The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands using the opencode CLI. It uses variables and positional arguments for user-provided prompts, which could lead to command injection if the input contains shell metacharacters. The documentation suggests using heredocs to mitigate this.
[PROMPT_INJECTION]: The skill facilitates the transfer of instructions to an external CLI tool. Ingestion points: The <prompt> and <follow-up> arguments in shell command templates in SKILL.md. Boundary markers: Double quotes are used around prompts, and the skill advises using single-quoted heredocs for complex inputs. Capability inventory: The target tool (opencode) has significant capabilities, including file system modification, repository editing, and session tracking. Sanitization: The skill relies on an external 'workspace authority contract' for permissioning and does not provide built-in input sanitization or validation. Additionally, the skill describes an --auto flag that bypasses permission prompts, which could be exploited by malicious instructions to perform unauthorized changes.

opencode-cli-runtime