gate-news-briefing

gate-news-briefing

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 / gate-news-specific rules (tool degradation, report standards, security, routing degradation, per-skill version checks when scripts/ is present, and legacy wrapper routing).

• 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.
• Legacy / routing mode: when Step 0 emits __FALLBACK__, use only the MCP tools listed in this file. When Step 0 emits __ROUTE_CLI__, do not call those MCP tools; delegate to the mapped primary skill per Step 0.

The crypto industry "morning briefing" Skill. The user asks what's been happening; the system calls 3 MCP Tools in parallel to fetch major events + trending news + social sentiment, then the LLM aggregates into a layered news briefing.

Trigger Scenarios: User asks about recent/today's news, headlines, or what's been happening.

Per-skill updates: This directory includes scripts/update-skill.sh and may include scripts/update-skill.ps1. ClawHub packages uploaded to the marketplace often omit update-skill.ps1 (upload-page restriction); GitHub / Bitbucket source trees keep both. Policy: info-news-runtime-rules.md §1. The Trigger update steps below apply in addition to gate-runtime-rules.md; when scripts are present, use this flow for version checks before execution.

Update check — user visibility: Technical failures during version check (missing script, sandbox, network, non-zero exit, no parseable Result= line) must not be explained to the user; continue with Execution per info-news-runtime-rules.md. Only successful check / apply outcomes may be summarized (including update_available / strict exit 3, which is still a success path that requires user confirmation before apply). Do not auto-download update-skill.* from the network. Static reference if .ps1 is missing: canonical scripts live in gate/gate-skills under skills/<name>/scripts/ (same <name> as frontmatter).

Step 0 — Wrapper routing probe

This legacy skill is a compatibility alias for the primary CLI skill gate-news-intel.

Before Trigger update, MCP tool selection, or any legacy Execution Workflow, run a deterministic shell probe:

PRIMARY_SKILL="gate-news-intel"
HAS_PRIMARY=0
for root in \
  "$HOME

Interpretation:

• __ROUTE_CLI__ → stop here. Do not run Trigger update or the legacy MCP sections below. Follow https://github.com/gate/gate-skills/blob/master/skills/gate-news-intel/SKILL.md and its playbook contract. Carry over the minimum context: optional symbol and time_range; if no ticker is present, let the primary skill choose market_wide_intel.
• __FALLBACK__ → continue with Trigger update and the legacy MCP workflow in this file.

Trigger update (with Execution)

Default (PD / blocking): On each trigger, run check first without GATE_SKILL_UPDATE_MODE=auto. The script only compares local vs remote; if Result=update_available, it does not overwrite files. STOP — ask the user whether to update before any apply, run, or skill Execution. In sandboxes (e.g. Cursor), use full / all permissions when later running apply. This matches blocking update confirmation in the independent-script TEST_REPORT: blocking is enforced by these SKILL instructions plus check-only behavior (not by silent auto-apply).

Strict check (Agent / IDE — recommended on check): Use GATE_SKILL_CHECK_STRICT=1. When Result=update_available, check exits 3 and prints GATE_SKILL_CHECK_EXIT=3 — it stops only that shell step. Hard rule: after you ask the user, end this assistant turn; do not call apply, run, or MCP in the same message. run applies without a prior user step — do not use run in chat for the blocking flow; use check → user confirms → apply.

Step 1 — Check

Install root: check / apply read $DEST/SKILL.md. Typical $SKILL_ROOT: $HOME/.cursor/skills/<name>/, $HOME/.codex/skills/<name>/, $HOME/.openclaw/skills/<name>/, $HOME/.agents/skills/<name>/, $HOME/.gemini/antigravity/skills/<name>/. Single-arg: the script resolves DEST in that order when SKILL.md exists; otherwise DEST is scripts/../. Workspace-only trees need two-arg check / apply with explicit DEST.

Agent selection rules (run at most one of the fenced examples below; choose by OS and files under $SKILL_ROOT/scripts/):

• macOS / Linux / WSL / Git Bash: If update-skill.sh exists → use only the Bash example.
• Windows (native PowerShell): If update-skill.ps1 exists → use only the PowerShell example.
• Windows without .ps1 (typical ClawHub install): If only update-skill.sh exists → try WSL or Git Bash (bash) using the same command shape as the Bash example. If bash is unavailable or execution fails → skip Step 1 silently (no script stderr or “version check failed” message to the user); proceed to Execution.
• Neither script exists: Skip Step 1 silently; proceed to Execution. Do not invent GATE_SKILL_CONFIRM_TOKEN or run apply / revoke-pending.

Bash (blocking — no auto; example Cursor):

PowerShell:

Result semantics: skipped = no action. update_available = remote newer; with blocking flow, do not apply until the user agrees. check_failed = could not compare — proceed with current version per info-news-runtime-rules.md; do not surface technical check failure details to the user.

Agent parse (stdout): GATE_SKILL_UPDATE_AGENT_ACTION=…. BLOCK_UNTIL_USER_CONFIRMS_UPDATE → Step 2 before Execution. CONTINUE_SKILL_EXECUTION → no block from the check script.

Step 2 — Confirm or Reject (blocking)

Runtime: Use the same shell family for Step 2 as for Step 1 (Bash vs PowerShell). If Step 1 was skipped, do not run apply or revoke-pending.

