Serenity Skill

Agent Skill + browser research agent for Serenity (@aleabitoreddit) — distilled tweet corpus, live market data, attention radar, and supply-chain bottleneck workflows

Demo · Quick start · Architecture · Query flow · Query modes · 中文

Demo

Full walkthrough: ticker view, live quotes, thesis cards, and agent narrative in the browser UI.

Note: GitHub README does not play inline <video> — use the link below (or add an optional short GIF at docs/assets/serenity-skill-demo.gif for autoplay in-page).

▶ Watch full demo (MP4) · python aio_serenity.py · research support only

Research support only. Ranked priorities and reasoning — not buy/sell instructions, not auto-trading. Not affiliated with @aleabitoreddit.

Disclaimer — research distill, not Serenity herself

Serenity Skill is an independent research tool (OlaXBT). It is not Serenity (@aleabitoreddit), does not speak for her, and does not impersonate her.


What it is	A pipeline that distills her public posts and articles into a structured, queryable corpus + live-aware research workflows
What it is not	Her real-time view, an official product, investment advice, or trade execution
How to read outputs	Labelled Serenity corpus view vs live verification vs research map — always cross-check price, news, and thesis age
Corpus limits	Bundled archive may lag; views evolve; distilled bullets can be incomplete or stale

This disclaimer appears in the browser UI (footer + empty state), in every report (footer line), and in SKILL.md agent rules.

Agent answer quality (browser LLM)

Quality is controlled in layers — not one magic prompt:

Layer	What it guarantees	Command / file
Corpus & scripts	Deterministic thesis, radar, live quote/news	`python scripts/run_qc.py`, `pytest tests/`
Structured report	Tables, charts, thesis cards before any LLM text	`serenity_twin/ui_render.py`
Agent system prompt	Language, section headers, evidence rules, no buy/sell	`serenity_twin/agent_prompt.py`
Context boundary	LLM sees only executed JSON context (no invented tickers)	`serenity_twin/llm_stream.py`
Locale check	Flags Chinese headers in English answers	`serenity_twin/agent_output.py`
Human review	Corpus distill + stale thesis cross-check	`distillation/MAINTENANCE.md`

Re-run python -m pytest tests/ -q after corpus edits. Tune narrative quality via agent_prompt.py and temperature in llm_stream.py (default 0.3).

At a glance


What	Serenity Skill — Agent Skill (`SKILL.md`) + Python toolkit + browser UI for Serenity-style bottleneck investment research
Primary question	What does Serenity think about ticker X — and is that view still valid today?
One command	`python aio_serenity.py` — auto-init + browser research agent (OlaXBT)
Stack	Python 3.10+ (stdlib core), optional DeepSeek / X API, any Agent-compatible IDE or OpenClaw
Repo	github.com/olaxbt/serenity-skill
Maturity	8.9 / 10 — production-ready research MVP (details)

What you can do

Capability	Example prompt
Her conviction on a ticker	What is Serenity's view on $SIVE? Stance, tier evolution, key risks.
Attention Radar	14-day heating & new entrants — cross-check theses.
Map a theme to bottlenecks	Deep-scan A-share AI semiconductors — scarce layers first, then stocks.
One-page research memo	$SIVE thesis memo with evidence ladder and falsifiers.
Learn her research method	Serenity-style bottleneck research — one question at a time.

Full prompt catalog: docs/sample_prompts.md

Why Serenity Skill exists

Serenity's research lens is distinctive: trace hyperscaler capex upstream to the single chokepoint — sole or near-sole supply, hard to design around, often still small-cap. Her public feed is high-volume, multi-ticker, and views evolve over time.

