The Agent Skills Directory

REMOTE_CODE_EXECUTION (MEDIUM): The skill is designed to generate files (.opencode/commands/*.md) that support the execution of arbitrary shell commands using the !command syntax. This capability allows the resulting commands to perform any action the user's shell can, including file modification or network access.
Evidence: The workflow in SKILL.md and documentation in references/opencode-commands.md explicitly describe and encourage using !command (e.g., !git diff --cached, !npm test).
COMMAND_EXECUTION (MEDIUM): The skill enables the creation of persistent automated scripts. While intended for benign automation (e.g., testing, reviewing), there are no safeguards preventing the creation of malicious commands.
Evidence: Templates in assets/opencode-command-template.md provide the structure for embedding executable logic.
PROMPT_INJECTION (LOW): The skill is vulnerable to Indirect Prompt Injection (Category 8) because it ingests untrusted user input and writes it into a high-privilege execution context.
Ingestion points: User-provided 'Prompt Body' described in SKILL.md (Step 2).
Boundary markers: Absent. The user input is directly written to markdown files without escaping or delimiters to prevent command injection within the prompt.
Capability inventory: The skill itself has file-write capabilities. The environment that consumes its output (OpenCode) has full shell execution capabilities via the !command syntax.
Sanitization: Absent. There is no validation or filtering of the 'Prompt Body' to detect or block dangerous shell commands.

reusable-commands