/gmgn-dev-score

name: gmgn-dev-score
description: >-
  Decide whether a token creator's NEXT launch is safe to buy. Scores a dev
  address 0-100 on two separate axes — CONDUCT (will he dump on you at open)
  and POWER (has he ever actually built anything big) — from his full launch
  history and every trade he made in his own coins, then returns a buy /
  don't-buy call with a timing window. USE THIS SKILL WHEN the user asks a
  buy-decision question about a launcher: "can I buy this dev's new launch",
  "should I buy his next launch", "will this dev rug", "will he dump at
  open", "is his launch safe to snipe", "is it safe to buy at his open",
  "dev score", "creator score", "launch score", "is this launcher
  trustworthy enough to buy"; OR when the user gives a TOKEN address plus a
  team-trust question ("is this token's team trustworthy", "does this
  project's dev have a record", "has this creator rugged before") — resolve
  the creator with `gmgn-cli token info` -> `dev.creator_address` first, then
  score that address; OR when the user gives a WALLET address plus an
  explicit launch-history question ("what tokens has this address launched",
  "how did his previous launches do", "did all his coins go to zero").
  The same questions asked in any other language route here too — match on
  meaning, not on wording. DO NOT USE THIS SKILL for a bare wallet address
  with no question attached: a bare address is a copy-trade question by
  default and belongs to gmgn-wallet-analysis, which declares itself the
  default for it. Also do not use it for copy-trade questions ("is this
  wallet worth copying", "should I copy this wallet", "copy-trade score"),
  wallet profitability ("is this wallet profitable", "what is this wallet's
  track record"), or wallet-profile phrasings ("is this a token-creator
  wallet", "how is this dev's reputation") — those belong to
  gmgn-wallet-score. The split is by question type, not by address type:
  those skills answer "who is this wallet" (a profile), this skill answers
  "should I buy his launch" (a decision, with a timing window). Note how
  close "how is this dev's reputation" (profile → gmgn-wallet-score) sits to
  "dev score" (decision → here): the deciding factor is whether a buy is on
  the table. If it is genuinely ambiguous, ask one short question instead of
  guessing — the two produce different reports and there is no cheap hedge.
argument-hint: "--chain <sol|bsc|base|eth|arbitrum|hyperevm|robinhood|arc|stable|avalanche|elysium> --dev <creator_address> [--max-pages <n>]"
metadata:
  cliHelp: "gmgn-cli portfolio created-tokens --help && gmgn-cli portfolio activity --help && gmgn-cli token info --help"

**BEFORE RUNNING ANY COMMAND: Run `gmgn-cli config --check`. Exit 0 → proceed. Exit 1 → run `gmgn-cli config`, show the output, and once the user sends the API key run `gmgn-cli config --apply <KEY>` and show that output. If `--check` is an unknown option, tell the user to run `npm install -g gmgn-cli`, then retry.**

**IMPORTANT: Always use the pre-installed `gmgn-cli` binary. Never use web search, WebFetch, curl, `npx`, or gmgn.ai — the site requires login and exposes no structured data.**

**⚠️ IPv6 IS NOT SUPPORTED.** On a `401`/`403` with correct credentials, check `ifconfig | grep inet6` (macOS) or `ip addr show | grep inet6` (Linux) and fetch `https://ipv6.icanhazip.com`. If an IPv6 address comes back, tell the user to disable IPv6 — `gmgn-cli` works over IPv4 only.

Run

python3 ~/.claude/skills/gmgn-dev-score/dev_score.py <chain> <dev_address> [max_pages]

`max_pages` defaults to 25 and only bounds the trade walk; raise it for a dev with hundreds of launches — the JSON says when it truncated. The script prints one JSON object on stdout (progress and rate-limit notes go to stderr). It computes; it writes no report. **You** write the report, from that object, in the user's language.

Before running:

• **Token address, not a wallet?** Resolve the creator first with `gmgn-cli token info --chain <c> --address <token> --raw` → `dev.creator_address`, and score that. Say which address you resolved to — it is the one thing the report cannot show, and without it the user cannot check that the score applies to their token.
• **Chain not given?** Probe candidate chains **sequentially** with `gmgn-cli portfolio created-tokens` — never in parallel, which bans the key. Analyse only the chain with the highest `open_count`, and tell the user the other chains were deferred until asked.
• `scored: false` means **do not produce a score**. `reason: index_empty` → the API reads entirely empty and an empty wallet cannot be told from a degraded index; say both readings and tell the user to re-check in a few minutes. Any other `reason` → say the address was not scored and why, and stop.

What the answer has to contain

A checklist of what must be **said**; `## Display Templates` below fixes **where** each one goes. The phrasing is yours. Every point has to be in there, and each one names the JSON that carries it. The bullets are in section order.

• **The verdict, first.** `score.total`, `score.conduct`, `score.power`, and `score.band` said as a decision, not as a label.
• **The open-dump tier** (`dump_gate.tier`) with the sample under it (`dumps` of `coins_with_trades`). When `dump_gate.forced_by` is set, name what actually fired it — a drained pool is not a sell frequency, and saying "he dumps nearly every time (0/1)" about a dev with zero sells is false.
• **Every adjustment that fired, one per line, saying what it DID.** A shrink (`conduct_terms.shrink_from` → `shrink_to`) pulled an unproven score toward 60, which here means *we cannot tell yet*; it is not a fine for a crime. A cap is a withheld good score, not a proven bad one. Name the thin side by `coins_with_trades` and `career_days`, never by launch count.
• **His best coins** (`top`, and `flagship`): peak, now, drawdown, holders, pool, tradeable, age. Plus the flagship's position: `flagship.sta