Customization

Project Context

Create or update AGENTS.md (created on first run or via /init) with a short project summary:

# Project

## Tech
- Node.js, TypeScript, PostgreSQL

## Standards
- Async/await everywhere
- Tests for new features
- No hardcoded secrets

Edit .opencode/agents/[agent].md to adjust behavior (model, permissions).

Note: older installs may still recognize .opencode/agent/ temporarily, but .opencode/agents/ is canonical.

Add command files to .opencode/commands/ with frontmatter (description, agent). Type /command-name in the TUI to run.

Skills live in .opencode/skills/ and can be loaded on demand:

Example skills: ux-responsive, project-bootstrap, agent-diagnostics, docs-validation
For the canonical current list, use the Skills Matrix

Recommended policy:

Load skills on demand only when the task clearly matches the domain.
Use one relevant skill by default; add a second only for explicit cross-domain work.

Harden each agent with permission.skill allowlists:

permission:
  skill:
    "*": "deny"
    "blogger": "allow"
    "brutal-critic": "allow"

This prevents unrelated skills from being loaded and keeps behavior predictable.

Current distribution is intentionally core-only; no optional skill packs are shipped.

When evaluating potential additions, use this gate: