Call AI

Run prompts against external AI providers. This is a low-level building block skill.

Alias: /ca

Quick Usage

• /call-ai "prompt" — Codex + Gemini thorough (default)
• /call-ai :all "prompt" — all configured models
• /call-ai codex "prompt" — Codex only
• /call-ai gemini "prompt" — Gemini only
• /call-ai claude "prompt" — Claude only

Progressive Loading Contract (Skill Graph)

For each invocation, load in order:

1. graph/index.md
2. one MOC: graph/mocs/execution.md or graph/mocs/reliability.md
3. only required nodes under graph/nodes/
4. deep references/examples only when needed

Hard Requirements

• External AIs have zero conversation context; prompt must be self-contained.
• Use ai-registry.yaml as model source of truth.
• Use script-defined reliability behavior (retry/backoff/timeout); do not re-implement ad-hoc loops.
• Never fail silently; always return explicit success/failure outcomes.

Execution Router

Reliability Router

Use when debugging degraded runs:

• graph/nodes/retry-and-timeout.md
• graph/nodes/handle-failures.md

Script Surface

• ask-ai.sh
• ask-ai-runner.sh
• ask-ai-zellij.sh
• ask-ai-tmux.sh
• ask-ai-ghostty.sh
• scripts/run-parallel.sh
• scripts/validate-response.sh

References

• references/ai-registry.md
• examples/single-ai.md
• examples/multi-ai.md
• examples/error-recovery.md
• examples/all-models.md