The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill contains instructions for a manual update using the command curl -fsSL https://opencode.dev/install | bash. This pattern involves downloading a script from an untrusted domain and executing it directly in the shell, which is a significant security vulnerability as the remote content can be replaced with malicious code.
[COMMAND_EXECUTION]: The skill relies on the bash tool to execute commands such as opencode acp, opencode session list, and opencode --version. The use of shell commands with potentially unsanitized parameters like file paths poses a risk of command injection.
[EXTERNAL_DOWNLOADS]: The skill fetches version data and software from https://github.com/anomalyco/opencode and https://opencode.dev. These sources are not recognized as trusted vendors or well-known services, increasing the risk associated with the remote code execution pattern.
[METADATA_POISONING]: Significant inconsistencies exist between the SKILL.md frontmatter (version 1.0.2, name 'opencode-acp-control') and the _meta.json file (version 0.1.1, slug 'opencode-acp-control-3'), which may mislead users or automated systems about the skill's identity and provenance.
[INDIRECT_PROMPT_INJECTION]: The skill interacts with external process data via process.poll and presents a potential vulnerability to indirect prompt injection. 1. Ingestion points: Output from process.poll in SKILL.md. 2. Boundary markers: Absent. 3. Capability inventory: bash, process.write, process.kill. 4. Sanitization: Absent.

opencode-acp-control