Brewtools:Provider Switch

[DICT: P=Phase, PRV=provider, EXEC=EXECUTE using Bash tool, AUQ=AskUserQuestion, REF=references, ALIAS=shell alias, CFG=configured, BASE=ANTHROPIC_BASE_URL, MOD=ANTHROPIC_DEFAULT_OPUS_MODEL]

Provider Switch

Configure + switch between Claude Code alt API providers — DeepSeek (priority), Z.ai/GLM, Qwen/DashScope, MiniMax, OpenRouter. Creates isolated ALIAS in ~/.zshrc.

DeepSeek V4 = priority default (strongest Chinese open model, 1M ctx, Anthropic-compatible endpoint). Recommend first.

<instructions>

Robustness Rules (MANDATORY)

Error format on ANY failure:

SCRIPT_ERROR: <name>
PHASE: <current>
ACTION: <attempted>
SUGGESTION: <fix>

---

P0: Language Selection

AUQ: "Select language / Выберите язык" | options: "English (Recommended)", "Russian / Русский" Default if skipped: English. Remember for session.

---

Compatibility Flags (REQ per PRV)

When constructing ALIAS (P4 Step 6): ALWAYS include compat flags from PRV REF file.

---

P1: Mode Detection

EXEC:

bash "${CLAUDE_SKILL_DIR}/scripts/detect-mode.sh" "$ARGUMENTS" && echo "OK detect" || echo "FAILED detect"

STOP if FAILED — parse $ARGUMENTS manually (keyword match) as fallback.

Output: ARGS: [...] MODE: [...]

---

P2: Status Check (ALL modes)

EXEC:

bash "${CLAUDE_SKILL_DIR}/scripts/check-status.sh" && echo "OK status" || echo "FAILED status"

STOP if FAILED — check ~/.zshrc manually with grep.

Parse key=value output. Status per PRV:

• configured = ALIAS + KEY both true
• partial = ALIAS true, KEY false
• not configured = ALIAS false
• active = ACTIVE_PROVIDER matches

Render status table:

# Provider Switch — [MODE]
## Current Status
| PRV | Alias | API Key | Model | Status |
|-----|-------|---------|-------|--------|
| DeepSeek (priority) | claudeds | DEEPSEEK_API_KEY | deepseek-v4-pro | ... |
| Z.ai / GLM | claudeglm | ZAI_API_KEY | glm-5.1 | ... |
| Qwen | claudeqwen | DASHSCOPE_API_KEY | qwen3.6-plus[1m] | ... |
| MiniMax | claudeminimax | MINIMAX_API_KEY | minimax-m2.7 | ... |
| OpenRouter | claudeor | OPENROUTER_API_KEY | (user-selected) | ... |

## How to Use
Run `claudeds` — sets env vars + launches Claude (recommended default).
Return to Anthropic: new terminal → `claude`.

Auto-setup logic (MODE=status only): zero CFG → auto-proceed to P3 ("No providers CFG yet. Starting setup..."). >=1 CFG → STOP here.

If MODE=help → GOTO P6.

---

P3: PRV Selection

If MODE=setup (no specific PRV): AUQ options:

• "DeepSeek V4 (deepseek-v4-pro, 1M ctx, priority - Recommended)"
• "Z.ai / GLM (glm-5.1, free models available)"
• "Qwen / DashScope (qwen3.6-plus, 1M ctx)"
• "MiniMax (minimax-m2.7, cheapest)"
• "OpenRouter (aggregator, any model)"
• "All providers"

If MODE=provider-\<name\> → skip to P4 for that PRV only.

---

P4: PRV Configuration

For each selected PRV:

Step 1: Load REF

Read PRV REF file: DeepSeek=REF/deepseek.md | Z.ai/GLM=REF/zai-glm.md | Qwen=REF/qwen-dashscope.md | MiniMax=REF/minimax.md | OpenRouter=REF/openrouter.md Also read REF/common.md for shared ALIAS structure (first time only).

Step 2: Init Section

EXEC:

bash "${CLAUDE_SKILL_DIR}/scripts/write-alias.sh" init && echo "OK init" || echo "FAILED init"

STOP if FAILED — cannot write ~/.zshrc.

