gate-info-risk

gate-info-risk

General Rules

⚠️ STOP — You MUST read and strictly follow the shared runtime rules before proceeding. Do NOT select or call any tool until all rules are read. These rules have the highest priority. → Read gate-runtime-rules.md → Also read info-news-runtime-rules.md for gate-info and gate-news shared rules.

• Only call MCP tools explicitly listed in this skill. Tools not documented here must NOT be called, even if they exist in the MCP server.

CLI and playbook contract

1. Legacy commands MUST exist under gate-cli v0.5.2. When shortcuts_enabled (>= 0.7.6), token_risk may use info +token-risk per playbook shortcut; fall back to check-token-security on failure or older CLI. This skill does NOT call check-address-risk or +address-tracker for address_risk.
2. Call gate-cli info … / gate-cli news … only as listed in playbooks/gate-info-risk.yaml. When preflight route is CLI, do not call Gate MCP tools from this skill.
3. Missing data NEVER counts as "low risk". When the primary command for a chosen playbook is unavailable, the verdict is UNABLE_TO_ASSESS (scope limited; localized labels may appear in the user report) plus an explicit retry hint.
4. Always pass --format json. Required slots (token+chain, address+chain, symbol) must be present; if ambiguous, ask the user.
5. The report is safety-first: the verdict line comes BEFORE evidence, and high-severity items are always surfaced before medium or low.
6. No buy / sell advice. No price predictions. Use verdict bands HIGH / MEDIUM / LOW / UNABLE_TO_ASSESS in this skill’s logic; the final user report may localize those labels. Always include a “what to verify next” list.

---

Step 0 — Preflight

Follow the shared contract in skills/_shared/preflight.md verbatim. Summary:

1. Run gate-cli preflight --format json; parse .route, .status, .user_message.
2. Branch: CLI → continue; MCP_FALLBACK → emit __FALLBACK__ and halt; BLOCK → echo user_message and halt.
3. When status == "ready_with_migration_warning", Step 3 appends one migrate hint at the very end of the report.
4. Step 0.5: Set cli_version and shortcuts_enabled per skills/_shared/cli-version-routing.md.

Do NOT run any data-collection command until Step 0 returns route == "CLI".

---

<!-- Step 0.5 — Skill update check (OPT-IN, disabled by default). The script [scripts/update-skill.sh](https://github.com/gate/gate-skills/blob/master/scripts/update-skill.sh) is shipped and functional, but agents do NOT run it as part of the normal flow (zero overhead, no token needed). Skill authors can still invoke it manually when they want to sync SKILL.md from upstream: bash "$SKILL_ROOT/scripts/update-skill.sh" check gate-info-risk Result semantics, strict-check mode (`GATE_SKILL_CHECK_STRICT=1`) and the auto-apply path (`GATE_SKILL_UPDATE_MODE=auto`) are documented in [skills/_shared/update-workflow.md](https://github.com/gate/gate-skills/blob/master/skills/_shared/update-workflow.md). To re-enable this step inside the agent flow, delete this HTML comment wrapper so the block becomes part of the rendered SKILL.md again. -->

---

Step 1 — Intent routing

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

Playbook id	When to pick	Required slots
token_risk	"is this token safe", "is there a honeypot", "taxes too high", "holder concentration" — subject is a token contract.	token, chain (the token slot accepts ticker OR contract address)
address_risk	"is this address safe", "sanctioned / OFAC", "blacklist", "is this wallet risky".	address, chain
project_risk	"compliance or incident risk for this project", "has there been a security incident on {project}", "exchange announcement risk".	symbol

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

User signal	Route to
"just give me a general analysis of this coin", no safety framing	gate-info-research
"who is this address / what is it doing on-chain" (not risk-first)	gate-info-web3
"why did it drop" / "community view on incident"	gate-news-intel

---

Step 2 — Data collection

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

When shortcuts_enabled and playbook defines shortcut, follow skills/_shared/cli-version-routing.md. Map +token-risk output (risk_level, risk_items, token_identity) to the same verdict rules as check-token-security.

token_risk shortcut args: use playbook args (--symbol) for tickers; args_when with when: contract_address (--address + --chain) for contract-shaped token. After shortcut success, still run optional token_onchain (post_shortcut_optional_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 {symbol_or_token} --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 UNABLE_TO_ASSESS (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. .is_honeypot, .is_open_source) are JSON booleans (true / false). Compare with == true / == false, NOT with "1" / "0".
• Nested list entries inside high_risk_list[], middle_risk_list[], low_risk_list[], risky_list[], attention_list[] have .risk_value as strings ("0" / "1"). Compare with == "1" or cast via tonumber.
• Tax fields (.buy_tax, .sell_tax, nested .tax_analysis.token_tax.{buy_tax,sell_tax,transfer_tax}) are strings (percent without %). Cast with tonumber before numeric comparison.
• Percent fields under .data_analysis (.top10_percent, .top100_percent, .dev_holding_percent, .insider_percent, .max_holder_percent) are strings. Cast before comparing.
• Each nested entry has shape {risk_key, risk_value, risk_level, risk_name, risk_desc}.

