081 Troubleshooting B1b50ae5

# Troubleshooting

Solutions for common issues when using skene-growth.

## LM Studio

### Context length error

```
Error code: 400 - {'error': 'The number of tokens to keep from the initial prompt is greater than the context length...'}
```

The model's context length is too small for the analysis. To fix:

1. In LM Studio, unload the current model
2. Go to **Developer > Load**
3. Click on **Context Length: Model supports up to N tokens**
4. Set it to the maximum supported value
5. Reload to apply changes

Reference: [lmstudio-ai/lmstudio-bug-tracker#237](https://github.com/lmstudio-ai/lmstudio-bug-tracker/issues/237)

### Connection refused

Ensure:
- LM Studio is running
- A model is loaded and ready
- The server is running on the default port (`http://localhost:1234`)

For a custom port:

```bash
export LMSTUDIO_BASE_URL="http://localhost:8080/v1"
```

## Ollama

### Connection refused

Ensure:
- Ollama is running (`ollama serve`)
- A model is pulled and available (`ollama list`)
- The server is on the default port (`http://localhost:11434`)

Getting started with Ollama:

```bash
# Pull a model
ollama pull llama3.3

# Start the server (usually runs automatically after install)
ollama serve
```

For a custom port:

```bash
export OLLAMA_BASE_URL="http://localhost:8080/v1"
```

## API key issues

### "No API key" or fallback to sample report

If `analyze` runs without an API key, it falls back to showing a sample preview (equivalent to `audit`). Set your key using one of:

```bash
# CLI flag
uvx skene-growth analyze . --api-key "your-key"

# Environment variable
export SKENE_API_KEY="your-key"

# Config file (interactive)
uvx skene-growth config
```

### Wrong provider for API key

Make sure the API key matches the provider. An OpenAI key won't work with `--provider gemini`.

## Provider issues

### Unknown provider

Valid provider names:
- `openai`
- `gemini`
- `anthropic` or `claude`
- `lmstudio`, `lm-studio`, or `lm_studio`
- `ollama`
- `generic`, `openai-compatible`, or `openai_compatible`

### Generic provider: missing base URL

The `generic` provider requires a base URL:

```bash
uvx skene-growth analyze . --provider generic --base-url "http://localhost:8000/v1" --model "your-model"
```

Or set via environment variable:

```bash
export SKENE_BASE_URL="http://localhost:8000/v1"
```

## File not found errors

### Manifest not found (plan/build commands)

The `plan` and `build` commands look for files in `./skene-context/` by default. Make sure you've run `analyze` first:

```bash
uvx skene-growth analyze .   # Creates ./skene-context/growth-manifest.json
uvx skene-growth plan        # Reads from ./skene-context/
```

Or specify paths explicitly:

```bash
uvx skene-growth plan --manifest ./path/to/manifest.json --template ./path/to/template.json
uvx skene-growth plan --context ./my-output-dir
```

### Growth plan not found (build command)

```bash
uvx skene-growth plan    # Creates ./skene-context/growth-plan.md
uvx skene-growth build   # Reads from ./skene-context/

# Or specify explicitly
uvx skene-growth build --plan ./path/to/growth-plan.md
```

## Rate limit errors

When a provider returns a rate limit error, skene-growth silently falls back to a cheaper model. This keeps the workflow moving but means the output was generated by a different model than configured.

If you need output from a specific model (e.g. during benchmarking), use `--no-fallback`:

```bash
uvx skene-growth analyze . --no-fallback
```

With `--no-fallback`, the CLI retries the same model with exponential backoff.
If all 3 retries are exhausted, the command raises an error instead of switching models.

## Debug mode

Use `--debug` on any command to log all LLM input and output to `.skene-growth/debug/`:

```bash
uvx skene-growth analyze . --debug
uvx skene-growth plan --debug
uvx skene-growth chat --debug
```

Debug mode can also be enabled via environment variable or config:

```bash
export SKENE_DEBUG=true
```

```toml
# .skene-growth.config
debug = true
```

The debug logs show the full prompts sent to the LLM and the complete responses, which is useful for diagnosing unexpected output or provider-specific issues.

## Getting help

- GitHub issues: [github.com/SkeneTechnologies/skene-growth/issues](https://github.com/SkeneTechnologies/skene-growth/issues)
- Documentation: [www.skene.ai/resources/docs/skene-growth](https://www.skene.ai/resources/docs/skene-growth)