The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes the opencode2 command-line tool to interact with the Sessions API. This interaction is mediated by a Python wrapper (opencode_sessions.py) that uses subprocess.run with hardened argument handling.
[DATA_EXPOSURE]: The skill intentionally reads session metadata and message content from other OpenCode chats to enable coordination. However, it implements filtering to only expose text message parts, excluding reasoning bodies, tool arguments, and internal skill text to minimize data leakage.
[REMOTE_CODE_EXECUTION]: While the skill interacts with an external CLI (opencode2), all command arguments are properly quoted or handled as list items in subprocess.run, preventing shell injection vulnerabilities. The skill uses a machine-readable onboarding contract (references/onboarding.json) to verify the presence of required local tools.
[SAFE]: The skill implements a robust 'preview-then-apply' pattern for sending messages. It generates a unique message_id, creates a local reservation file with a SHA-256 hash of the message text, and requires the user to explicitly apply the message. The wrapper ensures that messages cannot be retried or modified after an uncertain outcome, requiring explicit reconciliation instead.

x9-opencode-sessions