gold-digger

name

gold-digger

description

Research early crypto-AI projects and generate daily compounding reports. Use whenever the user asks to research a crypto project, find low-cap AI-agent tokens, track KOLs like DegenSensei or resdegen, check whether a project has announced a token or points program, audit a watchlist, compute mention velocity for a ticker, compare a project against similar narratives, generate a daily briefing of crypto-AI signals, or discover new projects that look like ai16z, openserv, virtuals, or other AI-agent plays. Also handles adding projects and KOLs to the watchlist, running scout discovery, and producing structured daily markdown reports.

Gold Digger

Research engine for early crypto-AI projects. Watchlist enrichment + scout discovery + KOL signal digest, emitting daily markdown reports that compound over time.

When to use this skill

Trigger on any of:

• "research project X" where X is a crypto or AI-crypto project
• "find new AI agent / DePIN / intent / restaking projects"
• "what's DegenSensei saying about X"
• "what did resdegen mention this week"
• "how much has [ticker] moved"
• "does [project] have a token / points program / airdrop"
• "run the daily gold-digger report"
• "add [project] to my watchlist"
• "what's heating up"

Architecture (two-sentence version)

Gold Digger is a thin crypto-research layer over last30days + direct calls to CoinGecko / DeFiLlama / GitHub / Perplexity. It maintains a markdown-native watchlist (one .md per project with frontmatter), runs enrichment + scout + aggregation daily, and writes structured reports that link back to the project pages.

How to run

Always invoke via the Python CLI — never reimplement the research logic in conversation.

# Full daily run (enrich all tracked projects + scout + report)
python3 ~/projects/gold-digger/scripts/gold_digger.py daily

# Enrich a single project (handy for testing / on-demand research)
python3 ~/projects/gold-digger/scripts/gold_digger.py enrich <slug>

# Full cited coin/project DD (team, token/TGE, value capture, liquidity,
# code activity, catalysts, risks, and buy/no-buy read)
python3 ~/projects/gold-digger/scripts/gold_digger.py research <slug>

# Scout pass only
python3 ~/projects/gold-digger/scripts/gold_digger.py scout

# Add a project by slug + coingecko id
python3 ~/projects/gold-digger/scripts/gold_digger.py add-project <slug> --coingecko-id <id>

# First-run key check
python3 ~/projects/gold-digger/scripts/gold_digger.py setup

The CLI emits to stdout (compact summary) and writes full output to the data directory. When the user just wants a quick answer, parse the compact summary and respond. When they want the full report, show them the path and wiki-link it.

Key discovery

Do not ask the user for API keys in conversation. Keys live in the environment or in one of the supported config files (see README). The gold_digger setup subcommand shows which keys are present and which sources are therefore available. If the user reports "source X isn't working," run setup first and check the relevant row.

Degradation policy

Every source is optional. Gold Digger prefers to run in degraded mode rather than fail:

• No COINGECKO_API_KEY — skip price/mcap enrichment; still write identity + social frontmatter
• No XAI_API_KEY — skip KOL digest and first-mention scout; still run watchlist enrichment and market-data scout
• No PERPLEXITY_API_KEY — deep-dive subagent falls back to raw Brave/Exa results
• No last30days installed — skip social aggregation; still run direct-API enrichment
• All sources missing — still runs; emits a report explaining which keys would unlock which features and stops

Markdown conventions

Projects live at $GOLD_DIGGER_DATA/projects/<slug>.md with frontmatter holding structured fields and body holding notes. Never overwrite user notes in the body — only update frontmatter. When updating, preserve any frontmatter fields the user added manually that aren't in the schema.

Reports wiki-link to project pages ([[openserv]]) so Obsidian backlinks work.

Snapshots are daily tables in snapshots/YYYY-MM-DD.md — one row per project. The aggregator reads N days of snapshots to compute velocity and divergence.

Referenced files

• references/schema.md — full project frontmatter field spec (load when the user asks about tracked fields or you need to write a new project file)
• references/narratives.md — user-editable narrative taxonomy (load when tagging projects or running scout with a narrative filter)
• references/api-keys.md — detailed matrix of which keys unlock which sources (load when the user asks "what should I get" or setup reports a missing key)
• references/extending.md — how to add sources, extractors, narrative tags, or custom scoring (load when the user wants to extend Gold Digger)
• agents/gold-digger-researcher.md — deep-dive subagent definition (invoke for full project due-diligence beyond daily enrichment)

Output format

Daily report sections (in order):

1. New discoveries (scout finds from the last 24h)
2. Watchlist deltas (price moves, social jumps, announcements)
3. KOL digest (what each tracked KOL said in the last 24h, with extracted tickers)
4. Trending narratives (which categories are heating up)
5. Heating up (projects with rising mention velocity)
6. Action queue (projects worth deep-dive tomorrow)

Brief report: 5 bullets — best find, biggest mover, hottest KOL signal, action item, narrative of the day.

What NOT to do

• Do not ask the user for API keys in chat — read them from the environment
• Do not overwrite user-authored notes in project file bodies
• Do not commit keys or frontmatter data to the skill repo
• Do not fabricate prices, mcap, or funding data — if a source fails, mark the field null in the frontmatter
• Do not reimplement social research — call last30days via subprocess
• Do not treat the daily run as idempotent — each run appends a new snapshot and may add new scout-tier projects