Support opencode

[StudioLE]

Evaluate permissions for OpenCode tool calls

OpenCode has no external binary hook protocol (see anomalyco/opencode#12472)

Instead it offers JS/TypeScript plugins that intercept tool.execute.before events (see https://opencode.ai/docs/plugins/#tool-events)

The plugin receives raw JS objects and can block a tool call by throwing an error. (See: https://opencode.ai/docs/plugins/#env-protection)

Proposal

Two components: an ultra-thin plugin and the Rust binary.

Plugin (~15 lines of JS)

• Receives tool.execute.before with raw JS output.args
• Stringifies them to JSON
• Spawns hook-rs <tool> --opencode and pipes the JSON to stdin
• Parses the JSON response
• Throws if permissionDecision === "deny"
• Returns silently for allow or ask (falls through to OpenCode's native prompt)

Rust binary

• New --opencode flag on each subcommand (bash, read, glob, grep)
• When --opencode is present, deserialise OpenCode's native arg shapes directly
• Map internally to the existing rule evaluation pipeline
• Return standard JSON output with permissionDecision and permissionDecisionReason

Design principle

The plugin knows nothing about schemas. All field mapping, validation, and decision logic lives in Rust. The plugin is a pure JSON pipe.