InterviewOps SDK

InterviewOps SDK is a standalone Node.js package for running an interview-note collection pipeline on top of:

opencli for Xiaohongshu collection
oh-my-codex / omx for stable Codex-side orchestration

It is designed for the workflow you built earlier in opencli, but split into a dedicated SDK repository with:

reusable domain types
OpenCLI adapters
OMX stabilization wrapper
a nightly interview collection pipeline
seller / lead-gen note marking
JSON + HTML exports

Current release:

v0.1.0
Git tag: v0.1.0

What It Does

For Xiaohongshu interview notes, the SDK can:

incrementally search notes by query
hydrate note detail content
enrich comments
extract interview questions
mark likely seller / lead-gen accounts or notes
export:
- xhs_notes.json
- xhs_questions.json (loose / compatibility)
- xhs_questions_strict.json (default reporting set)
- topic buckets
- company / round summary
- filterable HTML overview
optionally auto-commit each cycle

Install

git clone https://github.com/jerry609/InterviewOps-SDK.git
cd InterviewOps-SDK
npm install
npm run build

Quick sanity checks:

npm run typecheck
npm test
npm run build

External prerequisites:

Node.js 20+
opencli
omx
Chrome logged into xiaohongshu.com

Package notes:

The published SDK is ESM-only
The primary supported surface is the bundled CLI plus ESM imports
TypeScript consumers should use a Node runtime config (moduleResolution: "Node16" or "NodeNext")

CLI

After build:

node dist/cli.js --help

Main commands:

npm run dev -- init
npm run dev -- template
npm run dev -- sources
npm run dev -- seed-import --source-notes /path/to/xhs_notes.json
npm run dev -- harvest
npm run dev -- hydrate --limit 12
npm run dev -- comments --limit 8
npm run dev -- normalize
npm run dev -- questions
npm run dev -- overview
npm run dev -- status
npm run dev -- doctor
npm run dev -- export
npm run dev -- seller-summary
npm run dev -- ralph "analyze the current dataset"
npm run dev -- ralph-loop 6 --workspace ./workspaces/xhs-agent-algo-feb2026
node dist/cli.js stats
node dist/cli.js template
node dist/cli.js doctor
node dist/cli.js export
node dist/cli.js seller-summary
node dist/cli.js cycle
node dist/cli.js nightly 8
node dist/cli.js ralph-loop 6 --workspace ./workspaces/xhs-agent-algo-feb2026
node dist/cli.js validate
node dist/cli.js omx-safe doctor

During development:

npm run dev -- init
npm run dev -- harvest
npm run dev -- hydrate --limit 12
npm run dev -- comments --limit 8
npm run dev -- normalize
npm run dev -- questions
npm run dev -- overview
npm run dev -- status
npm run dev -- doctor
npm run dev -- stats
npm run dev -- export
npm run dev -- seller-summary
npm run dev -- cycle
npm run dev -- nightly 8
npm run dev -- omx-safe doctor

Command notes:

template: copies the bundled LaTeX interview template into the workspace
sources: lists currently built-in source adapters
seed-import: imports scoped seed notes from an existing xhs_notes.json into the target workspace
harvest: runs incremental search only
hydrate: fills note detail content only
comments: enriches comments only
normalize: refreshes question extraction and seller flags only
questions: rebuilds both loose and strict xhs_questions*.json outputs
overview: rebuilds strict overview HTML/summary plus seller reports
status: shows current stats plus last recorded stage runs
doctor: verifies node, opencli, omx, config path, data dir, and report dir
export: rebuilds question/topic/overview/seller outputs from existing note data
seller-summary: refreshes seller-tagged reports from current note data
ralph: shortcut for omx-safe exec --full-auto '$ralph "..."'
ralph-loop: repeatedly runs bounded Ralph collection cycles for a dedicated workspace

Control Plane

The control plane is the typed orchestration layer that OMX/Codex reads before deciding the next action.

control-status prints the current typed orchestration snapshot:

npm run dev -- control-status --workspace ./workspaces/xhs-agent-algo-feb2026

run-operation executes exactly one typed operation and persists the result in workspace state and operation_journal.jsonl:

npm run dev -- run-operation hydrate --workspace ./workspaces/xhs-agent-algo-feb2026 --limit 12 --reason "pending_hydrate backlog dominates current cycle"

ralph-loop now uses control-status plus run-operation internally instead of a fixed stage sequence.

