The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill documentation recommends an installation method using curl -fsSL https://opencode.ai/install | bash, which executes an unverified script from a remote domain directly in the shell.
[EXTERNAL_DOWNLOADS]: Instructions include downloading packages from unverified external sources such as the opencode-ai NPM package and the anomalyco/tap/opencode Homebrew tap.
[PRIVILEGE_ESCALATION]: The CLI includes a --dangerously-skip-permissions flag that allows for the automatic approval of actions, bypassing standard permission checks and reducing user oversight.
[INDIRECT_PROMPT_INJECTION]: The tool ingests untrusted data from multiple sources which could be used for prompt injection.
Ingestion points: Attaching files via -f, fetching pull requests via opencode pr, and importing sessions from URLs.
Boundary markers: The documentation does not specify any delimiters or warnings to ignore embedded instructions in the processed data.
Capability inventory: The tool has access to high-privilege capabilities including shell execution (bash), file modification (edit), and network operations (webfetch).
Sanitization: There is no mention of input sanitization or validation for the data being processed.
[COMMAND_EXECUTION]: The TUI supports direct execution of shell commands using the !command syntax, providing a mechanism for arbitrary command execution.

opencode