The Agent Skills Directory

[SAFE]: The skill is designed for development workflow assistance and does not contain any malicious patterns. It lacks any indicators of credential exfiltration, remote code execution, or persistence mechanisms. The use of local directories like vault/ and scratch/ for project memory and evidence storage is a common and safe practice in development-focused AI environments.
[INDIRECT_PROMPT_INJECTION]: The skill analyzes external content such as code diffs and task plans, which presents a surface for potential indirect prompt injection if those sources contain adversarial instructions. However, the risk is negligible as the skill's logic is constrained to documentation generation and recommending local verification steps.
Ingestion points: Code diffs and project files located in vault/plans/ and vault/index.md.
Boundary markers: The instructions do not specify any markers or delimiters to separate the ingested data from the agent's instructions.
Capability inventory: The skill identifies and recommends standard local test commands (e.g., yarn jest), uses existing skills for screenshots, and writes evidence files to a local scratch/ directory.
Sanitization: No sanitization or validation of the input diffs or vault content is defined.

test-plan