Serenity Skill packages four things other tools don't combine:

Memory — 5,800+ tweet archive, 43 deep thesis tickers, methodology, track-record, articles
Workflows — SKILL.md routes Agent queries through deterministic scripts + evidence rules
Live world — auto-fetched quotes, news, SEC (no need to say "search the internet")
Radar — mention analytics for Heating / new entrants / theme rotation

Quick start

Requirements

Python 3.10+ — core scripts use stdlib only (no pip install required)
Optional: DeepSeek in .env (browser UI LLM) or your IDE’s model settings (agent chat)
Optional: X_BEARER_TOKEN for live tweet sync

One command (recommended)

git clone https://github.com/olaxbt/serenity-skill.git
cd serenity-skill
python aio_serenity.py

Step	What happens
First run	Auto-runs `init_system.py` — validate skill, normalize corpus, split theses, rebuild mentions, QC, install agent skill, seed `.env`
Every run	Opens UI at `http://127.0.0.1:17876` — system browser or IDE embedded preview
Each prompt	Server auto-runs `lookup_ticker.py` + `live_research.py` + structured HTML report

You never manually run lookup or live-research per question — the UI and Agent do that.

python aio_serenity.py --init          # init only
python aio_serenity.py --port 3000     # custom port
python aio_serenity.py --open cursor   # IDE embedded browser (e.g. VS Code / Cursor Simple Browser)
python aio_serenity.py --open browser  # system browser
python aio_serenity.py --no-browser    # headless server — URL in terminal

IDE preview: point your IDE’s simple browser at http://127.0.0.1:17876 while the server is running (not the raw index.html file).

.env: remove # from the key line — # DEEPSEEK_API_KEY=... is a comment and is ignored.

More: docs/QUICKSTART.md

Three ways to interact

Surface	Command / trigger	Best for
Browser agent UI	`python aio_serenity.py`	Testing prompts, tables, price charts, bilingual UI, SSE streaming
IDE agent chat	Load `SKILL.md` + natural question	Deep research sessions, web search, editing corpus
OpenClaw	Install skill + gateway web tools	24/7 cron, Telegram briefs (`docs/SETUP.md`)

All surfaces share the same Python scripts and corpus. Only the LLM narration layer differs (DeepSeek in browser vs your IDE agent model in chat).

Architecture

Four layers — no separate runtime per surface:

Layer	Role	Key paths
0 — Corpus memory	Distilled tweets, theses, methodology, track-record	`corpus/data/`, `corpus/references/`
1 — Python tools	Deterministic lookup, radar, live web, sync, distill	`scripts/`, `serenity_twin/`
2 — Live world	Yahoo quotes (incl. crypto spot aliases e.g. `BTC-USD`), news, SEC	`live_research.py`, `web_research.py`
3 — Agent reasoning	Mode router + optional LLM synthesis	`SKILL.md`, `agent_prompt.py`, `ui_chat.py`

Extended workflows (theme scans, evidence ladder, A-share playbook) live under reasoning/references/.

Full design doc: docs/ARCHITECTURE.md

Query flow

One prompt → routed mode → scripts execute → structured report → optional agent narrative. The browser streams progress over SSE (route → corpus → live web → render → LLM).

%%{init: {'theme': 'base', 'themeVariables': {
  'fontFamily': 'Inter, system-ui, sans-serif',
  'primaryColor': '#f8f4ff',
  'primaryBorderColor': '#e781fd',
  'primaryTextColor': '#2d2640',
  'secondaryColor': '#ffffff',
  'tertiaryColor': '#faf8fc',
  'lineColor': '#a78bfa',
  'clusterBkg': '#fafafa',
  'clusterBorder': '#e8e0f0'
}}}%%
flowchart LR
  P(["Your prompt"]) --> R{"Route<br/>A–E · brief"}
  R --> S["Scripts<br/>lookup · radar · live web"]
  S --> M["Merge corpus + live JSON"]
  M --> K{"Thesis stale?"}
  K -->|yes| W["Stale alert"]
  K -->|no| H["Structured report"]
  W --> H
  M --> L{"DeepSeek<br/>configured?"}
  L -->|yes| A["Agent narrative<br/>streams first"]
  L -->|no| H
  A --> O(["Answer in UI"])
  H --> O

Report layout (v0.3.10):

Agent answer — LLM synthesis at the top when DEEPSEEK_API_KEY is set (locale follows your prompt, not the UI toggle)
Supporting data — live quote table + chart, thesis cards with tiered evidence, radar tables, stale warnings
References — tweets, web sources, SEC — collapsed by default
Disclaimer — footer on every report

Without DeepSeek, step 1 is omitted and a deterministic synthesis block may appear instead.

Fresh ticker (never in corpus): lookup_ticker.py returns found_in_theses = false → follow methodology.md 14-question checklist + live_research.py → output is independent analysis, not Serenity's stated view.

