The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides templates for agents that process external inputs such as code diffs and API documentation. It proactively mitigates injection risks by instructing authors to define strict permission boundaries and by including negative constraints in the agent prompts to prevent unauthorized delegation or file modification.
[CREDENTIALS_UNSAFE]: The documentation includes security guidance on avoiding hardcoded credentials. It promotes the use of environment variable substitution ({env:VAR}) or file-based secret loading for agent configurations, ensuring sensitive keys are not persisted in the agent definitions.
[COMMAND_EXECUTION]: The skill uses OpenCode CLI commands for discovery and validation purposes. It enforces the principle of least privilege by providing templates that explicitly disable shell access (bash: deny) for subagents unless specifically required by the user.
[DATA_EXFILTRATION]: Permission templates include fine-grained controls for read access, specifically setting *.env files to ask mode, which ensures the user is prompted before the agent can access potentially sensitive environment configuration.

opencode-agent-creator