The Agent Skills Directory

[DATA_EXFILTRATION]: Disclosure of a local system username via a hardcoded absolute path. The SKILL.md file contains a link to a safety document located at a specific local file path (file:///Users/fred/.config/opencode/skills/global-rules/rules/bash-safety.md), which reveals the username 'fred' from the author's environment.\n- [PROMPT_INJECTION]: Detection of an indirect prompt injection vulnerability surface. Ingestion points: The skill instructs the agent to download and analyze academic literature (PDFs) and external source code repositories during the research phase. Boundary markers: The instructions do not provide explicit delimiters or warnings to ignore potentially malicious instructions embedded within the ingested external content. Capability inventory: The skill utilizes capabilities for file system modification (mkdir, touch) and bash command execution for installation and testing. Sanitization: No validation or sanitization of the content from external research sources is specified before the data is used to inform architectural decisions.\n- [SAFE]: The skill incorporates commendable safety rules for shell command execution. It explicitly prohibits high-risk patterns such as sudo, curl|bash, and unshielded rm -rf commands, while also mandating the use of timeouts and budget limits for loops.

software-planner