Model Recommendations by Phase

Expires: 2026-03-06 Last researched: 2026-02-20

When this file is expired, the agent MUST do a fresh web search for current model rankings (speed, reasoning, cost) and update the recommendations and expiration date (+2 weeks).

Detection

Check environment variables to determine the agent:

• CLAUDECODE=1 → Claude Code
• GEMINI_CLI=1 → Gemini CLI
• CODEX_CLI=1 → Codex
• Otherwise → unknown (show generic suggestion)

Phase-to-Model Mapping

Phases are grouped by reasoning demand:

Light (fast/cheap model preferred):

• /iikit-core (init, status, help)
• /iikit-04-checklist (scoring against spec)
• /iikit-09-taskstoissues (mechanical issue creation)

Medium (balanced model):

• /iikit-01-specify (structured extraction from description)
• /iikit-02-clarify (targeted questions)
• /iikit-05-testify (test spec generation)
• /iikit-06-tasks (task breakdown)
• /iikit-bugfix (bug analysis + task generation)

Heavy (strongest model preferred):

• /iikit-00-constitution (governance design, premise authoring)
• /iikit-03-plan (architecture, tech selection, data model)
• /iikit-07-analyze (cross-artifact consistency)
• /iikit-08-implement (code generation, parallel orchestration)

Agent-Specific Recommendations

Claude Code

Gemini CLI

Codex

Unknown Agent

Show generic suggestion: "Consider using a stronger/faster model for this phase."

Suggestion Format

When suggesting the next step, append a model hint:

Next: /iikit-03-plan
Tip: This phase benefits from deep reasoning. Switch to [model]: [command]

Only suggest if the current model tier doesn't match the recommended tier for the next phase. If already on the right tier or higher, skip the suggestion.