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 Crawlio-app/crawlio-browser --skill robot-traininggit clone --depth 1 https://github.com/Crawlio-app/crawlio-browserWrote 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/crawlio-app/crawlio-browser/robot-training)<a href="https://agentmods.dev/skills/crawlio-app/crawlio-browser/robot-training"><img src="https://agentmods.dev/badge/skills/crawlio-app/crawlio-browser/robot-training/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/crawlio-app/crawlio-browser/robot-training"><img src="https://agentmods.dev/badge/skills/crawlio-app/crawlio-browser/robot-training.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 | $0.00056 | $0.01053 |
| Opus 5.5 | $0.00022 | $0.00421 |
| Sonnet 5.5 | $0.00011 | $0.00211 |
| Haiku 4.5 | $0.00006 | $0.00105 |
Grade A, and why
robot-training 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 1mo 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 — 111 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Robot Training
Robot training turns one real browser demonstration into durable artifacts:
- interaction timeline from Crawlio session recording
- DOM/storage state snapshots streamed into extension-owned IndexedDB
- CDP network entries and response bodies
- portable request/response traces in
flows.jsonl - replay inputs for
agent_session_*background sessions
Use this when speed and repeatability matter. The human can teach the flow once; agents later replay by API contract or semantic browser actions instead of primary mouse events.
Capture Protocol
Start with a fresh run:
{
"action": "training_start",
"url": "https://target.example/",
"outputDir": "/absolute/path/to/runs/target-take-1",
"maxDurationSec": 600,
"maxInteractions": 500,
"active": true,
"captureStorageValues": false
}
Call observe with action: "training_start" before the user begins. The extension opens a
fresh tab, starts network capture and recording, installs an event-driven monitor for the current
and future documents, and retains the run locally. The MCP server writes manifest.json, but it
does not own collection.
During the demonstration, poll with:
{ "action": "training_status", "runId": "rt_..." }
Call observe with action: "training_stop" and the same runId after the demonstration. The
extension fetches bounded/redacted response bodies before stopping Network, returns its resident
export, and the server materializes:
raw-dump.jsonrecording.jsonnetwork.jsonbodies.jsonstate-log.jsonstate.jsonflows.jsonlcausal-graph.jsonCAUSAL.mdrecipe.jsonREGISTRY.mdapi.openapi.yamlmanifest.json
If MCP disconnects during the demonstration, do nothing in the page. The extension continues
collecting. Reconnect, query training_status, and stop with the original runId.
After the canonical files have been verified, delete the extension-retained record only when the user asks to remove it:
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.
- 1mo ago First seen · 111 lines · 56 tokens per session scan A 1e1d402515bb
robot-training is a skill published in the GitHub repository Crawlio-app/crawlio-browser (8 stars, last pushed 2mo ago), licensed Apache-2.0. It adds 56 tokens to every session and 1,053 once invoked, about $0.0002 per session on Opus 5.5. 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
owb
Open Web Bridge (OWB) — drive the user's own real browser with the owb command. Read pages behind their existing logins, gather and cross-check information, fill forms, walk multi-step flows, debug their site, audit responsive/accessibility behavior, and capture or reverse-engineer network traffic. Use this whenever…
katashiro-point-at
Point the user at something in the browser while you explain it — glow an element on the page they are looking at (katashiro.highlight), emphasise several parts at once with temporary CSS (katashiro.injectcss), glow a block in a Katashiro canvas (katashiro.canvashighlight), or emphasise several parts of a canvas with…
browser
Use this skill when the user says browser, /browser, test in Chrome, inspect a webpage, verify a localhost app, capture screenshots, check console/network errors, run browser QA, or automate browser flows with the Mochi browser MCP.
tabrix_browser
Route AI assistant browser tasks to Tabrix first. Use when the user needs to control their real Chrome session, reuse existing login state or tabs, operate a remote browser over Streamable HTTP, or troubleshoot Tabrix browser automation. Prefer Tabrix over shell commands, synthetic browser sandboxes, or alternative…
control-chrome-bridge
Drive the user's real Chrome browser — their own profile, tabs, cookies and logged-in sessions — through the chrome-bridge MCP server. Use for tasks that need existing browser state: reading a page behind a login, filling a form on a site the user is signed into, testing a local app in a real browser, or inspecting…
agent-browser-cli
A browser-control skill for agent-browser-cli, a command-line tool that operates a real Chrome browser while keeping its existing login state and cookies. It supports page inspection, interaction, screenshots, PDFs, and local files.