---
description: Ensure every Claude skill has a corresponding OpenCode skill alias
allowed-tools: [Read, Write, Glob, Grep, Bash]
---

# Sync Claude skills to OpenCode

Ensure that every Claude command has a corresponding OpenCode skill and command that delegates to it.

## Steps

1. **List all Claude commands** — Glob `.claude/commands/*.md` (excluding `opencode-sync.md` itself) and `.claude/commands/includes/*.md`.

2. **List existing OpenCode skills** — Glob `.opencode/skills/*/SKILL.md` and `.opencode/commands/*.md`.

3. **For each Claude command**, determine the OpenCode skill name:
   - For `.claude/commands/<name>.md` → skill name is `<name>`
   - For `.claude/commands/includes/<name>.md` → skill name is `recipe-<name>` (following the existing convention, e.g. `guidelines` → `recipe-guidelines`)

4. **Read each Claude command's YAML frontmatter** to extract:
   - `description` — reuse as the OpenCode description
   - `argument-hint` — use to generate the argument context line

5. **For each missing OpenCode skill**, create `.opencode/skills/<skill-name>/SKILL.md`:

```markdown
---
name: <skill-name>
description: <description from Claude command>
---
Read and follow the full instructions in `.claude/commands/<path-to-claude-command>.md`.
<argument context line, if the Claude command has an argument-hint>
```

The argument context line should describe what arguments are expected based on the `argument-hint`:
- If `argument-hint` contains `recipe-name` → `The recipe name will be provided by the user or calling agent.`
- If `argument-hint` contains `b1cc|t1cc` → `The instance name will be provided by the user or calling agent.`
- If there is an `argument-hint` but no specific pattern matched → `Arguments will be provided by the user or calling agent.`
- If there is no `argument-hint` → omit the argument context line entirely

6. **For each missing OpenCode command**, create `.opencode/commands/<skill-name>.md`:

```markdown
---
description: <description from Claude command>
---
Load the `<skill-name>` skill and execute it for $ARGUMENTS.
```

7. **Detect orphaned OpenCode skills** — For each existing OpenCode skill/command, check whether it references a Claude command (by reading the `SKILL.md` body for the `.claude/commands/...` path). If the referenced Claude command no longer exists:
   - Check if the content/description closely matches a current Claude command that lacks an OpenCode equivalent (i.e., it was likely renamed).
   - If a rename is detected: delete the old OpenCode skill directory and command file, then create new ones under the correct name.
   - If no rename match is found: delete the orphaned OpenCode skill directory and command file (it references a dead Claude command).

8. **Update stale descriptions** — For each existing OpenCode skill/command that maps to a valid Claude command, read the Claude command's current `description`. If the OpenCode description differs, update it to match.

9. **Report results** — Print a summary of:
   - Skills created (new)
   - Skills renamed (old name → new name)
   - Skills deleted (orphaned, no rename match)
   - Skills updated (description changed)
   - Skills unchanged (already in sync)

## Important

- Every Claude command must have exactly one corresponding OpenCode skill — no exceptions, including this `opencode-sync` command itself.
- The includes (`guidelines.md`, `logging.md`) use the `recipe-<name>` prefix convention for their OpenCode skill names.