Add serenity-radar skill; make Serenity skills bilingual

- serenity-radar (Route C): data-driven attention radar + candidate
  generator over the full mention archive (radar.js, patterns, signals)
- serenity-method / serenity-radar / follow-aleabito: output now
  follows the user's language (中文 default, English on request) with
  bilingual labels and disclaimer

Mirrors the public suite at github.com/lanfuli/aleabito-serenity-skills

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Author: lanfuli

skills/follow-aleabito/SKILL.md

@@ -7,7 +7,9 @@ description: Track Serenity / @aleabitoreddit on X and turn the feed into (1) a
 
 ## Purpose
 
-Track `@aleabitoreddit` on X and produce Chinese deliverables grounded only in fetched posts. Always include source URLs. Treat the content as market commentary, not investment advice.
+Track `@aleabitoreddit` on X and produce deliverables grounded only in fetched posts. Always include source URLs. Treat the content as market commentary, not investment advice.
+
+**Language / 语言:** deliverables default to **中文** (the digest and Xiaohongshu workflows are Chinese-first by design), but any of them can be produced in **English** when the user asks. The analytics CSVs and research map are language-neutral.
 
 This skill handles four workflows. Pick the one(s) the user asked for and read the matching reference file only when needed:
 

skills/serenity-method/SKILL.md

@@ -16,19 +16,21 @@ This skill is **method only**. It does not fetch data. To pull her live posts or
 - "Analyze $X the way Serenity / aleabito would."
 - "Is $X a critical chokepoint? Map its supply chain."
 - "Give me a first-principles + Buffett judgment on $X."
-- Turning a Serenity post (or any thesis) into a structured, beginner-friendly Chinese analysis.
+- Turning a Serenity post (or any thesis) into a structured, beginner-friendly analysis (中文 or English).
 
 ## What you produce
 
-For each ticker or theme, output these blocks **in Chinese**, beginner-friendly, defining jargon on first use (see `references/glossary.md`):
+**Language / 语言:** respond in the **user's language** — 中文 by default, **English** when the request is in English or the user asks. The structure is identical in both; bilingual block labels are shown below.
 
-1. **核心论点 / 她的观点** — the one-paragraph thesis (if analyzing a Serenity post, ground it in the post + cite the source URL; if analyzing your own idea, state it plainly).
-2. **小白解释** — re-explain in plain language a beginner can follow.
-3. **第一性原理** — decompose with the five levers (Step 2).
-4. **Buffett 直接判断** — answer the five fields (Step 3). *Answer them, do not pose them as questions.*
-5. **当前结论** — classify: `研究地图`(a lead worth tracking)vs `可投资结论`(only after moat + financials + valuation + margin-of-safety work). Default to `研究地图`.
+For each ticker or theme, output these blocks, beginner-friendly, defining jargon on first use (see `references/glossary.md`):
 
-End every deliverable with: **仅作信息跟踪，不构成投资建议。**
+1. **核心论点 / 她的观点 · Core thesis / Her view** — the one-paragraph thesis (if analyzing a Serenity post, ground it in the post + cite the source URL; if analyzing your own idea, state it plainly).
+2. **小白解释 · Plain-language** — re-explain in plain language a beginner can follow.
+3. **第一性原理 · First principles** — decompose with the five levers (Step 2).
+4. **Buffett 直接判断 · Buffett verdict** — answer the five fields (Step 3). *Answer them, do not pose them as questions.*
+5. **当前结论 · Conclusion** — classify: `研究地图 / research-map`(a lead worth tracking)vs `可投资结论 / investable-conclusion`(only after moat + financials + valuation + margin-of-safety work). Default to `研究地图`.
+
+End every deliverable with the disclaimer in the output language: **仅作信息跟踪，不构成投资建议。** / **For information tracking only; not investment advice.**
 
 For multi-name digests, compress blocks 2–4 into 1–3 paragraphs per name. Use the full template only for a single deep-dive. (This mirrors the dashboard digests in `reports/aleabito-digests/`.)
 
@@ -98,4 +100,4 @@ Default output is **研究地图** (a lead worth tracking), with the specific th
 If the user wants this applied to her *latest* posts, run the `follow-aleabito` skill's fetch first (`scripts/analyze-mentions.js --incremental --resume` for analytics, or `scripts/fetch-updates.js` for raw posts), then apply Steps 1–5 to the returned content. The output structure here is identical to the dashboard digests in `reports/aleabito-digests/`, so results drop straight into that pipeline.
 