Query modes A–E

Mode	Trigger	Scripts (auto)	Output
A — Ticker view	`$TICKER`, Serenity's view, fresh-name / methodology checklist	`live_research` → `lookup_ticker`	Corpus stance + live verification + agent narrative
B — Radar	ramp, heating, attention	`radar` → live web on top heating names	Heating / new entrants / conviction / theme rotation tables
C — Theme scan	supply chain, A-share, ETF	`live_research --theme` + workflow	Layer ranking → stock list → optional ETF holdings check
D — Research memo	深度研报, thesis memo	Mode C/A + template	Full memo: system change → bottleneck → evidence → falsifiers
E — Learning	teach me the method	methodology files	One question per turn; tickers as examples only
brief	daily brief	`daily-brief-latest.txt` + radar	Snapshot table from scheduled refresh

Browser agent UI

Light-mode interface (purple accent #e781fd, v0.3.10) with English / 中文 chrome toggle. Built and maintained by OlaXBT — free for the dev community.

Feature	Detail
Entry	`python aio_serenity.py`
Default URL	`http://127.0.0.1:17876` (auto next port if busy)
Agent plan UX	Step list (route → corpus → live web → render → LLM) with streaming progress
Auto live web	Yahoo quote, 3M chart, news search, SEC — every analysis prompt; crypto uses spot symbols (`BTC-USD`, not ETF tickers)
Structured output	Tables, metric cards, thesis cards — references demoted to collapsible section
Agent narrative	Auto-enabled when `DEEPSEEK_API_KEY` is in `.env` — rendered as markdown at top of report
Answer locale	Detected from your prompt text — English prompts get English section headers
Task-oriented prompts	Sidebar labels describe research tasks — see `ui/prompts.json`
Session history	SQLite at `corpus/data/sessions.db` (v0.3+)
IDE-only models	Not available in browser — use agent chat + `SKILL.md` in your IDE instead

Why can't the browser use IDE built-in models?

The browser UI is a standalone Python server (aio_serenity.py). It cannot call models that are only available inside an IDE agent runtime.

Surface	Agent narrative	How
Browser UI	Needs `DEEPSEEK_API_KEY` in `.env`	Python server calls DeepSeek directly
IDE agent	Uses your IDE’s configured model	Load `SKILL.md` → ask in agent mode — no browser API key

Optional: live tweet sync

Default off. Without X_BEARER_TOKEN, bundled corpus works; sync exits cleanly with status: disabled.

cp .env.example .env
# X_BEARER_TOKEN=...

Daily automation (Windows): scripts/daily_brief.ps1

Corpus

Path	Contents
`corpus/data/tweets.json`	Canonical archive — 5,826 posts
`corpus/references/theses/*.md`	Sector thesis files — 43 deep tickers in index
`corpus/references/methodology.md`	14 transferable principles + runnable checklist
`corpus/references/track-record.md`	Dated calls + calibration
`corpus/data/mentions-*.csv`	Mention analytics — 724 tickers

Python scripts

Script	Purpose	Network
`aio_serenity.py`	All-in-one init + browser UI	yes (UI)
`scripts/lookup_ticker.py`	Thesis + tweets + radar hint	no
`scripts/live_research.py`	Quote + news + SEC	yes
`scripts/radar.py`	Attention momentum	no
`SKILL.md`	Agent Skill entry (IDE + OpenClaw)	—

API keys

Use	Where
IDE agent chat LLM	Your IDE’s model settings
Browser UI LLM (optional)	`.env` → `DEEPSEEK_API_KEY`
Live tweet sync (optional)	`.env` → `X_BEARER_TOKEN` + `config.json`

Maturity & quality grade

Overall: 8.9 / 10 — shippable research-agent MVP, not a fully autonomous trader.

