lopi-cli

# lopi CLI lopi runs Claude Code agents concurrently, each in a git-isolated branch, with retry loops and persistent SQLite memory. This skill documents the CLI surface as it ships today — no `;` composer grammar (that's separate, unshipped work). If the `lopi_*` MCP tools are available in this session (via the lopi plugin's bundled MCP server), prefer them over shelling out to the `lopi` binary for task submission and status checks — they give structured JSON directly instead of parsed CLI text. This skill's command reference is what to fall back on when the MCP tools aren't available, or when the user explicitly asks to run a command. ## Submitting a task ``` lopi run --goal "<goal>" --repo <path> ``` `--repo` defaults to `.`. A good goal string is a single, concrete, scoped instruction — the same granularity you'd give a human PR reviewer, not a multi-part epic: - Good: `"fix the failing test in src/foo.rs"`, `"add input validation to the /api/tasks POST handler"` - Too broad: `"improve the codebase"`, `"add a bunch of features to the dashboard"` `lopi run` streams progress to stdout as the agent moves through its pipeline, one line per phase transition, e.g.: ``` 🚢 lopi run goal: fix the failing test in src/foo.rs repo: . task id: a1b2c3d4… use `lopi watch` in another terminal for the TUI [1] → 📋 planning [1] → 🔨 implementing [1] → 🧪 testing [1] → 📊 scoring ⚓ success ✅ branch=lopi/a1b2c3d4/1, pr=https://github.com/… ``` On failure the ...