OpenCLI Integration

By default the SDK calls:

opencli xiaohongshu search ...
opencli xiaohongshu note-detail ...
opencli xiaohongshu comments ...

If your working opencli is a local checkout instead of a globally installed binary, you can point the SDK at it:

export INTERVIEWOPS_OPENCLI_BINARY=npm
export INTERVIEWOPS_OPENCLI_ARGS_JSON='["-C","/path/to/opencli","run","dev","--"]'

That makes the SDK run commands like:

npm -C /path/to/opencli run dev -- xiaohongshu search ...

If live XHS search is unstable, you can seed a dedicated workspace from an existing notes library first:

npm run dev -- seed-import \
  --workspace ./workspaces/xhs-agent-algo-feb2026 \
  --source-notes /home/master1/opencli/interview_data/xhs_notes.json

OMX Stabilization

omx-safe wraps omx with a stable policy:

removes common proxy environment variables
forces USE_OMX_EXPLORE_CMD=0
creates .omx/state automatically

Example:

npm run dev -- omx-safe doctor

Default Layout

The SDK writes into the current workspace:

interview_data/
  xhs_notes.json
  xhs_questions.json
  xhs_questions_strict.json
  xhs_questions_nlp.json
  xhs_questions_nlp_strict.json
  xhs_questions_backend.json
  xhs_questions_backend_strict.json
  xhs_questions_algo.json
  xhs_questions_algo_strict.json
  company_round_summary.json
reports/xhs-miangjing/
  index.html
  status.json
  run_history.jsonl
  xhs_questions_nlp.html
  xhs_questions_backend.html
  xhs_questions_algo.html
  seller_candidates.json
  author_seller_summary.json
  seller_summary.md
  progress.log
templates/
  interview-note-template.tex
  interview-note-template.pdf

Default reporting behavior:

xhs_questions.json and xhs_questions_{topic}.json stay as loose compatibility exports
xhs_questions_strict.json and xhs_questions_{topic}_strict.json are the cleaned question-bank exports
index.html, topic HTML files, and company_round_summary.json default to the strict set

Workspace Init

Create a local config and output directories in the current workspace:

npm run dev -- init

That writes:

./interviewops.xhs.json
./interview_data/
./reports/xhs-miangjing/

You can also initialize another workspace:

npm run dev -- init --workspace /data/interviewops

Example PRD

See:

examples/xhs-miangjing.prd.json

By default the CLI uses:

./interviewops.xhs.json if it exists in the target workspace
otherwise the packaged example file

You can override it explicitly:

npm run dev -- cycle --prd ./examples/xhs-miangjing.prd.json

The PRD now includes:

source
query list
sellerWhitelist
data/report/state paths
search/detail/comment batch and timeout policy
harvest/sleep cadence

Seller / Lead-Gen Marking

The SDK does not drop seller-team notes.
It keeps them and marks them with:

seller_flag
seller_tags
seller_confidence

These fields are propagated into:

xhs_notes.json
xhs_questions.json
topic exports
overview HTML
seller_candidates.json
author_seller_summary.json
seller_summary.md

Whitelist config example:

{
  "sellerWhitelist": {
    "authors": ["可信作者A"],
    "note_ids": ["69c9d37b0000000023007921"],
    "title_keywords": ["内部分享"],
    "urls": ["example.com/trusted"]
  }
}

Whitelisted notes keep raw seller tags/confidence for debugging, but:

seller_flag will be forced to false
seller_whitelisted will be true
seller_whitelist_reason records the match source

Purchase Link Detection

The SDK also marks notes that appear to contain purchase links.

Current outputs include:

purchase_link_flag
purchase_links
purchase_link_tags
purchase_link_confidence

Detection combines:

explicit commerce URLs
purchase-link phrases
e-commerce platform mentions

These fields are surfaced in:

note JSON
question JSON
topic HTML
overview HTML
seller summary markdown

Source Adapters

Current built-in adapters:

xiaohongshu

The pipeline now resolves a source adapter from config:

{
  "source": "xiaohongshu"
}

That keeps the CLI stable while making it possible to add more sources later without rewriting pipeline orchestration.

