The Agent Skills Directory

[SAFE]: The skill defines a robust internal policy framework for 'Write Boundaries', distinguishing between read-only, artifact-only, and state-changing phases. This enforces the principle of least privilege by restricting tool usage based on the current workflow phase.- [SAFE]: It provides specific instructions for handling 'Durable Records' (ADRs) and 'Deferred Findings', using standard project directories (docs/decisions/, docs/reports/findings/) and structured schemas to maintain auditability without exposing sensitive information.- [SAFE]: The guidance for commit messages and changelogs emphasizes the use of 'durable anchors' and 'stable identifiers', explicitly instructing the agent to remove session-local or transient information (like local file paths or tool session IDs) from version history.- [SAFE]: Language selection for chat interactions uses a safe precedence (explicit instructions, then VIBE_CHAT_LANGUAGE environment variable), preventing accidental data leakage through locale inference from source code.

vibe-writing