opencode-skill-hush

OpenCode plugin that suppresses verbose skill content and command template display in the TUI.

Instead of dumping 100+ lines of <skill_content> into the conversation when an agent loads a skill, it shows a one-line placeholder. The skill content still reaches the LLM context -- only the TUI display is shortened.

Before / After

Skill load -- Before:

Loaded skill: bmad-advanced-elicitation

<skill_content name="bmad-advanced-elicitation">
# Skill: bmad-advanced-elicitation

# Advanced Elicitation
**Goal:** Push the LLM to reconsider, refine, and improve its recent output.
...
Base directory for this skill: /Users/katops/git/.agents/skills/...
</skill_content>

Skill load -- After:

skill: bmad-advanced-elicitation
[Skill "bmad-advanced-elicitation" loaded]

Command template -- Before:

/commit
feat: add new feature (multi-line resolved template...)

Command template -- After:

/commit
[Command: commit]

Installation

Global (recommended)

npm install -g opencode-skill-hush

Or as a local plugin

Clone the repo and reference it directly in opencode.json:

{
  "plugin": ["/path/to/opencode-skill-hush/dist/index.js"]
}

Configuration

Add to your opencode.json:

{
  "plugin": [
    ["opencode-skill-hush", {
      "suppressSkills": true,
      "suppressCommands": true,
      "showLineCount": false
    }]
  ]
}

Options

Option	Type	Default	Description
suppressSkills	boolean	true	Replace skill tool content with placeholder
suppressCommands	boolean	true	Replace command templates with placeholder
showLineCount	boolean	false	Append line count to skill placeholder

Restart opencode after changing config.

Structure

opencode-skill-hush/
├── src/
│   ├── index.ts            # Plugin entry, exports ContentHushPlugin
│   └── hooks/
│       ├── skill.ts        # tool.execute.after handler
│       ├── command.ts      # command.execute.before handler
│       ├── chat.ts         # chat.message handler
│       └── __tests__/      # Unit tests (vitest)
├── package.json
├── tsconfig.json
└── AGENTS.md

How it works

Three hooks, one plugin:

Hook	Target	What it does
tool.execute.after	skill tool	Replaces output.title and output.output with a minimal placeholder. The tool has already finished, so the LLM context is unaffected.
command.execute.before	slash commands	Replaces output.parts with a single [Command: {name}] text part before it renders.
chat.message	TUI chat messages	Detects command templates in chat text parts and replaces them with [Command: {heading}]. Matches if the text contains a <skill_content> tag (regardless of H2 structure), or starts with an H1 heading (#) that has a / prefix and includes H2 sections (##). Minimum 50 chars. Catches templates that arrive via chat.message rather than command.execute.before.

Development

npm install
npm run typecheck
npm test

Contributing

See CONTRIBUTING.md for contribution guidelines and CODE_OF_CONDUCT.md for community standards.

Security

Report vulnerabilities via SECURITY.md.

License

MIT License. See LICENSE.