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 agentmods add skills/m4ttstack/fast-browser/annotating-screenshotsnpx skills add m4ttstack/fast-browser --skill annotating-screenshotsgit clone --depth 1 https://github.com/m4ttstack/fast-browserWhat 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 | $0.00047 | $0.03794 |
| Opus 5 | $0.00023 | $0.01897 |
| Sonnet 5 | $0.00009 | $0.00759 |
| Haiku 4.5 | $0.00005 | $0.00379 |
Grade A, and why
annotating-screenshots 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 yesterday.
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 — 278 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Annotating Screenshots
One capture-annotated.js call produces the PNG and the boxes together. Every
number in the config comes from that one call. Nothing is estimated by reading
the image, and nothing is typed by hand.
The pipeline
- Capture and measure in one call.
browser_run_code_unsafewith the macro's installed path asfilename, never a bare name: a relativefilenameis resolved against the browser server's own working directory and then checked for containment, so a bare name fails before the macro runs. Pass theScript:path from~/.fast-browser/macros/MACROS.md,~/.fast-browser/macros/capture-annotated.js, with~written out as your own absolute home directory path. Thenargs: { targets: { <key>: '<selector>' }, out: '<name>', home: '<your $HOME>' }. The macro cannot read$HOMEitself, so pass your own absolute home directory path there too.targetsis for elements: the values you mark and anything an arrow points at. The empty space a label sits in is usually not an element, so the macro measures that for you: the return'sspacemap gives each resolved key up to four verified-empty bands ({ side, box }, sidesabove/below/left/right, a side omitted wherever no empty room of useful size exists). What that verification covers, and the blind spots it cannot, is stated under Composition. You cannot add a key after the call. Under the defaultsafeprofile this prompts for approval every time. Expect the prompt; it is not a failure. - Write the config from that call's return value.
- Draw:
fast-browser annotate <config-path>. - Read the output PNG yourself before reporting: nothing clipped at an edge, nothing covering content it should not, blurred text genuinely illegible. Then report the output name and every missed key.
If ~/.fast-browser/macros/capture-annotated.js is absent, run
fast-browser setup to install it. Do not write a substitute script.
What ships with it
1 file 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.
- yesterday First seen · 278 lines · 47 tokens per session scan A 39095db1b688
annotating-screenshots is a skill published in the GitHub repository m4ttstack/fast-browser (0 stars, last pushed 7d ago), licensed MIT. It adds 47 tokens to every session and 3,794 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.
Other skills, from other repositories
webcmd-adapter-author
Use when writing a Webcmd adapter for a new site or adding a new command to an existing site. Guides end-to-end from first recon through field decoding, adapter coding, and verify. Replaces webcmd-oneshot / webcmd-explorer. For ad-hoc browser driving without an adapter, use webcmd-browser instead; for top-level…
webcmd-browser
Use when no deterministic Webcmd adapter command covers a live browser task requiring Playwright interaction, authenticated handoff, visible UI verification, or ad-hoc page inspection.
webcmd-usage
Use at the start of any Webcmd session. This is the top-level map of what webcmd can do, how to discover adapters, what flags and output formats are universal, and which specialized skill to load next. Point here when an agent asks "what can webcmd do?" or "how do I find the right command?".
webcmd-autofix
Automatically fix broken Webcmd adapters when commands fail. Load this skill when a webcmd command fails; it guides you through collecting a trace artifact, patching the adapter, retrying, and safely reporting reproducible upstream defects. Works with any AI agent.
smart-search
Use when a request needs search, research, source discovery, direct URL fetch, the first-choice Webcmd fetch path, evidence fetching, or search-capable Webcmd adapter discovery.
webcmd-sitemap-author
Use when creating or maintaining Webcmd site sitemaps: agent-facing navigation, page-state, action, workflow, API-reference, pitfall, and fallback knowledge for a website. Use after browser exploration discovers durable site context, when a sitemap is stale, or when promoting local site knowledge into the repo.