freeclaude-plugin

name: freeclaude-plugin description: "Use the FreeClaude plugin runtime instead of raw shell commands when delegating coding tasks." metadata: { "openclaw": { "emoji": "⚡", "category": "coding" }

}

FreeClaude Plugin Runtime

Prefer the freeclaude_run tool when you need a coding specialist.

When to Use

• feature work
• debugging
• refactors
• code review
• test generation
• code explanation

Preferred Interfaces

1. freeclaude_run tool — best default for agent delegation
2. FreeClaude MCP tools — good when the MCP server is available
3. /fc ... command — for direct operator use

Tool Guidance

Pass:

• task
• workdir
• mode (code, review, debug, explain, test, refactor)
• optional model
• optional timeout
• optional background=true for long-running runs
• optional sessionKey for session-native continuity inside OpenClaw
• optional resume=false to force a fresh FreeClaude session
• optional resumeSessionId to explicitly resume a known FreeClaude session
• optional retryRunId to retry a stored run

Keep includeMemory enabled unless the task explicitly needs a cold run.

Background Runs

For long-running work, prefer:

1. freeclaude_run(background=true, ...)
2. poll with freeclaude_run_status
3. cancel with freeclaude_run_cancel if needed
4. inspect recent runs with freeclaude_run_list
5. inspect stored session bindings with freeclaude_session_list

Both list tools accept optional filters like status, mode, sessionKey, workdir, query, and limit.

Operator command equivalents:

/fc start ...
/fc poll <runId>
/fc result <runId>
/fc cancel <runId>
/fc retry <runId>
/fc runs
/fc runs --status failed --query retry
/fc sessions
/fc sessions --session team-telegram

Safety

• never point it at ~/.openclaw/ unless the operator explicitly wants that
• always use the real project directory
• prefer specific tasks over vague requests

Modes

• code — single-pass coding (default)
• quest — spec-driven async execution (see QUEST.md)
• ralph — verify→fix loop until score ≥ threshold (see RALPH.md)
• review — read-only audit (Reviewer hat, no file edits)
• debug — diagnostics + minimal targeted fix
• explain — explanation only, no edits
• test — author and/or run tests
• refactor — structure-preserving change, behaviour unchanged

Loaded protocols

The following skill files extend FreeClaude with structured execution protocols:

• RALPH.md — iterative PLAN→CODE→VERIFY→FIX loop with script-based scoring, struggle detection, and context rotation.
• QUEST.md — spec-driven async execution across SPEC→PLAN→CODE→TEST →VALIDATE→REPORT stages with checkpointing and milestone notifications.
• HATS.md — multi-perspective hat system (Planner / Coder / Tester / Reviewer) used within Ralph and Quest to enforce role separation.
• SESSION_TREE.md — branch/fork sessions to explore alternative paths without losing the original checkpoint.