/help

description: "Show Forge usage guide"

Forge Help

Display the following help text to the user:

Commands

**`/forge brainstorm [topic]`** — Turn an idea into concrete specs

• `--from-code` — Generate specs from existing codebase
• `--from-docs path/` — Generate specs from PRDs, API docs, research files

**`/forge plan`** — Decompose specs into task frontiers

• `--filter <name>` — Only plan a specific spec
• `--depth quick|standard|thorough` — Set task granularity

**`/forge execute`** — Run the autonomous implementation loop

• `--autonomy full|gated|supervised` — Set pause behavior
• `--max-iterations N` — Safety cap on loop iterations
• `--token-budget N` — Max tokens to spend
• `--depth quick|standard|thorough` — Override task depth

**`/forge resume`** — Continue after context reset or interruption

**`/forge backprop "bug description"`** — Trace a bug back to a spec gap

• `--from-test path/` — Trace from a failing test

**`/forge status`** — Show current progress and budget

**`/forge review-branch`** — Review an unmerged branch before merging

• `--base main` — Base branch to diff against
• `--spec path/` — Spec file for acceptance criteria verification
• `--depth quick|standard|thorough` — Review depth
• `--fix` — Auto-fix critical issues
• `--comment` — Post results as GitHub PR comment

**`/forge setup-tools`** — Detect and install CLI tools that enhance Forge

**`/forge help`** — Show this help text

Quick Start

1. `/forge brainstorm "build a REST API for task management"` 2. `/forge plan` 3. `/forge execute --autonomy gated`

Configuration

Edit `.forge/config.json` to customize autonomy, depth, token budget, multi-repo setup, and circuit breaker thresholds.

Token Reduction Kill Switch

Set `FORGE_TOKEN_OPT=0` to disable Wave 2 cache extensions and revert to the original 6-pattern, flat-120s-TTL behavior. All other Wave 2 additions (mtime keying, HEAD pinning, stats logging) also disable. Default is on.

The same `FORGE_TOKEN_OPT=0` flag also disables the Wave 3 token-reduction surfaces: the broader output filter, the router's per-phase `effort`/`max_tokens` policy, and the first-run wizard banner. Wave 1 instrumentation extras (capture, `usage_actual.*` ledger keys) bypass too, leaving the v1-shape ledger. This is the single end-to-end rollback contract across all three waves; the wizard prints the same flag on first run.