singularity-forge/docs/troubleshooting.md

# Troubleshooting

## `/gsd doctor`

The built-in diagnostic tool validates `.gsd/` integrity:

```
/gsd doctor
```

It checks:
- File structure and naming conventions
- Roadmap ↔ slice ↔ task referential integrity
- Completion state consistency
- Git worktree health (worktree and branch modes only — skipped in none mode)
- Stale lock files and orphaned runtime records

## Common Issues

### Auto mode loops on the same unit

**Symptoms:** The same unit (e.g., `research-slice` or `plan-slice`) dispatches repeatedly until hitting the dispatch limit.

**Causes:**
- Stale cache after a crash — the in-memory file listing doesn't reflect new artifacts
- The LLM didn't produce the expected artifact file

**Fix:** Run `/gsd doctor` to repair state, then resume with `/gsd auto`. If the issue persists, check that the expected artifact file exists on disk.

### Auto mode stops with "Loop detected"

**Cause:** A unit failed to produce its expected artifact twice in a row.

**Fix:** Check the task plan for clarity. If the plan is ambiguous, refine it manually, then `/gsd auto` to resume.

### Wrong files in worktree

**Symptoms:** Planning artifacts or code appear in the wrong directory.

**Cause:** The LLM wrote to the main repo instead of the worktree.

**Fix:** This was fixed in v2.14+. If you're on an older version, update. The dispatch prompt now includes explicit working directory instructions.

### `npm install -g gsd-pi` fails

**Common causes:**
- Missing workspace packages — fixed in v2.10.4+
- `postinstall` hangs on Linux (Playwright `--with-deps` triggering sudo) — fixed in v2.3.6+
- Node.js version too old — requires ≥ 20.6.0

### Provider errors during auto mode

**Symptoms:** Auto mode pauses with a provider error (rate limit, server error, auth failure).

**How GSD handles it (v2.26):**

| Error type | Auto-resume? | Delay |
|-----------|-------------|-------|
| Rate limit (429, "too many requests") | ✅ Yes | retry-after header or 60s |
| Server error (500, 502, 503, "overloaded") | ✅ Yes | 30s |
| Auth/billing ("unauthorized", "invalid key") | ❌ No | Manual resume |

For transient errors, GSD pauses briefly and resumes automatically. For permanent errors, configure fallback models:

```yaml
models:
  execution:
    model: claude-sonnet-4-6
    fallbacks:
      - openrouter/minimax/minimax-m2.5
```

**Headless mode:** `gsd headless auto` auto-restarts the entire process on crash (default 3 attempts with exponential backoff). Combined with provider error auto-resume, this enables true overnight unattended execution.

### Budget ceiling reached

**Symptoms:** Auto mode pauses with "Budget ceiling reached."

**Fix:** Increase `budget_ceiling` in preferences, or switch to `budget` token profile to reduce per-unit cost, then resume with `/gsd auto`.

### Stale lock file

**Symptoms:** Auto mode won't start, says another session is running.

**Fix:** GSD automatically detects stale locks — if the owning PID is dead, the lock is cleaned up and re-acquired on the next `/gsd auto`. This includes stranded `.gsd.lock/` directories left by `proper-lockfile` after crashes. If automatic recovery fails, delete `.gsd/auto.lock` and the `.gsd.lock/` directory manually:

```bash
rm -f .gsd/auto.lock
rm -rf "$(dirname .gsd)/.gsd.lock"
```

### Git merge conflicts

**Symptoms:** Worktree merge fails on `.gsd/` files.

**Fix:** GSD auto-resolves conflicts on `.gsd/` runtime files. For content conflicts in code files, the LLM is given an opportunity to resolve them via a fix-merge session. If that fails, manual resolution is needed.

## MCP Client Issues

### `mcp_servers` shows no configured servers

**Symptoms:** `mcp_servers` reports no servers configured.

**Common causes:**
- No `.mcp.json` or `.gsd/mcp.json` file exists in the current project
- The config file is malformed JSON
- The server is configured in a different project directory than the one where you launched GSD

**Fix:**
- Add the server to `.mcp.json` or `.gsd/mcp.json`
- Verify the file parses as JSON
- Re-run `mcp_servers(refresh=true)`

### `mcp_discover` times out

**Symptoms:** `mcp_discover` fails with a timeout.

**Common causes:**
- The server process starts but never completes the MCP handshake
- The configured command points to a script that hangs on startup
- The server is waiting on an unavailable dependency or backend service

**Fix:**
- Run the configured command directly outside GSD and confirm the server actually starts
- Check that any backend URLs or required services are reachable
- For local custom servers, verify the implementation is using an MCP SDK or a correct stdio protocol implementation

