Skip to content

Workflow Slash Commands

OpenCode Multi-Agent ships with a suite of high-leverage workflow slash commands.


Command Overview

CommandCategoryWhat it does
/prd <topic>SDD LifecycleScaffold & draft Product Requirements Document in docs/prd/
/adr [new|supersede|tree|check|migrate|mode]ArchitectureArchitecture Decision Record management: automated drafting, supersede lifecycle, DAG graph, link audits, bidirectional migrations & hierarchy mode switches
/plan <topic>SDD LifecycleScaffold & draft phased Implementation Plan in docs/plan/ with automatic PRD & ADR linking
/impl [task]SDD LifecycleExecute test-driven code implementation & verification adhering to specifications
/sdd [status|handoff|help]SDD LifecycleSpecification-Driven Development lifecycle navigator & session handoff (/sdd handoff)
/grill-me <topic>BrainstormingSocratic interview that rigorously pressure-tests a plan or design
/quick-dev <task>Dev LoopQuick-Dev Zero-Review Fast Track: Ultra-fast Flash model coding + dynamic domain persona injection (instant delivery, alias /flash-dev, see Four-Tier Dev Loops)
/fast-dev <task> [--max-rounds=N]Dev LoopFast-Dev Agile Single-Review Loop: High-velocity Flash model coding (dynamic domain persona) + flagship review (see Four-Tier Dev Loops)
/deep-dev <task> [--max-rounds=N]Dev LoopDeep-Dev Mission-Critical Dual-Review Loop: Flash coding + dual flagship review + Advisor consensus arbitration with full-stack multi-stage staging (see Four-Tier Dev Loops)
/ultra-dev <objective> [--max-rounds=N] [--max-phases=N]Dev LoopUltra-Dev Autonomous Multi-Phase Track: End-to-end autonomous execution — decomposes large objectives into phases, each with its own /deep-dev cycle, context compaction, git-commit isolation, and --resume support (see Four-Tier Dev Loops)
/review-fix-loop [scope] [--max-rounds=N]Quality LoopAutomated review-verify-fix-re-review loop until zero P0/P1 issues. Scope: last commit, HEAD~N, branch, PR, or uncommitted changes
/grill-improve-loop [subject] [--max-rounds=N] [--target=N]Score LoopScore-driven improvement loop: score → analyze → fix/refactor → verify → re-score until structural ceiling, stall, or max rounds. Triggers verification-honesty scoring (Rules 5–7) every round
/goal [text]Goal ExecutionStructured goal execution protocol with audit-friendly checklists and mechanically checkable stop conditions
/handoff [focus]Session StateCompacts current session state into a git-safe handoff bundle (.opencode/handoffs/) and outputs a paste-ready opener for a fresh session
/adr-guard [on|off|status]Quality GateProject-level ADR commit gate: enforces architecture decision records on feat: and refactor: commits
/e2e-guard [on|off|status]Quality GateProject-level E2E testing gate: requires end-to-end coverage verification on features and bug fixes
/env-guard [on|off|status]Security GateProject-level secret leak prevention: blocks reading or leaking .env files to external tools
/deepseek-anchor [on|off|status]Model EngineDeepSeek V4/Pro reasoning depth anchor: prevents reasoning degradation and gates tools during deliberation
/auto-advisor [off|lite|full]IntelligenceToggle auto-advisor mode (off, lite recommendations, full factual auto-answers)
/md-to-pdf <file.md> [output.pdf]PublishingExport Markdown to high-res A4 PDFs with 300 DPI Mermaid diagrams, CSS themes & --doctor diagnostics
/md-to-docx <file.md> [output.docx]PublishingExport Markdown to publication-grade Word (.docx) with pure TS engine, dual fonts & Mermaid rendering
/project [init|index|sync]Project SetupScaffold project baseline files (.opencode/opencode.jsonc etc.) and trigger CodeGraph / GitNexus indexing
/project-wizardTUI WizardInteractive project configuration wizard: toggle MCP services and plugins via visual terminal UI
/profileTUI WizardOpen model profile picker: easily switch or customize Auto / Ultimate / Performance / Economy / Lightweight tiers
/providerTUI WizardOpen provider wizard: configure credentials (baseURL / apiKey) and manage model catalogs
/queuedTUI WizardInteractive TUI dialog to inspect, edit, or cancel queued messages submitted while the agent was busy

Example: review-fix-loop

> /review-fix-loop last commit
  → @code-review finds P0/P1 issues
  → Verifies each finding (reads code, traces data flow, checks upstream guards)
  → If false positive → skipped only after @advisor confirms
  → If confirmed BUG → @<domain-dev> fixes each verified issue
  → @code-review re-reviews
  → Repeats until clean or max rounds reached (default: 5)
  → Summary output: verdict + stats

> /review-fix-loop HEAD~3 --max-rounds=8
  → Same flow, up to 8 rounds (good for larger diffs)

Released under the MIT License.