 ---
-仅作信息跟踪，不构成投资建议。
+仅作信息跟踪，不构成投资建议。 / For information tracking only; not investment advice.

skills/serenity-radar/SKILL.md

@@ -0,0 +1,67 @@
+---
+name: serenity-radar
+description: Use @aleabitoreddit ("Serenity")'s full mention archive (built by the follow-aleabito skill) to anticipate where her attention is moving and generate candidate ideas in her style. Two modes — (1) RADAR reads the live mention data for attention momentum (which tickers she is heating up on, new entrants, conviction core, theme rotation) via scripts/radar.js; (2) GENERATOR applies her empirically-mined patterns (theme-rotation logic, selection signature, catalyst playbook) to propose her likely next focus. Every candidate is gated through the serenity-method checklist. This is a CANDIDATE GENERATOR + CHECKLIST, never an oracle or buy/sell signal. Trigger on "what is Serenity ramping on / her next pick / aleabito radar / predict her next move / generate ideas like her / 她下一个可能看什么".
+---
+
+# Serenity Radar
+
+A data-driven companion to `serenity-method`. Where `serenity-method` teaches **how she analyzes**, this skill uses her **actual 11-month archive** (2025-07-02 → present, ~6,120 posts / 750 tickers) to estimate **where her attention is going** and to **generate candidates the way she would**.
+
+> **What this is NOT.** Not a predictor, not a buy/sell signal, not "she will pump X next." It is a *candidate generator + checklist*. A single account is fragile; virality ≠ correctness; her archive has survivorship bias (winners get re-cited, losers fade). Read the **Caveats** section before using output. Always end with: 仅作信息跟踪，不构成投资建议。
+
+## Prerequisites
+- The mention archive must exist (the `follow-aleabito` skill produces it). Default path: `$FOLLOW_ALEABITO_REPORTS_DIR/aleabito-mentions-events.csv` (else the workspace `reports/`).
+- Keep it current with `follow-aleabito`'s incremental fetch (`analyze-mentions.js --incremental --resume`) before running radar, so signals reflect the latest days.
+- For the analytical gate, use the local `serenity-method` skill.
+
+## Mode 1 — RADAR (data-driven, run this first)
+Run the signal extractor:
+```bash
+FOLLOW_ALEABITO_REPORTS_DIR="<reports dir>" node skills/serenity-radar/scripts/radar.js --window 14 --top 20
+# add --json for machine-readable output; --asof YYYY-MM-DD to evaluate a past date; --window 7 for a tighter read
+```
+It prints four signal blocks (see `references/signals.md` for the exact math):
+- **🔥 Heating** — tickers whose mention count is *accelerating* (recent window vs prior window). This is the core "she's ramping attention here" signal.
+- **🆕 New entrants** — tickers that first appeared within the window. Candidate *next focus* — she often seeds a name quietly, then ramps.
+- **🎯 Conviction watch** — high recent volume + sustained + still active. Her *core book* right now (defended, repeated).
+- **🔄 Theme rotation** — theme mention-share recent vs prior. Tells you which narrative she is rotating *into* / *out of*.
+
+**How to read it:** a name that is *both* a New entrant *and* Heating, in a theme that is *rotating up*, is the strongest "emerging focus" signal. A Conviction-watch name that is *cooling* (falling out of Heating) may be maturing toward exit/realization. Cross-check a heating name's recent posts (via `follow-aleabito`) to confirm it is a genuine thesis, not a one-off reply.
+
+## Mode 2 — GENERATOR (pattern-driven, for "her likely next move")
