external-ai
External AI Skill
Invoke external AI CLI tools using natural language.
When to Use
Activate when the user:
- Says
/call,/codex, or/gemini - Asks for a second opinion from another AI
- Wants external review of code or design
- Needs verification from a different model
Instructions
Natural Language Interface
Users provide natural language. Translate it into CLI commands.
Examples:
- "review my changes" ->
codex review --uncommitted - "review against main with high reasoning" ->
codex review --base main -c model_reasoning_effort="high" - "check this file for bugs" ->
codex exec "Review for bugs: $(cat file.ts)" - "give me a second opinion" ->
gemini "Second opinion on..."
Self-Healing Behavior
On CLI Syntax Errors
- Run
<cli> --helpto get the current syntax - Update the reference docs at
reference/cli-reference.md - Retry with the correct syntax
- Inform the user of the fix
On Upgrade Available
- Run the upgrade command
- Update the version in
reference/cli-reference.md - Re-run
--helpto capture new options - Inform the user of the upgrade
On Gemini 403 Project/License Errors
If Gemini fails with 403 and messages like lack a Gemini Code Assist license or (#3501):
- Check project env vars:
env | rg "GOOGLE|GCP|CLOUDSDK|PROJECT" - Retry with project vars unset:
env -u GOOGLE_CLOUD_PROJECT_ID -u GOOGLE_CLOUD_PROJECT -u CLOUDSDK_CORE_PROJECT gemini ...
- If still blocked, report auth/licensing clearly and continue with other available models
- Record the fix path in
reference/cli-reference.md
On Gemini ModelNotFound / Capacity Errors
If Gemini fails with 404 or 429 style model/capacity issues, retry with this fallback order:
gemini-3.1-pro-previewgemini-3-pro-previewgemini-3-flash-previewgemini-2.5-progemini-2.5-flash
Command Interface
/call [codex|gemini] <natural language request>
/codex <natural language request>
/gemini <natural language request>
Translation Guidance
Codex
- Prefer
codex reviewfor git-aware reviews - Use
codex execfor custom prompts - Default to high reasoning for reviews when helpful
- Use
--full-autoonly for simple, low-risk tasks
Gemini
- Prefer preview model IDs first
- For 403 auth issues, retry with project env vars unset
- For 404/429 failures, continue through the fallback sequence automatically
Context Gathering
When a request needs context, gather it before invoking the external CLI:
- "review my changes" -> inspect uncommitted changes
- "review this file" -> read the file
- "review the dev skill" -> gather the skill files
- "check my implementation" -> identify the relevant files
Process Flow
- Parse the natural language request
- Determine the target CLI (
codexorgemini) - Translate to the appropriate CLI command
- Execute the command
- If syntax error -> self-heal and retry
- If upgrade available -> upgrade and refresh references
- Return the results to the user
More from codewithjv/agent-skills
retro
Run a retrospective on the current task or chat history. Use when the user wants to understand mistakes, dead ends, wasted effort, root causes, and what should change to improve performance next time.
33call-external-ai
Route natural-language prompts into external AI CLIs like Codex, Claude, and Gemini using safe command templates and minimal context handling.
19skill-extract-scripts
Review a skill and extract deterministic, mechanical steps into shell scripts. Makes skills more reliable by separating precision work (scripts) from judgment work (AI). Use when asked to extract scripts from a skill, make a skill more deterministic, or split a skill into script + prompt.
19review-context-hub
Review a repo as a context hub. Use when the user wants an overview of the repo structure, recent changes, standing instruction files, installed skills, remotes, or who has access.
19create-locked-down-skill
Create a restricted workflow directory with locked-down permissions for Claude Code, Codex CLI, Gemini CLI, and OpenCode. Use when asked to set up a new workflow, create a sandboxed workspace, restrict permissions for a task, or scaffold a locked-down skill.
10