Dedicated Agent / LLM Algorithm Workspace

Bundled workspace:

This workspace is scoped to:

Xiaohongshu
2026-02-01 onward
internet major companies
Agent / 智能体 / LLM / 大模型应用开发
算法岗 / NLP / 大模型算法

Run the dedicated Ralph loop:

npm run dev -- ralph-loop 6 --workspace ./workspaces/xhs-agent-algo-feb2026

What this dedicated loop is meant to collect:

Xiaohongshu notes only
internet major companies
Agent / 智能体 / LLM / 大模型应用开发
算法岗 / NLP / 大模型算法
interview-note style content

It now uses a two-step strategy:

broad Xiaohongshu search queries
local scopeFilter narrowing to the exact Agent/LLM + 算法岗 + 大厂 + 2026-02-01+ slice

This is intentional because overly narrow XHS queries like 腾讯 agent 算法面经 were timing out in practice.

Each ralph-loop cycle now reads control-status, selects one operation, and dispatches it through run-operation rather than stepping through a fixed stage list.

Primary broad query family includes:

腾讯面经
字节面经
阿里面经
美团面经
百度面经
京东面经
快手面经
LLM 算法面经
LLM 面经
智能体面经
Agent 面经
算法面经
NLP 面经

Persisted outputs land in that workspace:

interview_data/xhs_notes.json
interview_data/xhs_questions.json
reports/xhs-agent-algo-feb2026/

Curated filtered outputs land here:

reports/xhs-agent-algo-feb2026/scope_candidates.json
reports/xhs-agent-algo-feb2026/scope_candidates.md

If live search keeps timing out, recommended flow is:

seed-import from an existing xhs_notes.json
export
status
resume ralph-loop only after opencli xiaohongshu search is stable again

Recommended ways to monitor it:

tmux attach -t interviewops-agent-llm-algo

sed -n '1,120p' /tmp/interviewops/agent-llm-algo-loop.log
sed -n '1,120p' ./workspaces/xhs-agent-algo-feb2026/reports/xhs-agent-algo-feb2026/ralph-loop.log

And inspect structured status:

npm run dev -- status --workspace ./workspaces/xhs-agent-algo-feb2026

LaTeX Template

Bundled assets:

Copy them into your workspace:

npm run dev -- template

Auto Commit

By default the SDK does not auto-commit.

Enable it per command:

npm run dev -- cycle --auto-commit
npm run dev -- nightly 8 --auto-commit

Tests

npm test
npm run typecheck

Name		Name	Last commit message	Last commit date
Latest commit History 27 Commits
examples		examples
interview_data		interview_data
reports/xhs-miangjing		reports/xhs-miangjing
schemas		schemas
scripts		scripts
src		src
templates		templates
workspaces/xhs-agent-algo-feb2026		workspaces/xhs-agent-algo-feb2026
.gitignore		.gitignore
README.md		README.md
package-lock.json		package-lock.json
package.json		package.json
tsconfig.json		tsconfig.json

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Repository files navigation

InterviewOps SDK

What It Does

Install

CLI

Control Plane

OpenCLI Integration

OMX Stabilization

Default Layout

Workspace Init

Example PRD

Seller / Lead-Gen Marking

Purchase Link Detection

Source Adapters

Dedicated Agent / LLM Algorithm Workspace

LaTeX Template

Auto Commit

Tests

About

Uh oh!

Releases 1

Packages

Uh oh!

Contributors

Uh oh!

Languages

Folders and files

Latest commit

History

Repository files navigation

InterviewOps SDK

What It Does

Install

CLI

Control Plane

OpenCLI Integration

OMX Stabilization

Default Layout

Workspace Init

Example PRD

Seller / Lead-Gen Marking

Purchase Link Detection

Source Adapters

Dedicated Agent / LLM Algorithm Workspace

LaTeX Template

Auto Commit

Tests

About

Resources

Uh oh!

Stars

Watchers

Forks

Releases 1

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

Packages