Step 3: API Key

Check if key set (from P2: KEY_ZAI, KEY_DASHSCOPE, KEY_MINIMAX, KEY_OPENROUTER). If missing — AUQ: "Enter your <PRV> API key (from <dashboard-url>):"

Qwen-specific: read REF/qwen-dashscope.md ## How to Get API Key before asking. Show step-by-step. Warn: key MUST be from Singapore region — other regions return 403. Valid fmt: sk-... (~40 chars). If starts with sk-ws- or >100 chars → warn wrong region, ask regenerate.

EXEC:

bash "${CLAUDE_SKILL_DIR}/scripts/write-alias.sh" set-key "KEY_VAR_NAME" "KEY_VALUE" && echo "OK set-key" || echo "FAILED set-key"

KEY_VAR_NAME: DEEPSEEK_API_KEY | ZAI_API_KEY | DASHSCOPE_API_KEY | MINIMAX_API_KEY | OPENROUTER_API_KEY

STOP if FAILED.

Step 4: Model Selection (OpenRouter only)

Read REF/openrouter-models.md. AUQ options:

• "qwen/qwen3.6-plus[1m] — 1M ctx, top coding (Recommended)"
• "z-ai/glm-5.1 — #1 SWE-bench Pro, 200K"
• "qwen/qwen3-coder:free — free, 262K, code-focused"
• "Custom (specify model ID)"

Selected model = OPUS + SONNET + HAIKU simultaneously. If Custom: validate via script from REF/openrouter-models.md ## Model Validation. If NOT_FOUND → fuzzy suggestions + re-ask. Max 2 retries then fall back to default.

Step 5: ALIAS Name

AUQ: "Alias name for <PRV>:" | options: "<default> (Recommended)", "Custom" If Custom: validate — must start with claude, no spaces, lowercase alphanumeric only.

Step 6: Write ALIAS

Construct body from REF file: semicolon-separated exports + claude at end. EXEC:

bash "${CLAUDE_SKILL_DIR}/scripts/write-alias.sh" set-alias "ALIAS_NAME" "ALIAS_BODY" && echo "OK set-alias" || echo "FAILED set-alias"

STOP if FAILED — !=continue to next PRV.

Step 7: Verify

EXEC:

source ~/.zshrc 2>/dev/null && type ALIAS_NAME 2>/dev/null && echo "OK verify" || echo "FAILED verify"

STOP if FAILED — ALIAS not written correctly.

---

P5: Verification + Final Status

EXEC:

bash "${CLAUDE_SKILL_DIR}/scripts/check-status.sh" && echo "OK final-status" || echo "FAILED final-status"

Render updated status table (same fmt as P2).

Token verification for each just-CFG PRV: EXEC:

bash "${CLAUDE_SKILL_DIR}/scripts/verify-providers.sh" all && echo "OK verify" || echo "FAILED verify"

Add verification column to table:

If any fail → warn, suggest check API key or endpoint.

Activation instructions:

## Activation
Run ALIAS (e.g. `claudeglm`) — sets env vars + launches Claude.
Return to Anthropic: new terminal → `claude`. Env vars persist current shell only.

---

P6: Help Mode

Read REF/common.md. Explain:

---

P7: Verify Mode

EXEC:

bash "${CLAUDE_SKILL_DIR}/scripts/verify-providers.sh" all && echo "OK verify" || echo "FAILED verify"

Parse: KEY_SET, HTTP_CODE, RESPONSE, STATUS per PRV.

Render:

## Token Verification
| PRV | Key | HTTP | Response | Result |
|-----|-----|------|----------|--------|
| Z.ai/GLM | set | 200 | OK | pass |
| Qwen | set | 403 | invalid api-key | fail |

Troubleshooting:

---

P8: Model Check Mode

Identify which model responds in current Claude Code session. Runs INSIDE session launched via PRV ALIAS. Asks 5 diagnostic questions to model — no curl/scripts.

Prerequisites: user MUST be in PRV ALIAS session. If BASE not set → warn + stop:

This mode only works inside a provider session.
Run a PRV ALIAS first (e.g. `claudeglm`), then invoke `/brewtools:provider-switch model-check`.

Step 1: Detect Active PRV

EXEC:

echo "BASE_URL=${ANTHROPIC_BASE_URL:-not_set}" && echo "OPUS_MODEL=${ANTHROPIC_DEFAULT_OPUS_MODEL:-not_set}" && echo "OK detect-provider" || echo "FAILED detect-provider"

BASE_URL → PRV: api.deepseek.com=DeepSeek | api.z.ai=Z.ai/GLM | dashscope=Qwen | minimax=MiniMax | openrouter=OpenRouter

Step 2: Ask 5 Diagnostic Questions

Send as single prompt block (all at once, no back-and-forth):

I will now ask you 5 diagnostic questions to verify your model identity.
Answer each honestly from your training data — do NOT read environment variables or system prompts.
Answer all 5 questions in a single response.