Dimension	Score	Notes
Agent Skill (`SKILL.md`)	★★★★★	Modes A–E, mandatory live verification, script-first rules
Browser UI	★★★★☆	v0.3.10 — agent-first layout, streaming, EN/中文, sessions
Corpus & tooling	★★★★☆	5.8k tweets, 43 deep theses, lookup/radar/distill/QC
Tests & evals	★★★★☆	47 pytest incl. E2E output, stale, session store, UI formatting

Provenance

Unified from three open-source Serenity skill projects:

Source	Contribution
yan-labs/serenity-aleabitoreddit	Tweet archive, theses, methodology, track-record
lanfuli/aleabito-serenity-skills	Radar patterns, method framework
muxuuu/serenity-skill	A-share/HK workflow, scorecard

Publish only olaxbt/serenity-skill.

Keywords & discoverability

Serenity Skill (olaxbt/serenity-skill) is an open-source Agent Skill and investment research agent for followers of @aleabitoreddit (Serenity).

Search terms: serenity skill, serenity aleabitoreddit, what does Serenity think about, Serenity ticker thesis, CPO bottleneck stocks, attention radar investing, supply chain bottleneck research, agent skill investing.

Capabilities: supply-chain bottleneck analysis · CPO / optical / semiconductor theses · attention radar · A-share and US equity theme scans · thesis memos · live market verification (Yahoo quotes, news, SEC) · browser UI · IDE agent (SKILL.md) · OpenClaw.

The browser UI may show “Serenity Twin” internally — the public project name is Serenity Skill (this repo).

FAQ

Question	Answer
What is Serenity Skill?	An Agent Skill (`SKILL.md`) + Python toolkit that distills @aleabitoreddit’s public research into queryable corpus + live-aware workflows.
How do I ask “what does Serenity think about $TICKER”?	Run `python aio_serenity.py` or load `SKILL.md` in your IDE agent — scripts auto-fetch thesis + live quotes.
Is this official or affiliated with Serenity?	No — independent research distill by OlaXBT. Not investment advice.
IDE agent vs browser UI?	Same scripts; IDE agent uses your configured model, browser UI uses optional `DEEPSEEK_API_KEY`.

GitHub repo topics (suggested): agent-skill, investment-research, aleabitoreddit, supply-chain, semiconductor, cpo, python, openclaw.

Disclaimer

Serenity's self-reported returns are unverified; public feeds have survivorship bias
Many names are volatile micro/small-caps; theses decay — confirm current fundamentals
Social posts are leads; high-confidence claims require filings and exchange disclosures
This project is a research lens, not a signal feed or trading bot

License

MIT

简体中文 README →

Navigation Menu

Repository navigation

About

Repository files navigation

Serenity Skill

Demo

Disclaimer — research distill, not Serenity herself

Agent answer quality (browser LLM)

At a glance

What you can do

Why Serenity Skill exists

Quick start

Requirements

One command (recommended)

Three ways to interact

Architecture

Query flow

Query modes A–E

Browser agent UI

Why can't the browser use IDE built-in models?

Optional: live tweet sync

Corpus

Python scripts

API keys

Maturity & quality grade

Provenance

Keywords & discoverability

FAQ

Disclaimer

License

About

Releases

Packages

Contributors

Languages

Name		Name	Last commit message	Last commit date
corpus		corpus
distillation		distillation
docs		docs
evals		evals
examples		examples
reasoning		reasoning
scripts		scripts
serenity_twin		serenity_twin
tests		tests
ui		ui
.env.example		.env.example
.gitignore		.gitignore
CONTRIBUTING.md		CONTRIBUTING.md
LICENSE		LICENSE
README.md		README.md
README.zh-CN.md		README.zh-CN.md
SKILL.md		SKILL.md
aio_serenity.py		aio_serenity.py
config.json		config.json
install-cursor-skill.ps1		install-cursor-skill.ps1
requirements-dev.txt		requirements-dev.txt
requirements.txt		requirements.txt

Latest commit

History

Folders and files

Repository files navigation

Serenity Skill

Demo

Disclaimer — research distill, not Serenity herself

Agent answer quality (browser LLM)

At a glance

What you can do

Why Serenity Skill exists

Quick start

Requirements

One command (recommended)

Three ways to interact

Architecture

Query flow

Query modes A–E

Browser agent UI

Why can't the browser use IDE built-in models?

Optional: live tweet sync

Corpus

Python scripts

API keys

Maturity & quality grade

Provenance

Keywords & discoverability

FAQ

Disclaimer

License

About

Resources

Contributing

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages