Openorder

OpenOrder — Investment Research Wiki Skill

Order out of investment chaos. An open AI-maintained research wiki that compounds your edge across every conversation.

1. Core Mission

OpenOrder turns scattered investment research into a persistent, compounding wiki maintained by your AI agent.

Following the Karpathy LLM Wiki pattern:

• The agent is the maintainer; you are the prompter + source curator
• Knowledge is compiled once, kept fresh — not re-derived on every query
• The wiki is a persistent, compounding artifact — every conversation makes it smarter

Every conversation that touches a ticker / industry / earnings / framework MUST execute:

READ → REASON → WRITE → LOG → (periodically) LINT
 ↓        ↓        ↓       ↓             ↓
look up  combine  archive  timeline    health-check
existing real-time new       append-only contradictions,
data     + history insights  record       gaps, staleness

---

2. Three-Layer Architecture

${OPENORDER_HOME}/
│
├── raw/                        ← Layer 1: Raw Sources (immutable, read-only)
│   ├── earnings/                  Earnings transcripts / 8-K filings
│   ├── articles/                  Articles, tweets, blog posts
│   ├── filings/                   SEC 10-K/10-Q PDFs
│   ├── research-notes/            Raw research notes from conversations
│   └── README.md
│
├── [Wiki Layer]                 ← Layer 2: Wiki (LLM-owned, continuously maintained)
│   ├── INDEX.md                   📌 Content index (always read first)
│   ├── log.md                     📌 Activity log (append-only timeline)
│   ├── README.md
│   ├── companies/{TICKER}.md      Company profiles
│   ├── industries/{NAME}/         Industry deep-dives
│   ├── frameworks/{NAME}.md       Investment frameworks
│   ├── earnings/{TICKER}-{Q}.md   Earnings deep-dives
│   ├── portfolios/{NAME}.md       Portfolio configurations
│   └── templates/                 Templates for new entries
│
└── [Schema Layer]               ← Layer 3: Schema (this file)
    ~/.claude/skills/openorder/SKILL.md (or equivalent for other agents)

Layer rules:

• Raw is immutable: only append metadata headers; never modify content
• Raw must produce Wiki updates: every raw file must trigger at least one wiki page update (otherwise it's dead data)
• Wiki references Raw: use relative paths [source](../raw/earnings/{TICKER}-{Q}.txt)
• Schema rules everything: this SKILL.md defines all rules

---

3. Mandatory Trigger Scenarios

🟢 MUST trigger (no user reminder needed)

The first action MUST be Read ${OPENORDER_HOME}/INDEX.md when any of the following appears:

3.1 Ticker / company mentions

• Any equity ticker: US (4-letter), HK (4-digit), A-share (6-digit + .SZ/.SH), TW (.TW/.TWO), JP (.JP), KR, etc.
• Any company name (English or local language)
• Crypto: any token symbol (BTC, ETH, SOL, etc.) or protocol name

3.2 Industry / theme keywords

Replace this list with your own domain. Defaults below cover common AI / hard-tech / energy themes:

• Semiconductors: foundry, fab, EUV, lithography, advanced packaging, CoWoS, chiplet
• AI infra: hyperscaler capex, data center, scale-out, scale-up, NVLink, InfiniBand
• Photonics: optical interconnect, CPO, CW laser, EML, DFB, VCSEL, InP, SiPh
• Memory: HBM, HBM3E, HBM4, DDR5
• Energy / storage: SOFC, fuel cell, BESS, sodium-ion, long-duration storage
• Crypto: L1/L2, DeFi, perp DEX, stablecoin, restaking
• (extend with your own domain keywords in the customization layer below)

3.3 Framework / methodology keywords

• chokepoint, bottleneck, moat, value chain, supply chain
• super-cycle, supply-constrained, structural shortage
• target price, valuation, P/E, EV/EBITDA, DCF
• thesis, conviction, base/bull/bear case

