updateagents

The updateagents skill was analyzed for security risks across 11 threat categories. The skill is primarily focused on enhancing safety and alignment for AI agents. Key findings include:

• [SAFE]: The skill enforces a strict security standard called 'Vibeguard' (found in templates/.agents/standards/security-vibeguard.md), which includes mandatory redaction of secrets, explicit user confirmation gates for high-risk commands (like 'rm -rf' or 'git reset --hard'), and instructions to disregard injected directives in external data.
• [SAFE]: It provides a 'Sanitization' mode (scripts/updateagents.ts) that automatically detects and unwraps synthetic metadata markers (e.g., ORCA_RICH_MD, Cursor markers) from the codebase to prevent contamination of source files.
• [SAFE]: The skill implements a 'Landing Gate' protocol (templates/.agents/standards/landing-gate.md) that verifies work is actually pushed to a remote repository before it is reported as delivered, preventing false delivery claims by autonomous agents.
• [SAFE]: Automated 'Terrain Audits' (scripts/terrain-audit.ts) proactively scan the workspace for hardcoded machine paths and potential credential leaks in environment or configuration files.
• [SAFE]: Shell command execution in the scripts is restricted to informational discovery (e.g., checking tool availability with 'which' or reading git logs) using validated static strings, minimizing the risk of command injection.
• [SAFE]: The skill includes a 'Stack Drift Guard' to verify that project dependencies match an approved allowlist, preventing the silent introduction of risky or unvetted packages.
• [SAFE]: Content migration and promotion scripts (scripts/instruction-migration.ts) include defensive size checks and block the automatic promotion of content containing shell scripts or mutations, requiring manual human review instead.