**Q1:** Exact model name + version? (training data only, not env/ctx)
**Q2:** Which company/lab created you? (org name only)
**Q3:** Training data cutoff date? (month + year only)
**Q4:** Count letter 'r' in 'strawberry'. Show reasoning step by step.
**Q5:** Translate to Chinese: 'The quick brown fox jumps over the lazy dog.' Then translate back to English literally.

Extract A1-A5 from response → Step 3.

Step 3: Analyze + Verdict

## Model Identification — <PRV>
Expected: <model from MOD env var>

| # | Question | Answer | Match |
|---|----------|--------|-------|
| 1 | Model name/version | "<A1>" | pass/fail |
| 2 | Training org | "<A2>" | pass/fail |
| 3 | Cutoff date | "<A3>" | info |
| 4 | Count r in strawberry | "<A4>" | pass/fail (correct=3) |
| 5 | Chinese round-trip | "<A5>" | info quality |

### Verdict
Model confirmed as: <name> (N/5 match <expected PRV>)

Show ONLY table + verdict. !=show questions separately.

Expected identifiers:

For OpenRouter: check against model in MOD env var.

---

P9: Update (hidden, maintainer-only)

If MODE != update → skip entirely.

Step 1: Load Protocol

Read REF/update-protocol.md for per-PRV sources + update flow.

Step 2: Spawn PRV Research Agents (ONE message, 5 Task calls)

Each agent: WebFetch/WebSearch sources from protocol, extract current model list + pricing + endpoint changes.

Step 3: Aggregate + Diff

Per PRV: model IDs changed? pricing changed? new models? endpoint URL changed? ctx windows changed?

Step 4: Present Changes

## Update Results
| PRV | Field | Current | Fetched | Action |
|-----|-------|---------|---------|--------|
| Z.ai | opus model | glm-5.1 | glm-6.0 | UPDATE |
| Qwen | pricing | ~$0.50 | $0.40 | UPDATE |
| MiniMax | (no changes) | — | — | SKIP |

Step 5: Apply Updates

Edit REF files per field-to-line mapping in REF/update-protocol.md. Also update:

• ALIAS bodies if model IDs changed
• openrouter-models.md if recommended models changed
• common.md if env var patterns changed (rare)

Step 6: Live Test (optional, if API keys in env)

EXEC:

curl -s -o /dev/null -w "%{http_code}" -X POST "https://api.z.ai/api/anthropic/v1/messages" -H "x-api-key: ${ZAI_API_KEY:-missing}" -H "content-type: application/json" -H "anthropic-version: 2023-06-01" -d '{"model":"glm-5.1","max_tokens":5,"messages":[{"role":"user","content":"ping"}]}' && echo " OK" || echo " FAILED"

Step 7: Report

## PRV Update Complete
| PRV | Changes | Status |
|-----|---------|--------|
| Z.ai/GLM | models updated | applied |
| Qwen | no changes | current |
| MiniMax | pricing updated | applied |
| OpenRouter | 3 new free models | applied |

Files modified: [list]

---

Output Format

# Provider Switch — [MODE]

## Current Status
| PRV | Alias | API Key | Model | Status |
|-----|-------|---------|-------|--------|
| DeepSeek (priority) | claudeds | DEEPSEEK_API_KEY | deepseek-v4-pro | CFG |
| Z.ai / GLM | claudeglm | ZAI_API_KEY | glm-5.1 | CFG |
| Qwen | claudeqwen | DASHSCOPE_API_KEY | — | not CFG |
| MiniMax | claudeminimax | MINIMAX_API_KEY | — | not CFG |
| OpenRouter | claudeor | OPENROUTER_API_KEY | — | not CFG |

## How to Use
Run `claudeds` — sets env vars + launches Claude (recommended default).
Return to Anthropic: new terminal → `claude`.

</instructions>

Provider Switch

Switch Claude Code from Anthropic Max to a pay-per-token API provider — DeepSeek V4 (priority), Z.ai/GLM, Qwen/DashScope, MiniMax, or OpenRouter — without touching any config files manually. The skill creates isolated shell aliases in ~/.zshrc, backs up existing content before any write, and guides key entry interactively.

