The Agent Skills Directory

[SAFE]: The skill provides guidelines for 'Standalone Commits' and 'Ordered Waves', ensuring each commit is a coherent unit that can be independently audited. This methodology significantly improves the security posture of a project by making code history transparent and easier to inspect for anomalies.
[INDIRECT_PROMPT_INJECTION]: The skill involves the agent processing git diffs, which are external untrusted data sources. 1. Ingestion points: The agent reads git diff and git diff --staged output as described in SKILL.md. 2. Boundary markers: The 'Acceptance Checks' table (Reviewable, Buildable, Complete, Focused, Auditable, Revertible) serves as a logical boundary for the agent to evaluate the staged changes. 3. Capability inventory: The skill assumes access to git for version control and shell commands for verification (e.g., tsc). 4. Sanitization: Verification is performed via logical checks against the acceptance criteria to ensure the changes are focused and auditable.

standalone-commits