メニュー
Get a second opinion from Google Gemini via its headless CLI. Use when the user says "/consult-gemini", "ask gemini", "gemini second opinion", or wants an outside model's take on a question, plan, or code.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Semantic memory layer for long-term knowledge storage and retrieval across sessions. Use for /agentmemory remember, search, context, forget, status commands. Distinct from holographic memory (live cross-agent state) — agentmemory is a searchable knowledge base backed by BM25+vector hybrid search at localhost:3111.
Alias for /consult-ring261T. Consult Ring 2.6 1T (inclusionai/ring-2.6-1t) on OpenRouter. Use when user says "/ask-ring261T" or "ask ring".
Use when tracing how a pick on findtorontoevents.ca/audit got there — what source emitted it, which gates it passed or was rejected by, why it is or is not in Smart Picks / High Conviction / Money Ready, or why a class shows the win-rate it does. Also for "which DB tables matter", per-asset-class pick-funnel case studies, and onboarding someone to the audit pipeline.
Use when user asks to check GitHub Actions for failing jobs, chronic cancellations, or jobs without a subsequent successful run. Also for CI/CD health audits, rerunning failed workflows, or monitoring workflow reliability. Aliases - checkGHA, gha_health, github_actions_health, gh_actions, check_actions, fixjob, github_actions, actions_health
Get a second opinion from OpenAI Codex via its headless CLI. Use when the user says "/consult-codex", "ask codex", "codex second opinion", or wants an outside model's take on a question, plan, or code.
Get a second opinion from Cursor Agent via its headless CLI (the `agent` command). Use when the user says "/consult-cursor-agent", "ask cursor-agent", "cursor-agent second opinion", or wants an outside model's take on a question, plan, or code.
| name | consult-gemini |
| description | Get a second opinion from Google Gemini via its headless CLI. Use when the user says "/consult-gemini", "ask gemini", "gemini second opinion", or wants an outside model's take on a question, plan, or code. |
Query Google Gemini for a second opinion, non-interactively.
User asks "ask gemini" / "/consult-gemini" / second opinion from Gemini.
powershell -NoProfile -Command 'Set-Location $env:TEMP; gemini -p "PROMPT" --output-format json -y --skip-trust'
Output is one JSON object; the answer is the .response field (Gemini puts the
model reply there; fall back to scanning the object if absent).
| Flag | Purpose |
|---|---|
Set-Location $env:TEMP (via PowerShell) | Runs the CLI from a clean temp directory outside any repository — prevents Gemini discovering repo-local .gemini/ settings, GEMINI.md, rules, or MCP servers that pollute the second opinion or break headless mode. |
-p "PROMPT" | Non-interactive prompt mode. Without it the CLI starts an interactive REPL instead of exiting after one reply. |
--output-format json | Returns one machine-readable JSON object instead of TTY/streaming text, enabling reliable parsing. |
-y (or --approval-mode yolo) | Auto-approves every tool call. Without an auto-approve flag the headless process blocks forever waiting for terminal confirmation. |
--skip-trust | Required. Gemini refuses to run headless in an "untrusted directory" (even $env:TEMP) — it errors Gemini CLI is not running in a trusted directory. --skip-trust (or env GEMINI_CLI_TRUST_WORKSPACE=true) bypasses the trusted-folder gate. |
powershell -NoProfile -Command 'Set-Location $env:TEMP; gemini -p "reply with exactly: PONG" --output-format json -y --skip-trust'
--skip-trust Gemini aborts before answering. Verified failure on this machine 2026-05-17.Set-Location $env:TEMP is mandatory for an uncontaminated opinion.gemini interactively once to complete login / API-key setup; subsequent headless calls succeed.