The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user to install the opencode-ai package via the npm registry or a Homebrew tap (anomalyco/tap/opencode), which involves downloading and executing third-party software.\n- [COMMAND_EXECUTION]: The skill makes extensive use of the terminal tool to run CLI commands, including complex shell workflows that use command substitution ($()) and directory traversal to automate coding tasks.\n- [DATA_EXFILTRATION]: The skill facilitates the transmission of local source code, environment configurations, and configuration files (e.g., .env.example, config.yaml) to external AI model providers (such as OpenRouter or Anthropic) for analysis.\n- [INDIRECT_PROMPT_INJECTION]:\n
Ingestion points: The PR Review Workflow described in SKILL.md ingests untrusted code from external git repositories via git clone and git diff.\n
Boundary markers: No delimiters or specific safety instructions are provided to the agent to distinguish between its instructions and the content of the external code being reviewed.\n
Capability inventory: The skill uses terminal and process tools, which grant the agent the ability to modify the file system and access the network based on its analysis of the ingested code.\n
Sanitization: There is no evidence of input validation or sanitization before passing external code to the coding agent.

opencode