3.4 Earnings / financial keywords

• earnings, guidance, beat, miss, transcript, 8-K, 10-Q, 10-K, FY/CY quarters
• any concrete metric (revenue, gross margin, EPS, capex, backlog, inventory) tied to a company

3.5 Trading / portfolio keywords

• "buy", "sell", "trim", "add", "short", "hedge"
• "position", "weight", "allocation", "portfolio", "basket"

3.6 Explicit operation keywords

• Ingest: "read this article", "ingest", "archive", "save this source"
• Lint: "lint wiki", "health check", "find contradictions", "self-audit"
• Query: any normal question (default mode)

🟡 Suggested triggers

• "value chain", "upstream/downstream", "picks and shovels"
• "AI beneficiaries", "second derivative plays"
• Any "X vs Y" comparison

---

4. Operation 1: READ (mandatory every time)

Trigger → Read INDEX.md (mandatory, ~5s)
    ↓
Determine relevance, then read as needed:
    ├── Company question → companies/{TICKER}.md
    ├── Industry question → industries/{NAME}/README.md
    ├── Framework question → frameworks/{NAME}.md
    ├── Historical earnings → earnings/{TICKER}-{Q}.md
    ├── Portfolio → portfolios/{NAME}.md
    └── Timeline → log.md (newest first)
    ↓
When answering:
    1. Cite information timestamps ("based on YYYY-MM-DD data")
    2. Reference specific files ("see companies/{TICKER}.md")
    3. If data is stale → use WebSearch to confirm current state

---

5. Operation 2: INGEST (when user provides a new source)

Triggered by: user pastes a URL, sends a tweet/article, says "read this", "ingest", "archive".

Standardized 5-step flow:

Step 1. Fetch full content
   ├── URL: WebFetch / Browser
   ├── Tweet: browser MCP
   ├── Earnings: pull from filing / SEC direct link
   └── Write to raw/<subdir>/ (immutable; only add metadata header)

Step 2. Extract entities + key findings
   ├── Which companies are mentioned? (existing/new in wiki)
   ├── Which industries / concepts?
   ├── Key numbers, quotes, contradictions
   └── Confirm with user what to emphasize

Step 3. Update Wiki (may touch 10-15 files)
   ├── New company → use templates/company-template.md to create
   ├── Existing company → update "analysis log", revise bull/bear, adjust scores
   ├── Industry page → add "event tracking" entry
   ├── Framework → if validated/challenged, update framework file
   └── Each file: top "Last updated" date + bottom changelog entry

Step 4. Append log.md (mandatory)
   Add entry:
   ## [YYYY-MM-DD HH:MM] ingest | {title}
   - Source: raw/.../{file}
   - Key findings: ...
   - Files touched: companies/X.md, industries/Y.md, ...

Step 5. Update INDEX.md
   ├── Update top "Last updated"
   ├── Update "Wiki navigation" table (if new files)
   └── Update bottom "Update history"

After ingest, tell the user:

• Which files were modified (so they can review in Obsidian)
• Whether new findings need their decision

---

6. Operation 3: WRITE (when conversation produces new insights)

6.1 When to write

Trigger event	Required write
User mentions an uncatalogued company	Immediately create companies/{TICKER}.md (use template)
New earnings released or discussed	Create earnings/{TICKER}-{Q}.md + update company profile
Conversation produces a new insight/analysis	Append to relevant file's "Analysis log" + log.md + INDEX.md
User makes a trading decision (buy/sell)	Create or update portfolios/{NAME}.md + log.md
Industry change (M&A, partnership, product launch)	Update industries/{NAME}/README.md "Event tracking"
Framework validated or revised	Update relevant frameworks/ file + log.md

6.2 Writing rules

Required metadata

• File top: > **Last updated**: YYYY-MM-DD
• File bottom: changelog (newest first)
• INDEX.md: update history + todos
• log.md: append one entry

