greenroom
# greenroom · 候场
**Interview prep that ends in words you can actually say.**
*Seven agent skills, a 27-role knowledge base, and an open Markdown workspace format. Hand an agent your resume and a target JD — get role intel, a sourced story bank, speakable scripts, mock rounds, and debriefs that feed the next round.*
[](LICENSE)
[](https://github.com/YunyueLi/greenroom/releases)
[](https://github.com/YunyueLi/greenroom)
[](README.zh-CN.md)
### **[Open the app →](https://greenroom.ungetsu.net/)** · [Install the skills](#install-in-30-seconds) · [Read the spec](/greenroom/docs/workspace-spec.html)
候场 — the room where performers wait and prepare before going on stage.
resume + target JD ─→ role research (company, interviewers, likely questions)
─→ story bank (your facts, one source of truth)
─→ industry brief ─→ speakable scripts
─→ mock interview ─→ coach report
─→ debrief ─→ next-round prep
Why this one
It writes answers, not outlines. Most prep tools hand you bullet points and leave the hardest part — turning them into sentences under pressure — to you. greenroom writes the sentences.
Every number carries its own definition. A script line never states a figure without recording how it was counted and where it came from, so a follow-up question lands on something you can defend instead of something you improvised.
上线到现在，自助解决率从 31% 做到 52%，客服人力成本降了 18%。
**数字出处**
- 解决率 31%→52%，统计口径=未转人工且 24h 未重复进线（被追问时主动给口径）
Your prep is a folder you own. Plain Markdown under a published contract. No lock-in, no export button — the files are already yours, and any tool that reads the spec can read them.
Install in 30 seconds
# Claude Code plugin, versioned updates
/plugin marketplace add YunyueLi/greenroom
/plugin install greenroom@greenroom
# or any agent that supports skills
npx skills add YunyueLi/greenroom
Then say:
I’m interviewing for the AI PM role at Acme next Tuesday. Here’s my resume and the JD.
The greenroom skill runs the pipeline and writes a workspace folder. Say “just the scripts” or “run a mock” and it routes to one step instead.
The skills
| Skill | What it does |
|---|---|
greenroom |
Entry point — runs the full pipeline, or routes to a single step |
job-intel |
Deconstructs the JD, researches company and interviewers, forecasts the next round |
story-bank |
Mines your real experience into reusable story cards with sourced numbers |
industry-brief |
Writes the industry and role reading you need before you can sound informed |
interview-script |
Turns all of it into first-person answers you can read out loud |
mock-interview |
Plays the interviewer with pressure follow-ups, then scores the performance |
debrief |
Reconstructs the round you just finished into a revision list for the script |
What it writes
my-greenroom/
├── profile.md candidate profile
├── resume.md resume as Markdown, the fact source
├── story-bank.md story cards, repackaged per role type
├── library/ industry briefs, research, reference reading
└── jobs/<company-role>/
├── job.md the JD, plus a timeline of the process
├── intel.md company, interviewers, question forecast
├── script.md the verbatim script
└── rounds/ prep notes, mock reports, debriefs
A fictional example workspace lives in examples/demo-workspace/ — real format, invented people.
The role knowledge base
knowledge/ holds 27 roles across industries and functions: high-frequency questions, role fundamentals, follow-up patterns. It contains no candidate data and is safe to read, fork, and contribute to. The industry-brief skill checks it first, so the more complete it gets, the faster every new user starts.
Build on it
The workspace format is a published contract, not an internal detail. Anything that reads it works with everything else that reads it.
docs/workspace-spec.md |
File layout, frontmatter, the script.md card format, the HTTP read endpoints |
docs/realtime-bridge.md |
Contract for live-prompting and mock-interview backends |
serve.py |
Zero-dependency reference backend — Python standard library only |
git clone https://github.com/YunyueLi/greenroom.git
cd greenroom
./start.sh ~/my-greenroom # serves the workspace over HTTP on :8787
No API key is needed to read a workspace. Add an OpenAI-compatible key in .env to unlock live prompting, mock interviews, and one-pass generation.
The hosted app
greenroom.ungetsu.net is the official product: the full interface, accounts and cloud sync, hosted model access, and a maintained deployment. It reads and writes the same workspace format documented here, so nothing you build against the spec is wasted.
What’s here and what isn’t
Here, under AGPL-3.0: the skills, the role knowledge base, the workspace contract, the reference backend, the demo workspace. Run it, fork it, build compatible tools on it.
Not here: the product interface, accounts and cloud sync, the hosted service backend, and the brand assets. Those belong to the official product.
Privacy
Your workspace is a folder on your machine. Never commit real resumes, interview material, salary figures, private company information, API keys, or production logs — to this repository or any other public one.
License
greenroom Community Edition is licensed under the GNU Affero General Public License v3.0. See LICENSE.
The greenroom name, logo, wordmark, domain, and confusingly similar product branding are governed by TRADEMARK.md.