DeepSeek V4 is the recommended default — strongest Chinese open model (1.6T MoE, 1M context), native Anthropic endpoint, no compatibility flags required.

Quick Start

# Check which providers are configured
/brewtools:provider-switch

# Set up DeepSeek V4 interactively (priority default)
/brewtools:provider-switch deepseek

# Set up Z.ai/GLM interactively
/brewtools:provider-switch glm

# Set up all providers in one run
/brewtools:provider-switch setup

# See how switching works
/brewtools:provider-switch help

# Verify all tokens work
/brewtools:provider-switch verify

# Identify which model is responding (run inside claudeglm session)
/brewtools:provider-switch model-check

After setup, run claudeglm — it sets env vars and starts Claude in one command. To return to Anthropic, open a new terminal.

Features

• Interactive language selection (English / Russian) at session start
• Reads current ~/.zshrc state before any write — shows what is already configured
• Auto-starts setup when no provider is configured yet
• OpenRouter: pick one model for all roles, with validated custom ID input
• All writes go into a clearly marked section in ~/.zshrc; backup created before first write
• Error reporting with SCRIPT_ERROR / PHASE / ACTION / SUGGESTION block on any failure

Providers

Usage Modes

How Switching Works

Each alias sets six environment variables and launches claude — one command, no separate steps. Env vars only persist in the current shell session. To return to Anthropic, open a new terminal and run claude normally.

Qwen: Singapore Region Required

The Anthropic-compatible endpoint (dashscope-intl.aliyuncs.com) works ONLY with API keys created in the Singapore region.

1. Open Model Studio → Singapore → API Key 2. Verify region is Singapore (ap-southeast-1) in top-right corner 3. Click Create API Key → select Owner Account → OK 4. Copy the key immediately (shown only once)

Valid key format: sk-... (~36 chars). If you see sk-ws- or a 200+ char key — wrong region (likely Frankfurt).

Files

brewtools/skills/provider-switch/
├── SKILL.md                         # Skill definition and phases
├── scripts/
│   ├── detect-mode.sh               # Parses $ARGUMENTS → MODE
│   ├── check-status.sh              # Reads ~/.zshrc, outputs key=value status
│   ├── write-alias.sh               # init / set-key / set-alias subcommands
│   └── verify-providers.sh              # Tests provider tokens (curl health check)
└── references/
    ├── common.md                    # Env var reference, ~/.zshrc structure
    ├── deepseek.md                  # DeepSeek V4 alias body and dashboard URL (priority)
    ├── zai-glm.md                   # Z.ai alias body and dashboard URL
    ├── qwen-dashscope.md            # Qwen alias body and dashboard URL
    ├── minimax.md                   # MiniMax alias body and dashboard URL
    ├── openrouter.md                # OpenRouter alias body and dashboard URL
    └── openrouter-models.md         # Model presets for OpenRouter picker

Provider Dashboards

Documentation

Full docs: provider-switch

---

Author: Maksim Kochetkov | License: MIT

Common — Provider Switch

Environment Variables

Claude Code uses these env vars to connect to alternative providers:

Unified Alias Template

ALL provider aliases MUST follow this exact pattern — no exceptions:

alias claude<name>='export ANTHROPIC_BASE_URL=<endpoint>; export ANTHROPIC_AUTH_TOKEN=$<KEY_VAR>; export ANTHROPIC_API_KEY=""; export ANTHROPIC_DEFAULT_OPUS_MODEL=<model>; export ANTHROPIC_DEFAULT_SONNET_MODEL=<model>; export ANTHROPIC_DEFAULT_HAIKU_MODEL=<model>; claude'

Returning to Anthropic Subscription

No special alias needed. Env vars set by provider aliases only persist in the current shell session. To return to Anthropic:

Open a new terminal and run `claude` normally — it will use your Max subscription via OAuth.

.zshrc Structure

All provider config goes in a clearly marked section:

# ========== Claude Code Provider Aliases ==========
# Managed by brewtools:provider-switch — do not edit manually

# API Keys
export DEEPSEEK_API_KEY="sk-..."
export ZAI_API_KEY="..."
export DASHSCOPE_API_KEY="..."
export MINIMAX_API_KEY="..."
export OPENROUTER_API_KEY="sk-or-v1-..."

