The Agent Skills Directory

[DATA_EXFILTRATION]: The skill accesses and modifies the local configuration file at ~/.config/opencode/opencode.jsonc (and Windows equivalents). While this is the intended purpose, these files often contain sensitive provider credentials or API keys.
[DYNAMIC_EXECUTION]: A JavaScript snippet is provided in the SKILL.md for the agent to execute. It performs file reading via fs.readFileSync and manual string manipulation to strip comments from JSONC before parsing. This constitutes dynamic script execution during the validation phase.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through its data ingestion process.
Ingestion points: Fetches model lists from {baseURL}/models where baseURL is read from user configuration.
Boundary markers: None identified in the workflow to separate remote API content from the local configuration structure.
Capability inventory: File read/write capabilities (fs) and network access (GET) are present.
Sanitization: No explicit sanitization or validation of the fetched model IDs or metadata is described before they are written to the configuration file.

opencode-models