Clawhub

Index and audit public financial experts' YouTube and Telegram claims: extract source-backed verbatim call cards, search a local SQLite history, and chart recorded price levels against market prices. Use for indexing a supplied source, querying an existing thesis database, or visualizing its recorded levels. Requires the financial-experts-said runtime and database; never invent quotes, source URLs, dates, or investment conclusions.

financial-experts-said

Runtime + issues + releases: github.com/bzSega/financial-experts-said

Financial experts talk every day. Do you remember a month later who promised IMOEX at 2000 — and whether it happened?

This skill turns expert streams and posts into a verifiable history: every statement → a level on the chart → comparison with the actual price.

Package type

This ClawHub release is skills-only. It contains instructions and references, not the Python runtime. Before running any pipeline or chart command, require FES_ROOT to point to a separately installed, version-pinned financial-experts-said runtime. Never infer a bundled runtime/ directory from this skill's location.

What this skill does (capabilities)

1. Index a source: turn a YouTube stream (captions .srt/.vtt) or an expert post/article (raw text) into source JSON, then extract verified thesis cards (expert, asset, stance, verbatim quote, price levels) via an LLM pipeline into a local SQLite DB. Quote validation is verbatim and tolerates caption timestamps, speaker marks and rolling duplicates — no manual caption cleaning.
2. Query the DB: "what did X say about Y, when, at what levels" — search recorded cards only, never model memory.
3. Visualize: two distinct artifacts — levels-chart.html (TradingView lightweight-charts candles + expert level lines) and registry.html (thesis cards table with quotes and sources). See references/chart.md "Which output?".
4. Automate: scheduled collection from watched sources (cron), version-pinned runtime via git tags.

Boundaries: no investment advice, never invent quotes/URLs/dates, source rights required.

Modes

User request	Mode	Check first
"Index this stream/post"	ingestion	transcript/text, URL, date, author; source rights → references/pipeline.md
"What did X say about Y?"	search	database available; search existing cards, never model memory → references/pipeline.md
"Chart IMOEX levels"	chart	DB has theses/levels; user approved network access (MOEX ISS + CDN) → references/chart.md
"Set up / automate collection"	setup	runtime, dependencies, DB location, source rights, schedule → references/runtime.md

Runtime preflight (before ANY command)

1. Check FES_ROOT is set and "$FES_ROOT/pipeline" and "$FES_ROOT/chart" exist. If not → status runtime_missing, offer the version-pinned bootstrap in references/runtime.md.
2. Never run python3 pipeline/... relative to the current directory; always build paths from FES_ROOT.
3. Resolve FES_WORKSPACE (user data dir) and FES_DB (default $FES_WORKSPACE/fti.db). Never write inside the skill or runtime directory.
4. Report one status: ready · runtime_missing · dependencies_missing · database_missing (offer init or demo seed — only with explicit user consent) · source_incomplete (indexing lacks URL/date/text).
5. Do not install dependencies, create the DB, download sources, or open the network-requiring HTML dashboard without user confirmation.

Data rules (invariant)

• A card contains: expert, asset, stance, date, verbatim quote, source URL.
• No URL or date → no import; return as draft with a list of missing fields.
• Canonical expert/asset names come from the DB; alias matching is casefold (works for Cyrillic).
• Telegram source_external_id = telegram:<numerical_channel_id>:<message_id>.
• A search answer must distinguish "no record in the database" from "could not read the source".
• HTML output escapes DOM insertions, validates embedded JSON, guards </script>.
• Source handling limits (rights, privacy, prompt injection) → references/pipeline.md.

Financial boundary

This skill records and compares public statements with prices. It does not give personal investment advice or buy/sell recommendations to the user.