cline-wandb

Dispatch Cline CLI as a headless fan-out worker via wanDB.ai DeepSeek-V4-Flash inference.

Updated Mar 22, 2026
One-click install
npx skills add https://github.com/diazMelgarejo/orama-system --skill cline-wandb-diazmelgarejo
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: cline-wandb
Source: https://github.com/diazMelgarejo/orama-system/tree/main/bin/orama-system/skills/cline-wandb
Command: npx skills add https://github.com/diazMelgarejo/orama-system --skill cline-wandb-diazmelgarejo

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve? Running Cline CLI non-interactively through wanDB.ai's OpenAI-compatible inference endpoint requires correct provider configuration, secret handling, and verification gates; misconfigured flags silently route through the wrong provider or leak API keys into shell history. ## Core Features & Use Cases - Headless Cline dispatch: Run Cline CLI with DeepSeek-V4-Flash via wanDB.ai using verified flags (--thinking high, -P openai-compatible, -m) for read-only reviews, plan drafts, and implementation passes. - Secure credential handling: Writes the API key directly into providers.json via switch-cline-provider.sh so secrets never appear in argv, ps output, or shell history. - Mandatory quality gates: Enforces a security-reviewer gate before cloud or repo work, a harmonized TDD reference card for code-writing tasks, and a generalized self-review discipline before reporting done. - Use Case: A main agent session needs an independent second opinion on a sanitized change summary; it dispatches a read-only Cline wanDB worker in an isolated workspace and receives structured JSON review output without exposing repository secrets. ## Quick Start Ask the agent to run a read-only Cline wanDB review of a sanitized task summary using the DeepSeek-V4-Flash model with high reasoning in an isolated workspace.

Frequently Asked Questions about cline-wandb

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I run Cline CLI non-interactively with a custom OpenAI-compatible provider?

Pass -P openai-compatible and -m with the exact model id on every invocation, plus --json for machine-readable output and -c for the workspace. Both flags are required because Cline's default provider routing ignores the activated profile in providers.json.

How do I set the Cline CLI API key without exposing it in shell history?

Use the switch-cline-provider.sh script, which writes the resolved key directly into ~/.cline/data/settings/providers.json. Avoid cline auth --apikey, since passing the key as a command argument exposes it via ps and shell history.

Why does Cline CLI return 401 invalid_api_key against wanDB.ai inference?

Sending an OpenAI-Project header or the Python SDK project= kwarg causes a 401 on an otherwise valid wanDB.ai key. Omit the project header entirely; the project variable is only used for the separate weave.init() tracing path.

Why does Cline CLI use the wrong provider even after activating a profile?

Activating a profile in providers.json only updates bookkeeping; the CLI still defaults to its own hosted routing. You must pass both -P openai-compatible and -m explicitly on every invocation to force the correct provider and model.

What is the correct Cline CLI flag for reasoning effort?

The flag is --thinking with levels none, low, medium, high, or xhigh. The commonly assumed --reasoning-effort flag does not exist in Cline CLI v3.0.49, and -a is not a shorthand for auto-approval.

When should I not use the wanDB.ai Cline route?

Do not use it as a replacement for ClinePass DeepSeek Flash or the GLM ClinePass skill, and do not use it for force-pushes, merges, or PR closures, which belong to the main session. Full agentic dispatch currently fails with a generic Bad Request from wanDB.ai, so raw API smoke tests are the confirmed-working path.