Commands & Skills

Commands

Type /command-name in the OpenCode TUI to run.

Most commands accept an optional argument to scope the task: /command-name [argument] — the argument is passed directly to the target agent.

Command	Argument hint	Purpose
`/api-docs`	`[module, file, or endpoint path]`	Generate API reference docs
`/architecture-decision`	`[decision topic or system name]`	Create an ADR
`/architecture-review`	`[system, component, or design document]`	Review design for quality attributes
`/blog-post`	`[topic or title]`	Write a blog post
`/code-review`	`[file, PR, or scope — blank for current changes]`	Security, perf, and style review
`/content-review`	`[content text, file path, or topic]`	Score and critique content quality
`/create-readme`	`[project name or path]`	Generate or improve a README
`/generate-tests`	`[file, class, or function name]`	Generate targeted tests
`/plan-project`	`[feature, objective, or epic]`	Multi-phase implementation plan
`/execution-loop`	`[task goal or deliverable]`	Bounded verify-and-continue execution loop
`/refactor-plan`	`[target module, file, or scope]`	Safe staged refactor strategy
`/security-audit`	`[scope, file, component, or 'full project']`	Security-focused code review
`/stop-loop`	`[optional reason or scope]`	Halt loop execution and report state
`/1-on-1-prep`	`[person] [context]`	Prepare for a 1-on-1 meeting

Skills

Skills are reusable behaviors loaded on demand. Common utility skills include:

project-bootstrap - Create a minimal AGENTS.md scaffold
agent-diagnostics - Validate agent setup and instruction coverage
docs-validation - Outline docs lint and link checks

For the canonical current skill inventory and agent allowlists, see Skills Matrix.

Using Skills

Skills are defined in .opencode/skills/[skill-name]/SKILL.md. Agents load them via the skill tool when the task matches. To invoke a skill manually, ask an agent to use it:

@codebase Use the project-bootstrap skill to create AGENTS.md
@codebase Run agent-diagnostics to check my setup

Skills cannot be called directly from the TUI — they are invoked through agents.

External Link Validation (Optional)

Internal docs links are validated by default via npm run validate:docs.
External links can be validated on demand via npm run validate:docs:external.
CI runs external checks in a separate scheduled, non-blocking workflow to avoid flaky PR gates.

Command Matrix Parity Validation

Command tables are validated against canonical command frontmatter in .opencode/commands/*.md.
Run npm run validate:commands to verify command presence and metadata parity.

Risk-Scored Review Path (CI)

PR validation includes a risk-path gate that classifies changed files as low/medium/high impact.
Medium/high changes require the PR template Risk level field to acknowledge at least that level.
Risk-path logic runs in CI via validate-risk-path and is covered by local regression tests (npm run doctor).

Changelog Capability Labels

Unreleased changelog bullets should be prefixed with [capability:<label>].
Validate locally with npm run validate:changelog.
CI enforces this via the validate-changelog gate.

Skill Selection Guardrails

Load skills on demand for matching tasks only.
Use one relevant skill by default; add a second only for clear cross-domain need.
If technology/domain is ambiguous, ask for clarification before loading.

Skill Permission Hardening

OpenCode supports skill-level permissions. Use these in agent frontmatter:

permission:
  skill:
    "*": "deny"
    "docs-validation": "allow"
    "agent-diagnostics": "allow"

Pattern notes:

"*": "deny" first, then explicit allows.
Use wildcard patterns for groups (for example, "internal-*": "allow").
Prefer narrow allowlists per agent role for least privilege.

Task Permission Hardening

Control which subagents can be launched via the Task tool:

permission:
  task:
    "*": "deny"
    "explore": "allow"
    "review": "allow"

Pattern notes:

Keep "*": "deny" first, then explicit allows.
Keep allowlists role-specific.
Rules are evaluated in order; last matching rule wins.

Custom Commands

Add command files to .opencode/commands/ with frontmatter:

---
description: What this command does
agent: recommended-agent
subtask: true
---

See OpenCode Commands docs for full reference.

Next Steps

Agents
Customization