Writing style

• Concrete data: not "revenue grew strongly" — say "revenue $1.81B (+27% YoY)"
• Cite sources: not "analysts think" — say "GS report 2026/3" or "CEO on Q3 FY26 transcript", with raw path
• Flag assumptions: model predictions must say "assuming X conditions"
• Bull/bear balance: every company profile MUST have both bull and bear arguments

Naming conventions

• Company: companies/{TICKER}.md (e.g. AAPL.md, 0700-HK.md, 688012-SH.md)
• Earnings: earnings/{TICKER}-{QUARTER}.md (e.g. AAPL-Q1-FY26.md)
• Industry: industries/{NAME}/README.md
• Framework: frameworks/{NAME}.md (kebab-case)
• Portfolio: portfolios/{NAME}-{DATE}.md
• Raw: see raw/README.md

6.3 Templates

See templates/company-template.md and templates/earnings-template.md.

---

7. Operation 4: LOG (append on every write)

Location: ${OPENORDER_HOME}/log.md

Format (Karpathy convention — uniform prefix for grep):

## [YYYY-MM-DD HH:MM] {action} | {one-line summary}
- Trigger: {user prompt or event}
- Files touched: {file1}, {file2}, ...
- Key findings/decisions: {1-3 lines}
- Raw source: {path if applicable}

action values:

• ingest — new source absorbed
• query — user Q&A (only log if it produced new insight)
• lint — health check
• decision — real trading decision
• revise — corrected old thesis
• framework — framework update

How to append: use StrReplace at the top of the # Wiki Activity Log heading (newest first).

Benefit: grep "^## \[" log.md | tail -10 instantly shows recent activity.

---

8. Operation 5: LINT (periodic or explicit)

Triggered by: user says "lint wiki", "health check", "find contradictions"; or proactively offer monthly.

Lint 5 checks:

8.1 Contradictions

• Same company has conflicting target price / rating in companies/X.md vs portfolios/Y.md?
• Industry ranking in industries/.../README.md conflicts with frameworks/.../*.md scores?
• Same metric different across multiple earnings/ files?

8.2 Staleness

• Company profile Last updated predates its latest earnings → flag for update
• Earnings file >90 days old but catalysts have passed → need a new earnings file?

8.3 Orphans

• Files in companies/ not referenced by any portfolio / framework / industry?
• Raw sources without a corresponding wiki update?

8.4 Missing pages

• Companies/concepts mentioned multiple times across files but no profile (e.g. "Anritsu" cited 5 times but no companies/Anritsu.md)?
• Catalyst tables mention "watch X in YYYY/MM" — date passed without follow-up?

8.5 Data gaps

• Companies that haven't been refreshed in 30+ days
• Industries lacking a value-chain mermaid diagram

Lint output: a markdown report listing all issues + suggested next actions. Does not auto-modify — user decides.

---

9. Cross-Agent Compatibility

OpenOrder works with any AI agent that supports either SKILL.md or AGENTS.md:

Agent	Skill mechanism	Install location
Claude Code (Anthropic)	SKILL.md (native)	~/.claude/skills/openorder/
Cursor (IDE + CLI)	SKILL.md (compatible)	~/.cursor/skills-cursor/openorder/
Codex CLI (OpenAI)	AGENTS.md or SKILL.md	~/.codex/skills/openorder/
OpenCode (sst)	AGENTS.md	.opencode/agent/ or repo root
Hermes Agent (NousResearch)	AGENTS.md + ~/.hermes/skills/	~/.hermes/skills/openorder/
OpenClaw	AGENTS.md	per project config
Aider	CONVENTIONS.md	repo root (symlink to SKILL.md)
Cline / Roo Code	.clinerules	repo root

The repository ships both SKILL.md (this file) and AGENTS.md (mirror). The install.sh script auto-detects installed agents and creates the appropriate symlinks.