+When the user wants ideas she *hasn't surfaced yet*, apply her empirical patterns (full detail in `references/patterns.md`). Her behavior is remarkably consistent; the three levers that predict her next focus:
+1. **Move UP the supply chain** — from today's hot end-product to the upstream chokepoint that isn't priced. (She went interconnect → laser → InP substrate → **red phosphorus**.) Ask: *what is the bottleneck of the current bottleneck?*
+2. **Move EARLIER in the cycle** — front-run a dated catalyst (ETF approval, index inclusion, earnings read-through, government filing, M&A). Ask: *what catalyst is ~1-2 quarters out that the market hasn't mapped?*
+3. **Move SMALLER / less-covered** — toward a sub-$3B, designed-in, often FUD-labelled name. Ask: *who actually does the work (the subsidiary / upstream supplier), not the headline brand?*
+
+Generate 3-5 candidates by running these levers off the current Heating/Conviction themes, then gate each.
+
+## The gate (mandatory for every candidate)
+A radar signal or generated idea is **only a lead**. Before presenting it as a thesis, run it through `serenity-method` (Steps 1-5: chokepoint test → first principles → Buffett quality gate (default `unverified`) → narrative-vs-fundamentals hygiene → classify as `研究地图` vs `可投资结论`). Output should show the candidate **and** its gate result. Never promote a signal to a recommendation.
+
+## Output shape
+**Language / 语言:** respond in the user's language — 中文 by default, English on request. Bilingual labels below.
+
+For each surfaced candidate, give:
+1. **信号 · Signal** — why it surfaced (heating Δ, new entrant since X, conviction core, theme rotating up).
+2. **她的角度(推测) · Her angle (inferred)** — the likely Serenity-style thesis (chokepoint / catalyst / un-priced), clearly marked as inference.
+3. **闸门结果 · Gate result** — the `serenity-method` verdict (almost always `研究地图 / research-map`, with the specific things to verify).
+4. **可信度 · Confidence** — high/medium/low, with the caveat that drove it down (one-off reply, no fundamentals, single-account risk).
+
+## Caveats (read before trusting any output)
+- **Candidate generator, not oracle.** Attention momentum predicts *her interest*, not price or correctness.
+- **Survivorship bias.** Her archive over-weights names that worked; the radar inherits it. Treat "she ramped X and it ran" as *not* evidence X will repeat.
+- **Single-account fragility.** One person, one style, one era (a mostly-AI-up-cycle, though it does include the Nov-2025 drawdown where IREN −38% / NBIS −35% — proof she is *not* infallible and holds through pain).
+- **Reply noise.** A heating name driven by replies (conversation) ≠ a conviction post. Confirm with the source.
+- **No front-running.** This surfaces public attention patterns for research; do not use it to trade ahead of or against anyone, and never emit buy/sell calls.
+
+## References
+| Need | Read |
+| --- | --- |
+| Her empirical patterns: theme-rotation logic, selection signature, catalyst playbook, conviction tells, track record | `references/patterns.md` |
+| Exact radar math + how to read each signal + data caveats | `references/signals.md` |
+| The analytical gate every candidate must pass | the `serenity-method` skill |
+| Keeping the archive current / pulling raw posts | the `follow-aleabito` skill |
+
+---
+仅作信息跟踪，不构成投资建议。 / For information tracking only; not investment advice.

skills/serenity-radar/references/patterns.md

@@ -0,0 +1,58 @@
+# Serenity — Empirical Patterns (mined from the full archive)
+
+Distilled from @aleabitoreddit's complete history (2025-07-02 → 2026-05-30, ~6,120 posts / 750 tickers). These are *observed regularities*, used by the GENERATOR mode to anticipate her likely next focus. They describe her behavior, not market truth. Dates/names are evidence, not endorsements.
+
+---
+
+## 1. Theme-rotation logic (how her focus actually moved)
+The observed sequence, quarter by quarter (top names by mention):
+- **2025 Q3** — memes / value / squeeze / early neocloud: `UPWK, ALAB, HIMS, HOOD, GME, SG, KSPI, NBIS, IREN, RKLB`.
+- **2025 Q4** — neoclouds & AI compute: `NBIS, IREN, CIFR, CRWV, MSFT, META, GOOGL`. (Deep "neocloud GPU economics" margin models.)
+- **2026 Q1** — pivot to photonics / materials bottlenecks: `AXTI, LITE, AAOI, SIVE, COHR, MU`. The **"bottleneck" framework is formally named in her 2026-01-01 "Evolution / Disruption / Bottlenecks" newsletter.**
+- **2026 Q2** — CPO supercycle: `SIVE, LITE, AAOI, AXTI, MRVL, SOI, JBL, XFAB`.
+
+**The rule behind the rotation (use this to predict the next theme):** she consistently moves
+1. **up the supply chain** — end-product → component → material → the material's material (interconnect → laser/`SIVE` → InP substrate/`AXTI`/`IQE` → high-purity red phosphorus / Nippon Chemical 4092, her "bottleneck of the bottleneck");
+2. **earlier in the cycle** — toward names whose catalyst is ~1-4 quarters out and unmapped;
+3. **smaller / less-covered** — sub-$3B, designed-in, often the *subsidiary that does the work* rather than the headline brand (Foxconn → Shunsin 6451).
+
+To generate her likely next focus: take the current hot theme and ask **"what is the un-priced upstream chokepoint of THIS, and who is sole/primary source there?"**
+
+## 2. Selection signature (recurring traits of her picks)
+A name fits her profile when it has most of:
+- **Chokepoint / sole-or-primary source** at a real bottleneck (she says "designed-in", "sole source", "you can't make X without them").
+- **Small/mid-cap, un-priced** vs the opportunity (she repeatedly cites market cap vs TAM; likes <$3B).
+- **Contrarian setup** — high short interest and/or active media FUD ("meme / scam / overvalued"). She is drawn to *being right against the bears* (`SOI`, `RPI`, `HIMS` were all "stupid shorts" to her).
+- **A dated catalyst** (see §3).
+- **First-principles, self-computed case** — she normalizes margins / maps the supply chain herself rather than citing analysts.
+
+## 3. Catalyst playbook (what she front-runs)
+She times entries around concrete, dated events. Recurring types:
+- **ETF approvals / filings** (`LTC` Litecoin-ETF front-run, Sep 2025).
+- **Index inclusion** → forced passive flows (`HOOD` S&P 500; `SIVE` MSCI / Nasdaq inclusion).
+- **Earnings supply-chain read-throughs** (`MRVL` → `SIVE`/Celestial; `MSFT` Maia 300 → `MRVL`/`AAOI`).
+- **Government / regulatory filings** (NIST "only high-volume SiC foundry" → `XFAB`; CHIPS Act 2 blueprints).
+- **M&A / board changes / dual-listing / private placements** (read as capacity-derisk or TAM-expansion signals).
+- **Short-squeeze setups** (very high short interest, e.g. `HIMS`, `RKT`, `DNUT`).
+
+## 4. Conviction tells (how to spot a forming high-conviction name in the data)
+When she is *graduating* a name to a core position, the archive shows:
+- **Mention-velocity ramp** — she starts posting about it repeatedly over days (this is exactly what RADAR's "Heating" detects).
+- **Position-size language** — "bought $Xk", "scaling to $Y", "won't sell a single share", "plan to acquire more" (`UPWK` $150k, `ALAB` $175k→$500k, `SIVE` "won't sell a share").
+- **Repeated defense vs FUD** — she reframes each bear argument as a falsified rung of a "doubt ladder".
+- **Supply-chain cross-linking** — she ties the name into a broader chokepoint map (the strongest tell; e.g. the multi-supplier `SIVE` map).
+
+RADAR's **Conviction watch** + **Heating** together approximate this; confirm with the actual recent posts.
+
+## 5. Track record she cites (context, not proof)
+Self-reported winners she repeats (entry → cited level): `RPI` $280→$800, `SOI` $44→$181, `SIVE` $4→$71, `IQE` $12→$47, `LPK` ~$6→$24, `ALRIB` $5→$15, `AXTI` "10x'd", plus "+900% YTD off 13 names triple-digit in 4 months." **Treat as track-record context only** — survivorship bias is severe (losers fade from the feed). The GENERATOR must never present these as repeatable.
+
+## 6. Anti-patterns (what she avoids / attacks — useful as negative filters)
+She is structurally hostile to, and rarely a long-term holder of:
+- **Zero-revenue hype at huge caps** (she mocked `IONQ`, `OKLO`, `QBTS` as "stupid" comparators).
+- **Heavy serial dilution** (crusaded against `SLNH`, `IREN`'s "$200M cap, $1B dilution"; flags ATMs as overhangs even on names she likes, e.g. `AAOI` $600M ATM).
+- **Paywalled "gurus" / snake-oil TA** (her whole ethos; a name promoted mainly by paywalled callers is a red flag to her).
+A candidate that trips these is *out of character* for her — down-weight it in the GENERATOR.
+
+## 7. Style constants (the through-line, unchanged from Day 0)
+Free, public, real-time, never-deletes, self-does-the-research, transparent positions, anonymous. These don't predict tickers but they calibrate *what kind of idea* she amplifies: original, contrarian, supply-chain-grounded, and explainable from first principles. An idea that can't be explained from first principles is unlikely to be hers.