How do I install the Second Opinion skill?

Run npx skillmds@latest add alitahir6001/second-opinion in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Second Opinion safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Second Opinion?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Second Opinion free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Second Opinion?

alitahir6001 (@alitahir6001) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Second Opinion by alitahir6001 — Agent Skill

Name: Second Opinion
Author: alitahir6001

Second Opinion

Get a second opinion from a specific advisor model the user names, run outside this session via a vendor CLI — Anthropic (claude -p), Google (gemini), or OpenAI (codex, if installed). Use for ad-hoc routing to a provider or model that is not in the configured .mcp.json, or that the consult_advisor MCP tool has no provider for. The user picks the provider, so there is no escalation bar - any question they route here qualifies.

alitahir6001 f759751 2.2 KB Updated Sep 17, 2026

File contents

Second opinion from a named provider

Prefer the MCP tool when it already covers the provider

If consult_advisor is registered and its configured provider is the one being asked for, use it instead. It logs the consult and scrubs the subprocess environment; this skill does neither. Come here when the user names a provider or model the server is not configured for.

Package the consult

Build ONE focused prompt. The advisor is stateless — it sees nothing but what you send. Include:

The question (one decision or problem, not the whole task)
Minimal context: options on the table, constraints, what was already tried
This output contract, verbatim: "Reply with: Recommendation (one clear choice), Reasoning (2-4 load-bearing points), Risks (what could make this wrong). Do not write code."

Run it

Pick the CLI for the provider you want. Run via Bash, passing the packaged prompt as one quoted argument:

Anthropic (bills Claude subscription): claude -p "<packaged prompt>" --model <model>
Google (gemini CLI, free personal tier or its own billing): gemini -p "<packaged prompt>"
OpenAI (only if codex is installed; bills ChatGPT subscription): codex exec "<packaged prompt>"

Unlike the MCP path, this inherits the session's environment. claude -p run this way sees the parent session's CLAUDE* variables and may behave as a nested session; prefer consult_advisor for Anthropic consults.

If the chosen CLI is missing or errors on auth, report that and fall back to the advisor subagent — do not silently skip the consult.

Apply the advice

Weigh the recommendation against repo reality (the advisor can't read this project). State in your reply that you consulted, which provider/model answered, and that this consult is not in the log.

---name: second-opiniondescription: Get a second opinion from a specific advisor model the user names, run outside this session via a vendor CLI — Anthropic (claude -p), Google (gemini), or OpenAI (codex, if installed). Use for ad-hoc routing to a provider or model that is not in the configured .mcp.json, or that the consult_advisor MCP tool has no provider for. The user picks the provider, so there is no escalation bar - any question they route here qualifies.---# Second opinion from a named provider## Prefer the MCP tool when it already covers the providerIf `consult_advisor` is registered and its configured provider is the onebeing asked for, use it instead. It logs the consult and scrubs thesubprocess environment; this skill does neither. Come here when the usernames a provider or model the server is not configured for.## Package the consultBuild ONE focused prompt. The advisor is stateless — it sees nothing butwhat you send. Include:1. The question (one decision or problem, not the whole task)2. Minimal context: options on the table, constraints, what was already tried3. This output contract, verbatim: "Reply with: Recommendation (one clear choice), Reasoning (2-4 load-bearing points), Risks (what could make this wrong). Do not write code."## Run itPick the CLI for the provider you want. Run via Bash, passing the packagedprompt as one quoted argument:- Anthropic (bills Claude subscription): `claude -p "<packaged prompt>" --model <model>`- Google (gemini CLI, free personal tier or its own billing): `gemini -p "<packaged prompt>"`- OpenAI (only if `codex` is installed; bills ChatGPT subscription): `codex exec "<packaged prompt>"`Unlike the MCP path, this inherits the session's environment. `claude -p`run this way sees the parent session's `CLAUDE*` variables and may behaveas a nested session; prefer `consult_advisor` for Anthropic consults.If the chosen CLI is missing or errors on auth, report that and fall backto the advisor subagent — do not silently skip the consult.## Apply the adviceWeigh the recommendation against repo reality (the advisor can't read thisproject). State in your reply that you consulted, which provider/modelanswered, and that this consult is not in the log.

Frequently asked questions