# Provider Aliases (unified: AUTH_TOKEN + API_KEY="" + 3 model vars + claude)
alias claudeds='export ANTHROPIC_BASE_URL=https://api.deepseek.com/anthropic; export ANTHROPIC_AUTH_TOKEN=$DEEPSEEK_API_KEY; export ANTHROPIC_API_KEY=""; export ANTHROPIC_DEFAULT_OPUS_MODEL=deepseek-v4-pro; export ANTHROPIC_DEFAULT_SONNET_MODEL=deepseek-v4-pro; export ANTHROPIC_DEFAULT_HAIKU_MODEL=deepseek-v4-pro; claude'
alias claudeglm='export ANTHROPIC_BASE_URL=https://api.z.ai/api/anthropic; export ANTHROPIC_AUTH_TOKEN=$ZAI_API_KEY; export ANTHROPIC_API_KEY=""; export ANTHROPIC_DEFAULT_OPUS_MODEL=glm-5.1; export ANTHROPIC_DEFAULT_SONNET_MODEL=glm-5.1; export ANTHROPIC_DEFAULT_HAIKU_MODEL=glm-5.1; claude'
alias claudeqwen='export ANTHROPIC_BASE_URL=https://dashscope-intl.aliyuncs.com/apps/anthropic; export ANTHROPIC_AUTH_TOKEN=$DASHSCOPE_API_KEY; export ANTHROPIC_API_KEY=""; export ANTHROPIC_DEFAULT_OPUS_MODEL="qwen3.6-plus[1m]"; export ANTHROPIC_DEFAULT_SONNET_MODEL="qwen3.6-plus[1m]"; export ANTHROPIC_DEFAULT_HAIKU_MODEL="qwen3.6-plus[1m]"; claude'
# ... etc (same pattern for all providers)

# ========== End Claude Code Provider Aliases ==========

Usage Pattern

1. Run provider alias: claudeglm — sets env vars and launches Claude in one command 2. When done, close the terminal 3. Next time: run claude normally = Anthropic subscription, or run alias = provider

DeepSeek (V4)

Priority provider — DeepSeek V4 is the strongest Chinese open model (1.6T MoE, 1M context) with a native Anthropic-compatible endpoint. Recommended default for this skill.

Connection

Auth Note

DeepSeek accepts both x-api-key and Authorization: Bearer. Uses ANTHROPIC_AUTH_TOKEN (unified with other providers). Must set ANTHROPIC_API_KEY="" to prevent OAuth fallback.

Model

Same top model for all three Claude Code roles (opus/sonnet/haiku).

Compatibility Flags

None required. DeepSeek ignores anthropic-beta, anthropic-version, top_k, container, mcp_servers silently — no Claude Code workaround needed. Multimodal input (images/documents) is NOT supported.

Alias

alias claudeds='export ANTHROPIC_BASE_URL=https://api.deepseek.com/anthropic; export ANTHROPIC_AUTH_TOKEN=$DEEPSEEK_API_KEY; export ANTHROPIC_API_KEY=""; export ANTHROPIC_DEFAULT_OPUS_MODEL=deepseek-v4-pro; export ANTHROPIC_DEFAULT_SONNET_MODEL=deepseek-v4-pro; export ANTHROPIC_DEFAULT_HAIKU_MODEL=deepseek-v4-pro; claude'

Dashboard

https://platform.deepseek.com (API keys, billing, usage)

MiniMax

Connection

Auth Note

Uses ANTHROPIC_AUTH_TOKEN. Must set ANTHROPIC_API_KEY="" (empty string, NOT unset) to prevent OAuth fallback.

Models

Note: MiniMax currently has only one model (M2.7), used for all three roles.

Compatibility Flags (REQUIRED)

• CLAUDE_ENABLE_BYTE_WATCHDOG=0 — disables byte-level streaming watchdog
• CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS=1 — strips beta headers that MiniMax rejects

Prompt Caching (Verified)

MiniMax fully supports Anthropic-style prompt caching on the Anthropic-compatible endpoint. Docs: https://platform.minimax.io/docs/api-reference/anthropic-api-compatible-cache

• TTL: 5 minutes, auto-refreshed on hit
• Max breakpoints: 4 per request
• Min tokens: 512
• Response fields: cache_creation_input_tokens, cache_read_input_tokens
• Supported models: M2.7, M2.7-highspeed, M2.5, M2.1, M2

