fix(opencode): expose skills in TUI slash command autocomplete - #19
Conversation
Skills were returned with source='skill' from GET /command, but the OpenCode TUI autocomplete explicitly filters out source==='skill' commands (autocomplete.tsx:363). Native OpenCode's oh-my-openagent plugin injects skills with source='command', bypassing the filter. Changes: - Change skill command source from 'skill' to 'command' in GET /command endpoint (agent_routes.py) to match native OpenCode behavior - Expand Command model to match native SDK Command.Info type: add agent, model, subtask, hints fields; make description and source optional - Add _extract_hints() helper to extract / placeholders from skill templates, matching native Command.hints() utility - Fix server.py CommandStore update: replace broken add_commands() call with register_command(cmd, replace=True) loop - Add fallback skill lookup via pool.skill_commands in session_routes when CommandStore misses dynamically-added skills
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request updates the Command model to align with the OpenCode SDK, adding fields for agent, model, subtask, and hints. It introduces a utility to extract input hints from command templates and implements a fallback mechanism in the execution route for skills not yet synchronized with the CommandStore. Feedback includes a suggestion to use numerical sorting for numbered hints to avoid lexicographical ordering issues and a recommendation to handle the 'skill:' prefix in the new fallback logic to ensure consistency with the execution handler.
| # Fallback: check pool.skill_commands directly when CommandStore misses | ||
| # This handles cases where skills were registered after CommandStore init | ||
| # or where the CommandStore sync callback hasn't fired yet | ||
| if state.pool.skill_commands and request.command in state.pool.skill_commands: |
There was a problem hiding this comment.
The fallback check does not account for the skill: prefix, which is supported by _execute_skill_command. If a user manually enters a command with the prefix and it hasn't been synced to the CommandStore yet, this fallback will fail to identify it as a skill command. Normalizing the command name ensures consistency with the execution logic.
cmd_name = request.command.removeprefix("skill:")
if state.pool.skill_commands and cmd_name in state.pool.skill_commands:Lexicographical sort produces $1, $10, $2 for double-digit placeholders; numeric sort yields the expected $1, $2, $10.
|
/gemini review |
There was a problem hiding this comment.
Code Review
This pull request aligns the Command model with the OpenCode SDK by adding fields for agents, models, subtasks, and input hints. It introduces a hint extraction utility and a fallback mechanism in the command execution route to handle skills that might not be present in the CommandStore. The review feedback recommends making the hint extraction more robust against null inputs and ensuring the command execution fallback correctly handles prefixed command names for consistency.
| # Fallback: check pool.skill_commands directly when CommandStore misses | ||
| # This handles cases where skills were registered after CommandStore init | ||
| # or where the CommandStore sync callback hasn't fired yet | ||
| if state.pool.skill_commands and request.command in state.pool.skill_commands: |
There was a problem hiding this comment.
The fallback check for skill_commands should handle the skill: prefix for consistency with how commands are resolved in _execute_skill_command. If a user manually enters a command with the prefix, this check would currently fail even if the skill exists in the registry.
cmd_name = request.command.removeprefix("skill:")
if state.pool.skill_commands and cmd_name in state.pool.skill_commands:…lify with extend Accept Gemini Code Assist review suggestion: add early-return guard for None/empty template (defensive against provider bugs returning None) and replace for-append loop with extend for conciseness. Add test for None input case.
1 participant
Skills were returned with source='skill' from GET /command, but the OpenCode TUI autocomplete explicitly filters out source==='skill' commands (autocomplete.tsx:363). Native OpenCode's oh-my-openagent plugin injects skills with source='command', bypassing the filter.
Changes: