Your Task

•
Gather context: Find and attach relevant files based on the conversation and "$ARGUMENTS":
- •Always include CLAUDE.md if it exists
- •Add 2-5 most relevant source files
- •Use relative paths when in the same project, absolute paths for cross-project reviews
•
Summarize (2-3 sentences): What's the project/problem? What's the current state?
•
Run OpenCode:

bash

opencode run "MESSAGE" -m MODEL -f FILE1 -f FILE2

Syntax rules: Message MUST come first as a positional arg (quoted string). Flags (-m, -f) come AFTER the message. Do NOT use --prompt.

Example:

bash

opencode run "Context summary here.

Question: $ARGUMENTS

Give me your honest take - be direct and concise." -m opencode/kimi-k2.5-free -f CLAUDE.md -f src/relevant.py

Models

Pick based on task complexity:

Model	Use for
`opencode/minimax-m2.1`	Quick questions, lightweight
`opencode/glm-4.7`	Evals, comparisons, reviews
`opencode/kimi-k2.5-free`	Default — strong agentic model (free tier, temporary)
`opencode/kimi-k2.5`	Same as above, paid — use when free tier ends
`opencode/gpt-5.1-codex-mini`	Everyday tasks (cheap)
`opencode/gpt-5.2-codex`	Bug checking, 2nd opinions on complex tasks and architecture decisions (paid)

Default model is opencode/kimi-k2.5-free. Add -m <model> to use a different one.

Ref: OpenCode Zen models

Examples

•/opencode what do you think of adding caching here?
•/opencode kimi how should I structure this feature?
•/opencode is this the right approach for auth?
•/opencode review this plan as a PM
•/opencode codex complex architecture decision here

Notes

•This skill runs in a forked context (context: fork) to save main conversation space
•Cross-project reviews: use absolute paths (-f /path/to/other/project/file.md)
•The second opinion is returned to the main conversation as a summary