Step 1 — Intent routing

Pick exactly one playbook id from playbooks/gate-info-risk.yaml:

Slot extraction rules

• check-token-security takes exactly one of --token or --address — the CLI explicitly rejects passing both (see the CLI error message; mutually exclusive). This skill always uses --token because the flag is polymorphic: it accepts BOTH tickers (USDT) and contract addresses (0xdAC17F..., SPL mints, TRX mints). Agent slot-extraction rule: when the user mentions a contract address for a token-risk query, store that address in the token slot directly — do NOT create a parallel address slot for this playbook. Always pair with --chain.
• chain canonical form (always send the short id on --chain): eth, bsc, polygon, arb, op, base, avax, solana, tron, ... Accept common long-form aliases from users (ethereum, arbitrum, optimism, avalanche, bnb, matic) and normalize to the short id before calling the CLI. Response .chain echo is endpoint-dependent — sometimes short (eth→eth), sometimes long (arb→arbitrum). Parsers MUST accept both alias forms; see skills/gate-info-risk/references/troubleshooting.md (canonical alias table for input normalization and response parsing).
• address is a wallet. If the user asks about a token contract but passed an address, confirm which they mean.
• symbol for project_risk is the listed coin ticker; resolve ambiguous project names before calling news commands.
• Missing or ambiguous required slot → STOP, ask the user. Do not guess.

Cross-skill routing at Step 1

Step 2 — Data collection

All commands use --format json. Commands inside the same parallel_group run concurrently.

When shortcutsenabled and playbook defines shortcut, follow [skills/shared/cli-version-routing.md](../shared/cli-version-routing.md). Map +token-risk output (risklevel, riskitems, tokenidentity) to the same verdict rules as check-token-security.

tokenrisk shortcut args: use playbook args (--symbol) for tickers; argswhen with when: contractaddress (--address + --chain) for contract-shaped token. After shortcut success, still run optional tokenonchain (postshortcutoptional_ids) — shortcut does not include on-chain holder data.

2.A token_risk

Shortcut (>= 0.7.6): resolve args from playbook shortcut.args / shortcut.args_when (never default to --address for tickers). On native-coin resolution failure, fall back to legacy commands below.

Required (legacy):

• gate-cli info compliance check-token-security --token {token} --chain {chain} --scope full

Optional (run in parallel with the above):

• gate-cli info coin get-coin-info --query {symbolortoken} --chain {chain} --scope basic — project fundamentals for context.
• gate-cli info onchain get-token-onchain --token {token} --chain {chain} --scope full — holder concentration, smart-money flow.

2.B address_risk

Required:

• gate-cli info onchain get-address-info --address {address} --chain {chain} --scope basic — risk labels, tags, scoped balance summary.

Optional:

• gate-cli info onchain get-address-transactions --address {address} --chain {chain} --time-range 30d --min-value-usd 100000 --limit 50 — recent large flows; helps flag mixing / exchange hopping.

Scope-limited rule: gate-cli v0.5.2 does not expose info compliance check-address-risk. If get-address-info returns no risk labels or tags, the final verdict MUST be UNABLETOASSESS (localized in user output) with a note to re-check once the dedicated address-risk command ships.

2.C project_risk

Required (run in parallel):

• gate-cli news feed search-news --coin {symbol} --limit 10 --sort-by importance --time-range 7d
• gate-cli news events get-latest-events --coin {symbol} --time-range 7d --limit 20 (server enum is 1h|24h|7d — use --start-time / --end-time unix-ms for longer windows)

Optional:

• gate-cli news feed get-exchange-announcements --coin {symbol} --limit 10 — listing / delisting / maintenance / emergency notices.

Failure policy

• Required command failure → abort playbook, surface CLI error verbatim (trim auth), stop. No verdict is emitted.
• Optional command failure → mark the corresponding section scope limited in Step 3 and downgrade confidence.
• Upstream returns "not found" or empty labels → treat as data gap, NOT as low risk.

Step 3 — Synthesis

The report is always 5 sections, in this order:

Report template

## Risk assessment — {subject}

### 1. Verdict
- **Level**: {HIGH | MEDIUM | LOW | UNABLE_TO_ASSESS} (you may localize labels in the user-facing report)
- Three short sentences: main basis + data coverage + confidence (whether scope-limited)

### 2. Core risk items
Sort strictly HIGH → MEDIUM → LOW. Each item includes:
- Risk name + taxonomy: `TECHNICAL_CONTRACT` / `ADDRESS_COMPLIANCE` / `PROJECT_LEVEL`
- Evidence (real JSON paths), e.g. `check-token-security.high_risk_list[?risk_key=="is_honeypot"].risk_value == "1"`, `check-token-security.data_analysis.top10_percent`, `get-address-info.risk_labels`
- Impact + trigger (e.g. “tax rises after holding > 7 days”)

### 3. On-chain or compliance context
{Balances / tags / labels / large flows / holder concentration / fundamentals. Sources as hit by playbook:
`info onchain get-address-info`, `info onchain get-address-transactions`, `info onchain get-token-onchain`, `info coin get-coin-info`}

### 4. Event context (if any)
{Last 7–30d news / incidents / announcements from `news feed search-news`, `news events get-latest-events`, `news feed get-exchange-announcements`. Separate **sourced facts** from **community speculation** (mark “needs confirmation” where needed).}

If the playbook did not call news commands, mark **no data**; you may suggest `gate-news-intel` for follow-up.

### 5. What to verify next
- List concrete next commands to close each gap
- If **UNABLE_TO_ASSESS**, the first bullet must tell the user to re-run `check-token-security` / `get-address-info` with the correct `{chain}`
- No buy/sell advice; no price targets

Verdict rules (applied in section 1)

check-token-security intentionally surfaces two type conventions — agents must not mix them:

• Top-level flags (e.g. .ishoneypot, .isopen_source) are JSON booleans (true / false). Compare with == true / == false, NOT with "1" / "0".
• Nested list entries inside highrisklist[], middlerisklist[], lowrisklist[], riskylist[], attentionlist[] have .risk_value as strings ("0" / "1"). Compare with == "1" or cast via tonumber.
• Tax fields (.buytax, .selltax, nested .taxanalysis.tokentax.{buytax,selltax,transfer_tax}) are strings (percent without %). Cast with tonumber before numeric comparison.
• Percent fields under .dataanalysis (.top10percent, .top100percent, .devholdingpercent, .insiderpercent, .maxholderpercent) are strings. Cast before comparing.
• Each nested entry has shape {riskkey, riskvalue, risklevel, riskname, risk_desc}.

Optional summary shortcut: check-token-security.risksummary.highestrisklevel — an aggregate 0–3 severity score from upstream (3 = honeypot-grade). When highestrisklevel >= 3 without any flagged riskkey, still run through the rules above to cite the specific offending field rather than citing risk_summary alone.

Migrate hint (only when Step 0 status was readywithmigration_warning)

At the end of the report, append verbatim the migration blockquote line from skills/shared/report-style.md (line starting with ⚙️ and gate-cli migrate --dry-run).