The “it's not you, it's the context window” desk. 🎫
Take a number. Or just Ctrl+F.
Order of operations: slmcode doctor → permissions → prompt size →
“did I ask it to rewrite the universe?” Usually it’s #4.
??? question "🧱 How do I switch between Go and Python pipelines?"
bash slmcode blocks apply go # Go pipeline + go-worker + go-tester slmcode blocks apply python # Python pipeline + python-worker + pytest slmcode blocks apply react # React/TS pipeline + react-tester
In Studio: Settings → Pack Selector, or Pipeline tab → preset selector.
??? question "🧱 Can I create my own pipeline or agent blocks?"
Yes! Drop YAML files in .slmcode/blocks/pipelines/ or .slmcode/blocks/agents/.
Blocks are api_version: blocks/v1 documents with marketplace metadata (id, version, author, license, tags).
Run slmcode blocks validate to check your custom blocks. See 🧱 Blocks for the full schema.
??? question "🧱 Where are the built-in blocks?"
Embedded in the binary via go:embed — always available. Project-level blocks in .slmcode/blocks/ override them.
The discovery order is: project → user → env → builtin (first ID wins).
??? question "👻 slmcode: command not found"
Add the install dir to PATH and open a new shell:
- User install → `~/.local/bin`
- System → `$(brew --prefix)/bin` or `/usr/local/bin`
- Windows → `%LOCALAPPDATA%\slmcode\bin`
```bash
hash -r
which slmcode
slmcode version
```
??? question "🔐 Checksum mismatch on install" Re-run the installer. Corporate SSL inspection sometimes rewrites downloads. Try from a clean network or download the asset from GitHub Releases manually.
??? question "🍺 Homebrew SHA256 fails"
Formula SHAs track a specific release build. Update to latest main formula URL
or reinstall after a formula sync commit.
??? question "🔴 doctor shows unreachable LLM"
1. Is the server up? (omlx start, ollama serve, LM Studio running…)
2. Does endpoint match reality (note /v1 vs bare host)?
3. Is the API key set for cloud providers?
4. Try:
```bash
curl -s "$ENDPOINT/models" | head
slmcode config
slmcode doctor
```
??? question "🪐 Wrong model / wrong gateway"
Per-agent endpoints register unique backend keys. Check .slmcode/agents/*.yaml
and Studio → Agents. Global config can be fine while one specialist phones Mars.
??? question "🦙 Ollama vs OpenAI-compat confusion"
Ollama uses its native API. Everything else is Chat Completions.
Don't point provider: ollama at an OpenAI URL (or vice versa) unless you enjoy puzzles.
??? question "📂 It edits the wrong files"
- Set a correct model_profiles.<family>.context_limit, or lower context_role_budget / repo_map_tokens
- Pin atomic-coding
- Add a stern AGENTS.md
- Force a fresh explore once:
```bash
SLMCODE_FORCE_EXPLORE=1 slmcode run -v "…"
```
??? question "🥴 JSON / tool calls look drunk"
Expected on weaker SLMs. The harness repairs common breakage and retries.
Raise think_passes / retries. Prefer models with decent tool-calling.
??? question "🤷 Reviewer always rejects (or always accepts)"
Check disk evidence in the Live feed. Prefer permission: review on important repos.
Speculative cancel + disk rename acceptance exist specifically for flaky judges.
??? question "🐢 Second run is slow / re-explores everything"
It shouldn't — explore reuse should kick in. If CONTEXT is empty or wiped, explore returns.
Don't delete .slmcode/ for fun mid-project. The flywheel has feelings.
??? question "🚧 Studio port already in use"
bash slmcode studio --listen 127.0.0.1:7421
??? question "👻 UI looks stale after I hacked ui/"
Rebuild the Go binary — assets are go:embed'd.
```bash
make build && ./bin/slmcode studio
```
??? question "🛟 Resume doesn't continue where I left off"
Confirm .slmcode/queries/<id>/react/ exists. /resume needs a checkpoint from /stop or Ctrl+C —
not a crashed machine mid-write with no flush.
??? question "✍️ Did it write to disk?"
bash slmcode diff ls .slmcode/pending/ # if permission=review
??? question "🎭 How do I demo without touching files?"
bash slmcode config set permission dry-run # or slmcode run --dry-run "…"
slmcode doctor(redact keys)slmcode run -v "…"snippet of the failing phase- Open a GitHub issue: UnicoLab/smlcode
Emotional support tip: if the model invents an API that doesn’t exist, it’s not gaslighting you — it’s optimistic. Lower context. Pin a skill. Try again with snacks.
☀️ Made with ♥ by UnicoLab