The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes git diff --name-only to programmatically identify files changed during a session, helping populate the session log with accurate technical history.
[SAFE]: Implements a strict 'User Confirms Before Write' rule (Rule 2), ensuring the agent never modifies or creates files without the user reviewing and approving the gathered session data.
[SAFE]: Enforces a 'No Silent Defaults' policy (Rule 7) for directory resolution, requiring an explicit user selection via the AskUserQuestion tool before determining where brain documents are stored.
[SAFE]: The skill operates entirely within the local filesystem and uses standard tools (Read, Write, Edit) without requesting network access or downloading external code.
[SAFE]: Analyzed the attack surface for indirect prompt injection (Category 8). The skill reads external markdown files (ingestion point) and possesses write capabilities (capability). This is effectively mitigated by robust boundary markers (markdown headings) and a mandatory confirmation gate for all write operations.

project-brain