If update_available:

1. STOP — do NOT proceed to Execution yet.

2. Inform the user (e.g. newer version available; summarize if helpful).

3. Wait for the user’s reply — blocking step.

   Hard rule (Cursor / Agent): When check reports update_available, or BLOCK_UNTIL_USER_CONFIRMS_UPDATE, or strict exit 3, end this turn after asking. Only in the user’s next message run apply (if they agree) or revoke-pending (if they decline). Do not chain apply in the same turn as check for this flow.

   • User agrees → run apply with GATE_SKILL_CONFIRM_TOKEN from strict check stdout when required, then Execution.
   • User declines → revoke-pending, then Execution on the current install.

Two-step gate (strict check): apply / run (without GATE_SKILL_UPDATE_MODE=auto) fail until GATE_SKILL_CONFIRM_TOKEN matches .gate-skill-apply-token. User decline → revoke-pending.

If Step 1 was not strict (no pending token): apply without GATE_SKILL_CONFIRM_TOKEN is allowed.

If skipped or check_failed: no update step; proceed to Execution.

Optional — GATE_SKILL_UPDATE_MODE=auto

For CI / unattended automation only: setting GATE_SKILL_UPDATE_MODE=auto on check makes the script apply immediately when the remote is newer — no user confirmation and incompatible with blocking update confirmation tests. Do not use auto on check when reproducing the blocking PD flow.

Parameters

• name: Frontmatter name above; must match skills/<name>/ on gate-skills.
• Invoke: Use $SKILL_ROOT/scripts/update-skill.sh (or .ps1) where $SKILL_ROOT/SKILL.md is this skill — e.g. ~/.cursor/skills/<name>, ~/.codex/skills/<name>, ~/.openclaw/skills/<name>, ~/.agents/skills/<name>, ~/.gemini/antigravity/skills/<name>; do not treat ~/.cursor (or any host root without skills/<name>/SKILL.md) as the install. With one arg, the script resolves $SKILL_ROOT in that order before falling back to the script’s directory; workspace installs need explicit DEST. Two-arg check / apply / revoke-pending: canonical order is absolute DEST (skill root) first, then name; update-skill.sh / update-skill.ps1 auto-swap when only one normalized path contains SKILL.md (e.g. agent passes name then path).
• ClawHub vs full tree: Installs without update-skill.ps1 may copy it from gate/gate-skills under skills/<name>/scripts/ (manual only; agents must not auto-download).

Do not dump raw script logs into the user-facing reply except when debugging. On check exit 3 (strict), do not run Execution until Step 2 is resolved. On check_failed or apply failure, still run Execution when appropriate per runtime rules.

MCP Dependencies

Legacy path only — this section applies when Step 0 emitted __FALLBACK__.

Required MCP Servers

MCP Tools Used

Query Operations (Read-only)

• news_events_get_latest_events
• news_feed_get_social_sentiment
• news_feed_search_news

Authentication

• API Key Required: No
• Credentials Source: None; this skill uses read-only Gate Info / Gate News MCP access only.

Installation Check

• Required: Gate-News
• Install: Use the local Gate MCP installation flow for the current host IDE before continuing.
• Continue only after the required Gate MCP server is available in the current environment.

Routing Rules

Legacy path only — when Step 0 emitted __ROUTE_CLI__, routing is delegated to gate-news-intel.

Execution Workflow

Legacy path only — this section applies when Step 0 emitted __FALLBACK__.

Step 0: Multi-Dimension Intent Check

Before executing this Skill, check if the user's query involves multiple analysis dimensions:

• If the query is exclusively about recent news or headlines, proceed with this Skill.
• If the query also mentions coin analysis, risk check, technicals, fundamentals, or any other analysis dimension beyond news, route to gate-info-research — it handles multi-dimension queries with unified tool deduplication and coherent report aggregation.

Step 1: Intent Recognition & Parameter Extraction

Extract from user input:

• coin (optional): Whether a specific coin is targeted, e.g., "SOL news" → coin=SOL
• time_range: Time window (default: 24h; if user says "this week" → 7d)
• topic (optional): Whether focused on a specific topic (e.g., "regulation", "ETF", "DeFi")

Step 2: Call 3 MCP Tools in Parallel

General Briefing Mode (no specific coin):

Coin-Specific Mode (user specified a coin):

Step 3: LLM Aggregation

The LLM must:

1. De-duplicate events and news (the same event may appear in both feeds)
2. Rank by impact/importance
3. Categorize (Regulation, Projects, Market, Technology, etc.)
4. Combine with social sentiment to identify market focus areas

Step 4: Output Structured Briefing

Report Template

Legacy path only — this section applies when Step 0 emitted __FALLBACK__.

Coin-Specific Mode Template

De-duplication & Ranking Logic

Error Handling

Cross-Skill Routing

Safety Rules

1. Source attribution: Every news item must include the source (media name) and time
2. No interpretive speculation: News summaries must remain objective — no subjective extrapolation
3. No fabricated news: If data is insufficient, reduce the number of items rather than fabricate content
4. Neutral handling of sensitive topics: Maintain neutral tone when covering regulation, politics, or other sensitive subjects
5. Timeliness labeling: Clearly label the time range covered by the data