The Agent Skills Directory

[COMMAND_EXECUTION]: The script install-tools.sh uses sudo to perform system-level operations, such as updating package manager lists (apt-get update) and installing software. It also uses administrative privileges to modify the system's repository configuration by writing to /etc/apt/sources.list.d/trivy.list.
[EXTERNAL_DOWNLOADS]: The skill fetches an installation script from the public GitHub repository for Trivy at https://raw.githubusercontent.com/aquasecurity/trivy/main/contrib/install.sh.
[REMOTE_CODE_EXECUTION]: The skill executes the downloaded Trivy installation script by piping its content directly to the shell (sh), which allows remote code to be run with the user's current permissions.
[PROMPT_INJECTION]: The skill ingests untrusted source code and configuration files from the repository to generate security findings, creating a surface for indirect prompt injection if malicious instructions are embedded in the analyzed data.
Ingestion points: Reads all repository files through the various scanning scripts (scan-code.sh, scan-deps.sh, etc.).
Boundary markers: Absent; the findings are included in Markdown reports (.opencode/docs/SECURITY-AUDIT.md) without explicit sanitization or delimiters to prevent the agent from processing instructions hidden in the data.
Capability inventory: Authorized to execute bash scripts, write files to the filesystem, and interact with the GitHub API via the gh CLI.
Sanitization: Findings are processed through jq, but textual fields from findings are interpolated into the final report without filtering for potential instructions.

security-auditor