image at all (near-zero vision tokens); if something did change, you view a small crop of just the changed region instead of the full screenshot.
Token rule (do not break): on a skip decision, do not Read the screenshot — the JSON
result is the answer ("no visual change, X% diff"). On a crop decision, Read only
cropPath, never the original full screenshot. On a full decision (first time seeing this
artifact), there's no baseline yet to compare against — view the original if you need to,
that only happens once per artifact.
Step 1 — Prereq gate
Needs node on PATH plus the global pixelmatch/pngjs packages (npm install -g pixelmatch pngjs, one-time). If missing, install first — don't fall back to viewing the raw screenshot
just to avoid the install step, that's the exact token cost this skill exists to avoid.
Step 2 — Run it
node visual-diff.js "<screenshot-path>" --key "<stable-artifact-key>"
Point the path at wherever this tool is installed. <stable-artifact-key> should identify
this artifact, not this invocation — e.g. roadmap-deck/slide-04, or a Figma node id.
Reuse the same key across repeat checks on the same thing so the cache actually compares
against the right prior state.
Optional flags: --threshold <pct> (default 0.5 — % of pixels differing to count as a real
change), --pad <px> (default 24 — padding around the changed-pixel bounding box before
crop), --max-dim <px> (default 768 — long-edge cap after crop, downsampled if exceeded).
Step 3 — Act on the JSON result
The script always prints one JSON object to stdout:
"decision": "skip"— nothing meaningfully changed (diffPercentbelow threshold). Report this directly to the user ("no visual change, 0.1% diff") without viewing anything."decision": "crop"— something changed.ReadonlycropPath(a small PNG of just the changed region + padding), not the original screenshot."decision": "full"— first time this artifact's been checked, nothing to compare against yet. View the original screenshot if you need to; it becomes the baseline for next time.
What ships with it
2 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.
- 3d ago First seen · 67 lines · 0 tokens per session scan A 96a6ffba92ed
visual-diff is a skill published in the GitHub repository zwolf25/tokenminning (30 stars, last pushed yesterday), licensed MIT. It costs nothing until one of its globs matches a file; then it loads 1,043 tokens. 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 2026-09-15.
Other skills, from other repositories
html-ppt-zhangzara-block-frame
Rescuing a messy startup deck into a board-grade system — the diagnosis, the page grammar, and the rebuilt proof pages. Built as a decision-grade design craft deck for founders, exec presenter.
accesslint-scan
Audit a live page for accessibility issues, locate each WCAG violation precisely, and return a selector-grounded fix worklist without editing.
check
Reviews diffs, PRs, release readiness, and publishing follow-through. Use when asked to review, triage issues or PRs, or ship. Not for debugging root causes or prose.
health
Audits agent config, instruction drift, hooks or MCP, and AI maintainability. Use when Claude, Codex, or Pi setup looks wrong. Not for application bugs or PR review.
hunt
Finds root cause before any fix. Use when something errors, crashes, regresses, or used to work. Not for code review or new features.
think
Turns rough ideas into approved, decision-complete plans before coding. Use when planning architecture, judging whether to build, or writing a handoff. Not for bug fixes or small edits.