StarNet is a local-first desktop harness for creating AI agents and arranging them in a pixel-art space station whose layout represents their capabilities, permissions, and workflow. It lets people run real model calls and tools through multiple bounded agents, using their own provider keys or accounts and optionally messaging the station through chat services. The catalogue entries are skills that extend or support work performed by StarNet agents.
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 androoAGI/starnet --skill grounded-citationsgit clone --depth 1 https://github.com/androoAGI/starnetWrote 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/androoagi/starnet/grounded-citations)<a href="https://agentmods.dev/skills/androoagi/starnet/grounded-citations"><img src="https://agentmods.dev/badge/skills/androoagi/starnet/grounded-citations/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/androoagi/starnet/grounded-citations"><img src="https://agentmods.dev/badge/skills/androoagi/starnet/grounded-citations.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 | $0.00040 | $0.01074 |
| Opus 5.5 | $0.00016 | $0.00430 |
| Sonnet 5.5 | $0.00008 | $0.00215 |
| Haiku 4.5 | $0.00004 | $0.00107 |
Grade A, and why
grounded-citations 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 12d 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 — 39 lines — stays where its author put it; the contents beside it link to each section on GitHub.
A citation is a promise that the source says what you say it says. Keep that promise mechanically: open every source, save its text, copy the supporting sentence, and prove by a literal search that the sentence is really there. The bar: the Commander can check any claim in under a minute.
Method
- List the claims the answer will need. Break the question into the facts that must be established: the numbers, dates, names, and who said what. This is your search plan, and later your checklist.
- Find sources and open each one.
web_searchfinds candidates;web_fetchopens them. Prefer the original (the law, the paper, the filing, the official page, the dataset) over an article about it. A search-result snippet is not a source. A page that would not open, or showed only a login or paywall notice, was not read and cannot be cited. For a file the Commander gave you, read it withfs.readand cite its path and page or section. - Save what you read. Write each opened text with
fs.writetosources/<n>-<short-name>.txt, headed by its URL, title, publisher or author, the published date if the page shows one, and the date you read it (today's date from the run). Keepsources/index.md: the numbered list, one line per source. - Draft claim by claim. For each factual sentence, copy from the saved text the one sentence that supports it, exactly as written, and tag the claim
[n]. Collect these incitations.md: claim, source number, quote. Give a second, independent source for figures that matter and for anything contested. - Check every quote literally. Search the saved file for the quote with
fs.search(plain text match), or run onecode.runpass that reads each saved file withawait tool("fs.read", { path })and tests every quote after collapsing runs of spaces and straightening curly quotes. A quote that is not found is not a quote: open the page again, find the real sentence, or drop the citation. Your memory of the page does not count. - Check the quote says what the claim says. Same subject, same number and unit, same period, same scope. "May" in the source is not "does" in the answer. One country or one year does not support a general statement. Note who is speaking: the page asserting something is different from the page reporting someone else's view.
- Deal with what is left. A claim with no supporting quote is either marked
[unverified]with the reason or removed. Your own working — arithmetic on cited figures, a conclusion drawn from two cited facts — is labelled as yours, with its inputs cited. - Write the answer. Answer first, markers inline, then
Sources: number, title, publisher, URL, published date, date read. Where sources disagree, show both and say so. Savecitations.mdwith the quotes and name it withdeliverable_note.
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.
- 12d ago First seen · 39 lines · 40 tokens per session scan A af9f75bbe3e9
grounded-citations is a skill published in the GitHub repository androoAGI/starnet (1,468 stars, last pushed 3d ago), licensed MIT. It adds 40 tokens to every session and 1,074 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 2026-10-04.
Other skills, from other repositories
Review a GitHub PR (via gh)
Review a specific GitHub pull request with gh — fetch the diff, fan out reviewers, consolidate, and optionally post the review. Requires the gh CLI or the GitHub MCP server.
Implement (multi-agent loop)
Orchestrate an implement -> review -> fix loop with subagents until reviewers sign off.
Best of N (parallel attempts)
Delegate N parallel subagents on the same task, then pick the best result.
Check work (verify against criteria)
Verify an implementation against acceptance criteria with a reviewer and a tester.
Commit (clean, conventional)
Stage the right changes and write a clear, conventional commit message.
Design doc (write -> review loop)
Draft a design document and iterate writer/reviewer subagents until consensus.