nanocosmos-skills
Skills that teach your AI coding assistant the nanocosmos platform.
Install them once, and Claude Code (or Cursor, Copilot, and similar tools)
answers questions about nanoStream and
nanoStudio with verified, documentation-backed knowledge — and can run checks
against your own account instead of guessing.
Public home: https://github.com/nanocosmos/nanocosmos-skills
What this is, in plain words
A skill is a folder with a SKILL.md file (instructions written for an AI
assistant) plus, where useful, small scripts and reference documents. Your
assistant reads a skill only when a question matches it, so installing all of
them costs nothing until they are needed. The scripts are ordinary bash and
python3 files you can read before running; they call the public nanocosmos
APIs with your credentials and only ever reach your own organisation's
streams and metrics.
With the skills installed you can ask things like:
- "Is my stream
XXXXX-YYYYY really live, and why does it take five seconds to start?"
- "Give me a working nanoPlayer 5 config for secure playback of this stream."
- "What was our buffering ratio last night, per stream and per country?"
- "How do I set up nanoStudio Chroma for three branded outputs from one green-screen feed?"
- "Build me a dark, broadcast-style control panel for this player page."
What the skills are not: a support channel, and not a replacement for the
documentation. They are made for nanocosmos
customers and evaluating users, are updated frequently, and contain no
internal-only nanocosmos information. When a question needs account-side
checks or platform-side monitoring, the skills say so and point you to
support.
Generated repository. Everything here is generated from nanocosmos's
internal knowledge base and re-released on every sync. Don't edit files
here — corrections land in the source. Found a mistake? Open an issue.
Skills
| Skill | For | What it enables |
|---|
nanostream-analytics | anyone with a dashboard account | Query your own nanoStream Analytics metrics (buffering, latency, startup time, playback errors by geo/browser, usage volumes) with your dashboard credentials — including the query semantics that prevent wrong answers. Read-only |
nanostream-development | developers / integrators | Ingest with ffmpeg/OBS (RTMP/SRT), playback paths (JS player, iframe, emulated MP4/M3U8, RTMP pull), nanoPlayer 5 configuration with verified parameter values, JWT-secure playback, ABR groups, ready-to-fork player pages |
nanostudio | producers / operators | The real-time production tools product line. nanoStudio Chroma: server-side chroma-key and multi-brand compositing — one green-screen feed in, multiple branded streams out to nanoStream and global destinations. More production tools (audio/video improvement, denoising) are planned |
nanostream-diagnostics | anyone operating streams | Self-serve troubleshooting with your own organisation's API key: is the stream really live, GOP/keyframe measurement (the top cause of slow starts), stream architecture resolution, token validation, plus 30+ troubleshooting references (ingest, playback, quality, latency, errors, captions, recording) |
dark-neutral-studio | frontend developers | A complete portable dark-UI design system for AV/broadcast tooling — tokens, CSS, layout, status semantics; pairs with the nanostream-development player pages |
Install
Pick one route per machine — they all deliver the same five skills, and two
routes side by side would load every skill twice.
1. Claude Code plugin (recommended)
One install, all five skills, updated together:
claude plugin marketplace add nanocosmos/nanocosmos-skills
claude plugin install nanocosmos@nanocosmos
Working from a local clone instead? claude plugin marketplace add /path/to/nanocosmos-skills does the same. For everyone working in one
repository, add --scope project to both commands and commit the
.claude/settings.json they write.
2. Clone into your skills folder
Claude Code loads any folder under its skills directory that carries a
.claude-plugin/plugin.json as a plugin, so a plain clone is enough:
git clone https://github.com/nanocosmos/nanocosmos-skills.git ~/.claude/skills/nanocosmos-skills
The skills appear as nanocosmos:<skill> in the next session (or after
/reload-plugins). Update with git pull. Inside one project, use a git
submodule at .claude/skills/nanocosmos-skills instead; Claude Code asks you
to trust the workspace once before loading it.
3. Generic skills installers
Both installers read the Agent Skills format and place the skills where your
assistant expects them — Claude Code, Cursor, Codex, Copilot, Windsurf and
others. Default scope is the current project; the flags shown install for
every project.
npx skills add nanocosmos/nanocosmos-skills -g # choose the agent interactively
gh skill install nanocosmos/nanocosmos-skills --agent claude-code --scope user
Add --skill <name> (npx) or a skill name after the repo (gh) to install a
single skill.