### `mcp_discover` reports connection closed

**Symptoms:** `mcp_discover` fails immediately with a connection-closed error.

**Common causes:**
- Wrong executable path
- Wrong script path
- Missing runtime dependency
- The server crashes before responding

**Fix:**
- Verify `command` and `args` paths are correct and absolute
- Run the command manually to catch import/runtime errors
- Check that the configured interpreter or runtime exists on the machine

### `mcp_call` fails because required arguments are missing

**Symptoms:** A discovered MCP tool exists, but calling it fails validation because required fields are missing.

**Common causes:**
- The call shape is wrong
- The target server's tool schema changed
- You're calling a stale server definition or stale branch build

**Fix:**
- Re-run `mcp_discover(server="name")` and confirm the exact required argument names
- Call the tool with `mcp_call(server="name", tool="tool_name", args={...})`
- If you're developing GSD itself, rebuild after schema changes with `npm run build`

### Local stdio server works manually but not in GSD

**Symptoms:** Running the server command manually seems fine, but GSD can't connect.

**Common causes:**
- The server depends on shell state that GSD doesn't inherit
- Relative paths only work from a different working directory
- Required environment variables exist in your shell but not in the MCP config

**Fix:**
- Use absolute paths for `command` and script arguments
- Set required environment variables in the MCP config's `env` block
- If needed, set `cwd` explicitly in the server definition

## Recovery Procedures

### Reset auto mode state

```bash
rm .gsd/auto.lock
rm .gsd/completed-units.json
```

Then `/gsd auto` to restart from current disk state.

### Reset routing history

If adaptive model routing is producing bad results, clear the routing history:

```bash
rm .gsd/routing-history.json
```

### Full state rebuild

```
/gsd doctor
```

Doctor rebuilds `STATE.md` from plan and roadmap files on disk and fixes detected inconsistencies.

## Getting Help

- **GitHub Issues:** [github.com/gsd-build/GSD-2/issues](https://github.com/gsd-build/GSD-2/issues)
- **Dashboard:** `Ctrl+Alt+G` or `/gsd status` for real-time diagnostics
- **Forensics:** `/gsd forensics` for structured post-mortem analysis of auto-mode failures
- **Session logs:** `.gsd/activity/` contains JSONL session dumps for crash forensics

## Windows-Specific Issues

### LSP returns ENOENT on Windows (MSYS2/Git Bash)

**Symptoms:** LSP initialization fails with `ENOENT` or resolves POSIX-style paths like `/c/Users/...` instead of `C:\Users\...`.

**Cause:** The `which` command in MSYS2/Git Bash returns POSIX paths that Node.js `spawn()` can't resolve.

**Fix:** Updated in v2.29+ to use `where.exe` on Windows. Upgrade to the latest version.

### EBUSY errors during WXT/extension builds

**Symptoms:** `EBUSY: resource busy or locked, rmdir .output/chrome-mv3` when building browser extensions.

**Cause:** A Chromium browser has the extension loaded from the build output directory, preventing deletion.

**Fix:** Close the browser extension, or set a different `outDirTemplate` in your WXT config to avoid the locked directory.

## Database Issues

### "GSD database is not available"

**Symptoms:** `gsd_save_decision`, `gsd_update_requirement`, or `gsd_save_summary` fail with this error.

**Cause:** The SQLite database wasn't initialized. This happens in manual `/gsd` sessions (non-auto mode) on versions before v2.29.

**Fix:** Updated in v2.29+ to auto-initialize the database on first tool call. Upgrade to the latest version.

## Verification Issues

### Verification gate fails with shell syntax error

**Symptoms:** `stderr: /bin/sh: 1: Syntax error: "(" unexpected` during verification checks.

**Cause:** A description-like string (e.g., `All 10 checks pass (build, lint)`) was treated as a shell command. This can happen when task plans have `verify:` fields with prose instead of actual commands.

**Fix:** Updated in v2.29+ to filter preference commands through `isLikelyCommand()`. Ensure `verification_commands` in preferences contains only valid shell commands, not descriptions.

## LSP (Language Server Protocol)

### "LSP isn't available in this workspace"

GSD auto-detects language servers based on project files (e.g. `package.json` → TypeScript, `Cargo.toml` → Rust, `go.mod` → Go). If no servers are detected, the agent skips LSP features.

**Check status:**
```
lsp status
```

This shows which servers are active and, if none are found, diagnoses why — including which project markers were detected but which server commands are missing.

**Common fixes:**

| Project type | Install command |
|-------------|-----------------|
| TypeScript/JavaScript | `npm install -g typescript-language-server typescript` |
| Python | `pip install pyright` or `pip install python-lsp-server` |
| Rust | `rustup component add rust-analyzer` |
| Go | `go install golang.org/x/tools/gopls@latest` |

After installing, run `lsp reload` to restart detection without restarting GSD.