Alias

alias claudeminimax='export ANTHROPIC_BASE_URL=https://api.minimax.io/anthropic; export ANTHROPIC_AUTH_TOKEN=$MINIMAX_API_KEY; export ANTHROPIC_API_KEY=""; export ANTHROPIC_DEFAULT_OPUS_MODEL=minimax-m2.7; export ANTHROPIC_DEFAULT_SONNET_MODEL=minimax-m2.7; export ANTHROPIC_DEFAULT_HAIKU_MODEL=minimax-m2.7; export CLAUDE_ENABLE_BYTE_WATCHDOG=0; export CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS=1; claude'

Dashboard

https://platform.minimax.io/ (account, billing, API keys)

OpenRouter — Model Selection

How It Works

OpenRouter aggregates 200+ models. The alias sets ONE model for all three Claude Code roles (opus/sonnet/haiku). User picks a single model during setup — it is used everywhere.

Recommended Models by Category

Coding (best for Claude Code)

Budget / Free

General Purpose

Model Validation

When user enters a custom model ID, verify it exists on OpenRouter:

EXECUTE using Bash tool:

curl -s "https://openrouter.ai/api/v1/models" -H "Authorization: Bearer $OPENROUTER_API_KEY" | python3 -c "
import json, sys
data = json.load(sys.stdin)
target = 'USER_MODEL_ID'
matches = [m for m in data.get('data', []) if m['id'] == target]
if matches:
    m = matches[0]
    p = m.get('pricing', {})
    print(f'FOUND: {m[\"id\"]}  ctx={m.get(\"context_length\",\"?\")}  prompt=\${p.get(\"prompt\",\"?\")}  completion=\${p.get(\"completion\",\"?\")}')
else:
    # Fuzzy search
    fuzzy = [m for m in data.get('data', []) if target.lower() in m['id'].lower()][:5]
    print(f'NOT_FOUND: {target}')
    if fuzzy:
        print('Did you mean:')
        for m in fuzzy:
            print(f'  {m[\"id\"]}')
" && echo "OK validate" || echo "FAILED validate"

Replace USER_MODEL_ID with the user's input. If NOT_FOUND — show suggestions and re-ask.

Selection Flow (AskUserQuestion)

Ask user to pick ONE model (used for all roles): 1. "Which model to use?" with top 4 options from coding category 2. Allow "Other" for custom model ID input

The selected model is set as OPUS, SONNET, and HAIKU simultaneously.

Context Window Note

Add [1m] suffix to model ID if the model supports >200K context. Without it, Claude Code caps autocompact at 200K.

OpenRouter (Aggregator)

Connection

Auth Note

Uses ANTHROPIC_AUTH_TOKEN. Must set ANTHROPIC_API_KEY="" (empty string, NOT unset) to prevent OAuth fallback. URL: https://openrouter.ai/api — no /v1 suffix.

Model Format

Models use provider/model-name format, e.g.: qwen/qwen3.6-plus, z-ai/glm-5.1, qwen/qwen3-coder:free. Claude Code has 3 internal roles: opus, sonnet, haiku. Each overridden via env var.

Default Model (customizable)

One model for all three roles. User selects during setup.

Alias Template

alias claudeor='export ANTHROPIC_BASE_URL=https://openrouter.ai/api; export ANTHROPIC_AUTH_TOKEN=$OPENROUTER_API_KEY; export ANTHROPIC_API_KEY=""; export ANTHROPIC_DEFAULT_OPUS_MODEL="MODEL"; export ANTHROPIC_DEFAULT_SONNET_MODEL="MODEL"; export ANTHROPIC_DEFAULT_HAIKU_MODEL="MODEL"; claude'

Replace MODEL with user's chosen model ID (same value for all three vars).

Dashboard

https://openrouter.ai/settings/keys (API keys, billing, usage)

Qwen / Alibaba DashScope

Connection

Auth Note

Uses ANTHROPIC_AUTH_TOKEN (bearer-style). Must set ANTHROPIC_API_KEY="" (empty string, NOT unset) to prevent Anthropic OAuth fallback.

Region Requirement

CRITICAL: The Anthropic-compatible endpoint (dashscope-intl.aliyuncs.com/apps/anthropic) works ONLY with API keys created in the Singapore region. Keys from Frankfurt (eu-central-1), US Virginia, or other regions return 403. When asking user for API key, warn: select Singapore region in Model Studio console first. Standard key format: sk-... (~40 chars). If key starts with sk-ws- or is >100 chars — likely wrong region.