4. install.sh (manual copy)
A plain copy with no further tooling, for Claude Code's flat skills layout:
git clone https://github.com/nanocosmos/nanocosmos-skills.git
cd nanocosmos-skills
./install.sh # interactive: choose skills + scope
./install.sh --all --user # everything into ~/.claude/skills/
./install.sh nanostream-development nanostream-diagnostics --project
Any other assistant: copy the desired skills/nanocosmos/<name>/ folder
into your tool's skills or rules directory, or point your assistant at the
SKILL.md files directly.
First steps
-
Credentials, once. The diagnostics scripts need your organisation's
bintu API key (Dashboard → Organisation → API Keys); the analytics scripts
need a dashboard login or an access token. Create the credential file with
bash skills/nanocosmos/nanostream-diagnostics/scripts/init_env.sh \
skills/nanocosmos/nanostream-analytics/scripts/.env.example
and fill in ~/.config/nanocosmos/.env (same path on macOS, Linux and
Windows with Git Bash; it survives plugin updates). Exporting the variables
in your shell works too. Prefer an expiring access token over the raw key
wherever a credential leaves your machine — the diagnostics skill explains
how (create_access_token.sh).
-
Ask. Start a session and ask a question from the list above, or say
"use the nanocosmos diagnostics skill to check stream XXXXX-YYYYY".
-
Update. New syncs land on every push; pick them up with the update
command of your install route (see the table above), for the plugin:
claude plugin marketplace update nanocosmos && claude plugin update nanocosmos
Try it yourself
Every skill ships a handful of test questions with the answer shape to expect
(skills/nanocosmos/<skill>/tests/questions.txt), and tests/run.sh asks them
all through claude -p and checks the answers:
tests/run.sh # from a clone; ~5 minutes, a few cents per question
tests/run.sh --only 02,03 # a subset by question id or skill name
tests/run.sh --list # just show the questions
# plugin install without a clone: the marketplace copy has the same files
bash ~/.claude/plugins/marketplaces/nanocosmos/tests/run.sh
Or ask them yourself in a session and compare:
| Ask | A good answer |
|---|
| Is stream HX26g-YVe8D live? | runs get_stream_info.sh and reports the state; never guesses |
| Measure the keyframe interval of HX26g-YVe8D. | runs the probe and reports about 2 seconds plus codec and resolution |
| What was my organisation's buffering ratio yesterday? | uses your dashboard login, states the UTC range and the scope, or tells you which credential is missing — no invented numbers |
| Give me a secure-playback nanoPlayer 5 config for nNCcf-nMhQY. | a documented config with the demo token and a warning that the token expires |
| My ingest is being rejected — what can you check? | self-serve checks first, then "contact support with these details" |
Answers vary in wording; what matters is that the skill ran its scripts where
it could and did not invent values. Found a wrong answer? Open an issue with
the question and the answer file.
Requirements
- Your own nanocosmos account (trial accounts included): a bintu API key
for the diagnostics scripts and/or dashboard credentials for the
analytics scripts. The skills only ever access your own organisation's data.
bash, curl; python3 for a few scripts; ffprobe (ffmpeg) for the
keyframe/GOP checks and recordings; Node ≥ 18 only for the optional .js
variants. On Windows use Git Bash.
Privacy and security
- Credentials stay on your machine. No script sends anything anywhere except
the nanocosmos APIs you are calling; there is no telemetry in this repo.
- Never paste API keys or tokens into chat, tickets, web pages or client-side
code — the skills tell your assistant the same.
- The analytics scripts are read-only. Use a read-only dashboard account for
them where possible.
- Scripts are plain text: read anything before you run it, as with any
third-party code.
Support and feedback
- Product documentation: https://docs.nanocosmos.net
- Support requests: https://www.nanocosmos.net/support or
[email protected] — this is where questions about your streams and
account belong. GitHub issues in this repo are for mistakes in the skills
themselves (wrong parameter, outdated statement, broken script), not a
support channel.
- Pull requests are welcome as suggestions, but cannot be merged here
directly: this repository is generated. Accepted changes are ported into the
source and appear with the next sync, with credit in the changelog.
- Skill routing and the prefilled ticket-form parameters (for AI agents):
skills/nanocosmos/README.md
License
Commercial — not open source. These materials are part of the nanocosmos /
nanoStream product offering: use requires a valid nanocosmos account or
agreement under the nanocosmos Terms
(trial accounts included). Browsing this repository for evaluation is fine;
redistribution and competing use are not. Full text: LICENSE.
Trademarks are covered separately — see TRADEMARK.md.
