Atomic Agent is a local-first AI agent that runs its control loop and state on a user's machine while using local or cloud models. It drives browsers, edits files, runs approved commands, remembers context, schedules follow-ups, and connects to external tools, with the catalogue providing skills for its use.
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 AtomicBot-ai/atomic-agent --skill verify-workgit clone --depth 1 https://github.com/AtomicBot-ai/atomic-agentWrote 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/atomicbot-ai/atomic-agent/verify-work)<a href="https://agentmods.dev/skills/atomicbot-ai/atomic-agent/verify-work"><img src="https://agentmods.dev/badge/skills/atomicbot-ai/atomic-agent/verify-work/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/atomicbot-ai/atomic-agent/verify-work"><img src="https://agentmods.dev/badge/skills/atomicbot-ai/atomic-agent/verify-work.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 | $0.00058 | $0.00891 |
| Opus 5.5 | $0.00023 | $0.00356 |
| Sonnet 5.5 | $0.00012 | $0.00178 |
| Haiku 4.5 | $0.00006 | $0.00089 |
Grade A, and why
verify-work 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 — 86 lines — stays where its author put it; the contents beside it link to each section on GitHub.
verify-work
Writing a file is not evidence. Running it is.
The failure this skill exists to prevent: a turn writes source files and a test file, replies "done, tests included", and never executes a single one of them. The operator finds out later that nothing ran. If you wrote it, run it — and if you cannot run it, say so plainly instead of implying you did.
The rule
Never report code as working, done, or fixed unless a command you ran in this turn proves it. Quote the command and the part of its output that carries the verdict — an exit code, a passing count, a rendered element. "Should work" is not a verdict.
If verification is impossible here (no runtime installed, needs a device, needs a credential you must not use), say which check you could not run and why, in one sentence. That is an honest result. Silence is not.
Pick the smallest command that could fail
Verify in this order and stop at the first that applies.
- The project's own test command. Read
package.json/Makefile/pyproject.tomlfirst — do not guess a runner.npm test·npm run <script>·pytest -q·cargo test·go test ./... - The file you just wrote.
node path/to/file.mjs·python3 path/to/file.py·bash path/to/script.sh - The entry point, once. A CLI:
--helpplus one real invocation with real arguments. A server: start it, hit one endpoint, stop it. - A syntax floor, only when nothing above can run:
node --check file.js·python3 -m py_compile file.py·tsc --noEmit
Run it from the directory the project expects. Capture stderr, not just stdout — a script that prints nothing and exits 1 has failed.
Web pages and anything visual
A page that loads is not a page that works. Use the browser tools:
- Open the file or serve it, then
browser.*to navigate to it. - Read the console. An uncaught exception on load is a failure even when the page looks fine. Report it and fix it.
- Screenshot and actually look: is the thing the operator asked for on screen, or is the canvas blank, the layout collapsed, the sprite absent?
- Exercise one interaction — a click, a key, a submit — and confirm the state changed.
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 · 86 lines · 58 tokens per session scan A cf81f64d64ec
verify-work is a skill published in the GitHub repository AtomicBot-ai/atomic-agent (3,347 stars, last pushed today), licensed MIT. It adds 58 tokens to every session and 891 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 2026-09-10.
Other skills, from other repositories
BrowserBash Browser Automation
BrowserBash is a vendor-independent, natural-language browser automation CLI. Drive a real browser from plain-English objectives or committable Markdown tests, run on local Chrome, CDP/Playwright MCP, Browserbase, LambdaTest, or BrowserStack, and stream NDJSON results with CI exit codes — using free local Ollama…
agent-eval
Use when measuring whether an LLM or agent system actually got better and gating merges on it: golden sets, fixing an inflated LLM-as-judge, scoring RAG (faithfulness, contextual recall) or agent trajectories (tool correctness, completion), or picking an eval framework. NOT building the agent loop, tools or RAG…
agent-browser
Use the host-side agent-browser CLI for local browser smoke tests, screenshots, snapshots, and simple UI validation against forwarded localhost URLs.
checking-a-page
Check that a web page you wrote or changed actually works, by opening it in a browser and driving every control on it. Use after creating or editing an HTML page, or the JavaScript or CSS that a page loads.
browser-qa
Testing web apps through a real browser: reconnaissance before acting, selector discovery, state capture, console errors and screenshot evidence. Use on a web surface end to end, or to reproduce a front-end bug.
test-driven-development
Writing code with tests first: RED-GREEN-REFACTOR, reproduce bugs as failing tests before fixing, and avoid testing anti-patterns. Use when implementing behaviour, fixing a defect, or asked to add tests.