Make the repository the context system, not the prompt.

OpenCode provides primary and subagents, configurable permissions, and ACP/editor integrations. It fits well when you want plan/build separation around a repository-native workflow.

Recommended pattern

• Use a plan-style agent to read specs and propose work without edits.
• Use a build-style agent to implement once the spec and task path are clear.
• Keep repository rules in AGENTS.md and folder README.md files.
• Restrict write and bash permissions until the reading path is validated.

{
  "$schema": "https://opencode.ai/config.json",
  "agent": {
    "plan": {
      "permission": {
        "edit": "deny",
        "bash": { "*": "ask" }
      }
    },
    "build": {
      "permission": {
        "edit": "allow",
        "bash": { "*": "ask", "git status *": "allow" }
      }
    }
  }
}

MCP + prompts (auto-configured)

// opencode.json — generated automatically by install.py
{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "specnative": {
      "type": "local", "enabled": true,
      "command": ["./.specnative/.venv/bin/python3",
                  "./.specnative/specnative_mcp.py"]
    }
  },
  "prompts": {
    "spec-init":    { "description": "Guided project setup" },
    "spec-update":  { "description": "Detect gaps, refine docs" },
    "spec-status":  { "description": "Quick health check" },
    "spec-handoff": { "description": "Handoff for next agent" }
  }
}

Source: OpenCode agents and permissions docs

Claude Code reads AGENTS.md and supports custom slash commands via .claude/commands/. SpecNative installs four commands automatically so the developer can start with /spec-init on day one.

Recommended pattern

• Use /spec-init to fill context documents the first time.
• Use /spec-status to check active work at the start of each session.
• Use /spec-handoff before ending a session so the next agent can resume.
• Encode domain flows as MCP prompts (start_initiative, plan_tasks).

# Built-in slash commands (from .claude/commands/)
/spec-init     # Interview developer, fill spec-native/ docs
/spec-update   # Detect gaps, guide iterative refinement
/spec-status   # Session + specs + tasks + alerts
/spec-handoff  # Generate SESSION.md handoff for next agent

# MCP setup (auto-installed)
claude mcp add specnative \
  "$(pwd)/.specnative/.venv/bin/python3" \
  "$(pwd)/.specnative/specnative_mcp.py" \
  -- --repo "$(pwd)"

Source: Claude Code overview

Codex reads AGENTS.md and works best when the repository already exposes constraints clearly. SpecNative gives Codex a stable context surface and installs ready-to-use prompts via codex.toml.

Recommended pattern

• Start every session with codex --prompt spec-status to check active work.
• Delegate work by naming the spec or initiative, not by rewriting the context.
• Use resume() MCP tool to continue from the last agent's checkpoint.
• Use codex --prompt spec-handoff before closing to save session state.

# codex.toml — generated automatically by install.py
[mcp_servers.specnative]
command = "./.specnative/.venv/bin/python3"
args    = ["./.specnative/specnative_mcp.py"]
type    = "stdio"

# Ready-to-use prompts:
# codex --prompt spec-init
# codex --prompt spec-update
# codex --prompt spec-status
# codex --prompt spec-handoff

Sources: Codex cloud docs, Codex product overview