---

10. Special Scenarios

10.1 Uncatalogued company

1. WebSearch latest financials + industry position
2. Use templates/company-template.md to create companies/{TICKER}.md
3. Add to INDEX.md
4. Append log.md ingest entry
5. Answer the user

10.2 Stale data

1. WebSearch confirms latest data
2. Update file + INDEX.md timestamp
3. Note in changelog: "YYYY-MM-DD: data updated from X to Y"
4. Append log.md revise entry

10.3 Conflict with new data

1. New data wins
2. Note conflict reason in changelog
3. If thesis was wrong → revise bull/bear arguments
4. Append log.md revise entry

10.4 Real trading decision

1. Create or update portfolios/{NAME}.md
2. Record: entry/exit price, size, time, rationale
3. Append log.md decision entry
4. Add "review reminder" todo to INDEX.md

---

11. Self-check Checklist

Before ending any conversation that touched stocks / industries:

• Did I read INDEX.md?
• Did I read the relevant company profile(s)?
• Did I answer using wiki content?
• Are there new data / insights that need to be written back?
• Is there a raw source to ingest?
• Does log.md need a new entry?
• Does INDEX.md need updating?
• Are there new todos?

If any answer is "no but should be yes", do it now.

---

12. Anti-patterns to avoid

❌ Anti-pattern 1: Answer without reading the wiki

Wrong: "AAPL is a consumer hardware company..." (from memory) Right: Read INDEX.md → Read companies/AAPL.md → answer based on wiki

❌ Anti-pattern 2: Read but don't update

Wrong: Use stale data and stop Right: If conversation produced new conclusion → append to file + log.md + INDEX.md

❌ Anti-pattern 3: Create file but don't update INDEX

Wrong: Create companies/NEW.md but INDEX.md unchanged Right: Every new file → update INDEX navigation table + log.md

❌ Anti-pattern 4: Write data without timestamps

Wrong: "revenue +20% YoY" (which quarter?) Right: "Q3 FY26 revenue $1.81B (+20.5% YoY), reported YYYY-MM-DD"

❌ Anti-pattern 5: Bull-only profiles

Wrong: List only positives Right: Every profile must have a bear case section

❌ Anti-pattern 6: Skip Raw layer when ingesting

Wrong: Summarize an article straight into wiki without storing the original Right: First save raw/articles/{date}-{slug}.md, then wiki references it

❌ Anti-pattern 7: Wiki changes without log.md entry

Wrong: Modify 5 files, log.md unchanged Right: Every write operation must append log.md (even one line)

---

13. Customization (fork to your domain)

OpenOrder ships investment research as the default domain. To adapt:

1. Edit Section 3 trigger keywords to match your field (e.g. biotech, crypto, real estate)
2. Edit templates/ to fit your entity type (instead of "company profile" → "drug pipeline" / "protocol" / "property")
3. Edit frameworks/ to fit your scoring methodology
4. Keep Sections 2/4/5/6/7/8 unchanged — these are domain-agnostic operations

See docs/customize.md for examples adapting OpenOrder to:

• Crypto research wiki
• Academic literature review
• Personal journaling / quantified self
• Internal team knowledge base

---

14. External Tools Integration

Tool	Role	How agent uses it
Local markdown (this wiki)	Primary store	Read / Grep / Glob
git	Version history + collaboration	git log + auto-commit hook
Obsidian	Visual browse + bidirectional links	Vault → ${OPENORDER_HOME}
WebSearch / Browser MCP	Real-time data + ingest	Standard tools

---

15. Version History

Version	Date	Changes
v1.0.0	First public release	Open-sourced as OpenOrder

---

16. Credits

OpenOrder's three-layer architecture (Raw / Wiki / Schema), log.md timeline, Lint operation, and Ingest workflow are directly inspired by Andrej Karpathy's LLM Wiki pattern: https://gist.github.com/karpathy/442a6bf555914893e9891c11519de94f

The investment-research domain framing, chokepoint scoring methodology, and cross-agent install scaffolding are original to OpenOrder.

License: MIT. See LICENSE.