permission-manager – skill Lập trình cho agent AI

What I do

• Review and summarize currently always-allowed commands
• Suggest safe read-only commands for auto-approval
• Add or remove commands from the allow list in opencode.json
• Configure skill-level permissions (allow/deny/ask) with wildcard patterns
• Audit permission configs for security and usability

When to Use

Use this when optimizing opencode's permission settings, reviewing allowed commands, or configuring skill access controls.

Workflow Steps

1. Read current config: Load ~/.config/opencode/opencode.json or project-level opencode.json

2. Summarize permissions: Identify currently allowed commands and skill permissions

3. Suggest additions: Propose safe read-only commands for auto-allow (see recommended list below)

4. Apply changes: Edit the config to add/remove permission entries

5. Validate: Ensure JSON is valid after changes

Complements opencode's built-in allow/deny/ask permissions by auditing current config and recommending adjustments through conversation.

Key Rules

• Never allow commands that modify files, commit, push, or change system state
• Prefer exact command entries such as git status --short, git diff --stat, and ls -la
• Avoid trailing wildcards such as git status* unless the expanded command family has been manually reviewed as read-only
• Confirm with user before modifying permission config
• Distinguish between bash command permissions and skill permissions
• Keep config organized: group related commands together

Limitations

• This skill is scoped to opencode permission configuration and should not modify other agent hosts' permission stores.
• Treat all write-capable command permissions as high-risk; review them manually even when a pattern looks narrow.

How to trigger me

Use the Task tool with the permission-manager subagent type:

/permissions

Or in natural language, ask opencode to "manage opencode permissions" or "review allowed commands".