Provider Health Check Summary (Starter Pack)
Produce a one-screen health summary of every provider Octopus can seat, so the user knows what a workflow will actually dispatch to before spending money.
When to use
The user asks "what providers do I have?", a workflow banner showed (unavailable - skipping), or a session is starting on a new machine.
Steps
- Detect. Run
${CLAUDE_PLUGIN_ROOT}/scripts/helpers/check-providers.sh and ${CLAUDE_PLUGIN_ROOT}/scripts/helpers/check-versions.sh.
- Summarize per provider. For each seat report: available or not, auth method (env key, oauth session, subscription), and the model it would resolve to today. Include the claude-sdk seat when
CLAUDE_SDK_API_KEY is set (Agent SDK, Opus 5 by default, 1M context). Flag Claude Code older than v2.1.219 for Opus 5 and Codex CLI older than v0.144.0 for GPT-5.6.
- Flag problems, ranked. Expired auth first (breaks dispatch), then version drift (behavior skew), then missing optional providers (reduced diversity). One line each with the exact fix command (
grok login, opencode auth login, key export).
- State the cost posture. Split the roster into included (Claude, agy, copilot, ollama, cursor-agent) versus billed (codex, perplexity, grok, openrouter, atlascloud, claude-sdk) so the user can predict what a multi-provider workflow costs before running it.
Output shape
A single table, one row per provider, columns: Provider, Status, Auth, Model, Cost. Below it, at most three ranked fix-it lines. No prose beyond that.
Guardrails
- Report only what the detection scripts return this session. A provider that worked yesterday but fails detection now is DOWN; do not soften it.
1---2name: provider-health3description: Starter: one-screen provider health summary — availability, auth method, version drift, and cost posture for every seatable provider4---5
6# Provider Health Check Summary (Starter Pack)
7
8Produce a one-screen health summary of every provider Octopus can seat, so the user knows what a workflow will actually dispatch to before spending money.
9
10## When to use
11
12The user asks "what providers do I have?", a workflow banner showed `(unavailable - skipping)`, or a session is starting on a new machine.
13
14## Steps
15
161. **Detect.** Run `${CLAUDE_PLUGIN_ROOT}/scripts/helpers/check-providers.sh` and `${CLAUDE_PLUGIN_ROOT}/scripts/helpers/check-versions.sh`.
172. **Summarize per provider.** For each seat report: available or not, auth method (env key, oauth session, subscription), and the model it would resolve to today. Include the claude-sdk seat when `CLAUDE_SDK_API_KEY` is set (Agent SDK, Opus 5 by default, 1M context). Flag Claude Code older than v2.1.219 for Opus 5 and Codex CLI older than v0.144.0 for GPT-5.6.
183. **Flag problems, ranked.** Expired auth first (breaks dispatch), then version drift (behavior skew), then missing optional providers (reduced diversity). One line each with the exact fix command (`grok login`, `opencode auth login`, key export).
194. **State the cost posture.** Split the roster into included (Claude, agy, copilot, ollama, cursor-agent) versus billed (codex, perplexity, grok, openrouter, atlascloud, claude-sdk) so the user can predict what a multi-provider workflow costs before running it.
20
21## Output shape
22
23A single table, one row per provider, columns: Provider, Status, Auth, Model, Cost. Below it, at most three ranked fix-it lines. No prose beyond that.
24
25## Guardrails
26
27- Report only what the detection scripts return this session. A provider that worked yesterday but fails detection now is DOWN; do not soften it.
gabrielmoreira