model-probe

model-probe (builtin skill)

Role

Discover available LLM models in the current runtime, classify each into the three tiers (strong / balanced / cheap), and write the resulting catalog so config-resolve can expand tier_* symbols. Implements architecture §3.2.4.2 (Model Discovery & Tiering).

Triggers

• init.sh --refresh-models (manual)
• init.sh --install-plugin / --upgrade-core (auto on install events)
• Workspace catalog older than ttl_days (auto-refresh in M2+)
• Main session "刷新模型" natural-language command (M2+)

Probe sources (in order)

1. Runtime API — Claude Code list_models() / Copilot CLI registry. M1: not implemented; falls through to source 2.
2. CODENOOK_AVAILABLE_MODELS env var — comma-separated model ids.
3. Builtin fallback — minimum viable triple (opus-4.7,sonnet-4.5,haiku-4.5).

The runtime label is set to "env", "builtin-fallback", or one of the runtime names as appropriate.

CLI

probe.sh                                       print catalog JSON to stdout
probe.sh --output <file>                       write catalog JSON to <file>
probe.sh --tier-priority <yaml_file>           override built-in priority
probe.sh --check-ttl <file> --ttl-days <int>   exit 0 if fresh, 1 if stale

Any catastrophic probe error (e.g. unreadable --tier-priority file) → stderr starts with probe failed: and exits non-zero.

Tier classification

Each model id is matched against tier_priority (user-supplied or built-in). The model's tier is the first priority bucket containing it. resolved_tiers.<tier> is the first id from tier_priority[tier] that is also in available.

Built-in tier_priority mirrors implementation.md §3.5.1.2:

strong:   [opus-4.7, opus-4.6, sonnet-4.6, gpt-5.4]
balanced: [sonnet-4.6, sonnet-4.5, gpt-5.4, gpt-5.4-mini]
cheap:    [haiku-4.5, gpt-5.4-mini, gpt-4.1, sonnet-4.5]

Output schema

{
  "refreshed_at": "ISO-8601 UTC",
  "ttl_days": 30,
  "runtime": "env | builtin-fallback | claude-code | copilot-cli",
  "available": [
    {"id": "opus-4.7", "tier": "strong", "cost": "high", "provider": "anthropic"}
  ],
  "resolved_tiers": { "strong": "opus-4.7", "balanced": "sonnet-4.6", "cheap": "haiku-4.5" },
  "tier_priority": { "strong": [...], "balanced": [...], "cheap": [...] }
}