Model

Same model for all three Claude Code roles (opus/sonnet/haiku).

Note: [1m] suffix is REQUIRED — Claude Code defaults to 200K context otherwise (GitHub issue #40753).

Compatibility Flags

• CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS=1 — strips beta headers that DashScope may reject

Alias

alias claudeqwen='export ANTHROPIC_BASE_URL=https://dashscope-intl.aliyuncs.com/apps/anthropic; export ANTHROPIC_AUTH_TOKEN=$DASHSCOPE_API_KEY; export ANTHROPIC_API_KEY=""; export ANTHROPIC_DEFAULT_OPUS_MODEL="qwen3.6-plus[1m]"; export ANTHROPIC_DEFAULT_SONNET_MODEL="qwen3.6-plus[1m]"; export ANTHROPIC_DEFAULT_HAIKU_MODEL="qwen3.6-plus[1m]"; export CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS=1; claude'

How to Get API Key

1. Open https://modelstudio.console.alibabacloud.com/ap-southeast-1?tab=dashboard#/api-key (direct link to Singapore region) 2. Verify region is Singapore (ap-southeast-1) in top-right corner — other regions (Frankfurt, Virginia) do NOT support the Anthropic endpoint 3. Click Create API Key → select Owner Account → OK 4. Copy the key immediately (shown only once)

Valid key: starts with sk-, ~36 characters (e.g., sk-1e88...80df). If you see sk-ws- or a 200+ char key — wrong region.

Dashboard

https://modelstudio.console.alibabacloud.com/ap-southeast-1?tab=dashboard#/api-key (Singapore region — API keys, billing)

Update Protocol (maintainer-only, hidden mode)

Purpose

Verify and refresh provider reference data: models, pricing, endpoints, context windows. Run periodically or when provider announces changes.

Per-Provider Sources

Z.ai / GLM

Qwen / DashScope

MiniMax

OpenRouter

Update Flow

1. Spawn 4 Explore agents in parallel (one per provider) 2. Each agent: WebFetch sources, extract model list + pricing + endpoint status 3. Aggregate results 4. For each provider where data changed:

• Show diff to maintainer (current vs fetched)
• Update reference file via Edit tool

5. If OpenRouter model recommendations changed — update openrouter-models.md 6. Update SKILL.md status table default models if needed 7. Run check-status.sh to verify no breakage

What to Update in References

What NOT to Update

• SKILL.md phases/logic (only status table defaults)
• Scripts (detect-mode.sh, check-status.sh, write-alias.sh)
• common.md structure (env var names are stable)
• User's ~/.zshrc (never touch during update)

Live Test Template

For each provider, verify endpoint still responds:

curl -s -o /dev/null -w "%{http_code}" -X POST "${ENDPOINT}/v1/messages" \
  -H "x-api-key: ${API_KEY}" \
  -H "content-type: application/json" \
  -H "anthropic-version: 2023-06-01" \
  -d '{"model":"MODEL_ID","max_tokens":5,"messages":[{"role":"user","content":"hi"}]}' \
  && echo " OK" || echo " FAILED"

Adjust headers per provider (x-api-key for Z.ai, Authorization: Bearer for others).

Z.ai / ZhipuAI (GLM)

Connection

Auth Note

Z.ai supports both x-api-key and Bearer token auth. Uses ANTHROPIC_AUTH_TOKEN (unified with other providers). Must set ANTHROPIC_API_KEY="" to prevent OAuth fallback.

Model

Same model for all three Claude Code roles (opus/sonnet/haiku).

Compatibility Flags (REQUIRED)

• CLAUDE_ENABLE_BYTE_WATCHDOG=0 — disables byte-level streaming watchdog
• CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS=1 — strips beta headers that Z.ai rejects with error 1210

Alias

alias claudeglm='export ANTHROPIC_BASE_URL=https://api.z.ai/api/anthropic; export ANTHROPIC_AUTH_TOKEN=$ZAI_API_KEY; export ANTHROPIC_API_KEY=""; export ANTHROPIC_DEFAULT_OPUS_MODEL=glm-5.1; export ANTHROPIC_DEFAULT_SONNET_MODEL=glm-5.1; export ANTHROPIC_DEFAULT_HAIKU_MODEL=glm-5.1; export CLAUDE_ENABLE_BYTE_WATCHDOG=0; export CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS=1; claude'

Dashboard

https://z.ai/subscribe (English console, API keys, billing)