Signal (real JSON path)	Effect on verdict
high_risk_list[?risk_key=="is_honeypot"].risk_value == "1" (equivalently top-level is_honeypot == true — top-level is boolean, not string)	HIGH (immediate)
high_risk_list[?risk_key=="is_high_tax"].risk_value == "1" OR tonumber(buy_tax) > 10 OR tonumber(sell_tax) > 10	At least HIGH
middle_risk_list[?risk_key=="owner_change_balance"].risk_value == "1" (owner can rewrite any holder's balance)	HIGH (immediate)
middle_risk_list[?risk_key=="can_take_back_ownership"].risk_value == "1" OR middle_risk_list[?risk_key=="hidden_owner"].risk_value == "1" OR middle_risk_list[?risk_key=="self_destruct"].risk_value == "1"	HIGH (immediate)
middle_risk_list[?risk_key=="is_mintable"].risk_value == "1" OR middle_risk_list[?risk_key=="is_proxy"].risk_value == "1" OR middle_risk_list[?risk_key=="is_blacklisted"].risk_value == "1"	At least MEDIUM
high_risk_list[?risk_key=="is_open_source"].risk_value == "0" (equivalently top-level is_open_source == false — boolean at top level)	At least MEDIUM
tonumber(data_analysis.top10_percent) > 70 (or top-level top10_percent) and no lock-up evidence	At least MEDIUM
get-address-info.risk_labels includes OFAC / sanctioned / mixer / scam / exchange-hack-proceeds / darknet	HIGH (immediate)
get-address-info returns risk_labels=[] AND tags=[]	UNABLE_TO_ASSESS (scope limited) — do not downgrade to LOW
holders.holder_concentration[?label=="No.1-10"].percentage > 0.5 and active large turnover in 7d (get-token-onchain)	Add MEDIUM item
latest-events has event_type == "exploit" / "rug" / "depeg" and create_time ≤ 7d	Add MEDIUM or HIGH (by severity)
All required commands fail	UNABLE_TO_ASSESS + clear retry guidance

Optional summary shortcut: check-token-security.risk_summary.highest_risk_level — an aggregate 0–3 severity score from upstream (3 = honeypot-grade). When highest_risk_level >= 3 without any flagged risk_key, 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 ready_with_migration_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).

Never gate the verdict on migrate status.

---

Cross-skill routing + Safety rules

Routing

Follow the shared matrix in skills/_shared/routing.md. Risk-skill-specific follow-ups:

Follow-up intent	Target
"full research" / "is it worth following"	gate-info-research
"what is this address doing" / "smart money behavior"	gate-info-web3
"why it dropped" / "community take on the event"	gate-news-intel
User changes chain or token and wants a new assessment	Re-enter the same playbook with new slots; do NOT reuse prior verdict.

Safety rules

Follow the shared contract in skills/_shared/report-style.md. Risk-specific hardening:

1. Never say LOW when a required command failed or when no risk labels / security-check fields were returned; use UNABLE_TO_ASSESS (or localized equivalent).
2. Keep the three risk taxonomies (TECHNICAL_CONTRACT / ADDRESS_COMPLIANCE / PROJECT_LEVEL) strictly separate in Section 2.
3. HIGH items MUST appear before any MEDIUM or LOW items in Section 2.
4. No buy / sell advice. The closest allowed phrasing is “avoid / verify further” (neutral, non-prescriptive).
5. Cite the exact JSON path for every verdict-driving claim, e.g. info compliance check-token-security.high_risk_list[?risk_key=="is_honeypot"].risk_value or info compliance check-token-security.data_analysis.top10_percent — never cite a command-only hand-wave like "from check-token-security".

---

References

File	Load when...
skills/gate-info-risk/references/scenarios.md	Picking a playbook id is not obvious, or the user's prompt looks like Scenarios 4 / 5 (address risk, major-coin ambiguity).
skills/gate-info-risk/references/cli-reference.md	You need the full flag table for check-token-security, get-address-info, get-token-onchain, or the auxiliary news commands.
skills/gate-info-risk/references/troubleshooting.md	A command failed, or get-address-info came back with no risk labels (the check-address-risk gap).
skills/_shared/preflight.md	Step 0 contract.
skills/_shared/routing.md	Cross-skill routing.
skills/_shared/report-style.md	Safety + reporting rules.
skills/_shared/update-workflow.md	Skill update workflow.