The Agent Skills Directory

[SAFE]: The skill is implemented as a pure Markdown instruction set and does not include any executable scripts, binaries, or hidden configuration files that could run on the user's system.
[COMMAND_EXECUTION]: The skill directs the agent to use standard local development tools such as git and gh (GitHub CLI) to analyze project structure and manage branches. These commands are used as intended for project organization and do not involve arbitrary command injection or suspicious execution patterns.
[DATA_EXFILTRATION]: No patterns of sensitive file access (such as credentials, environment files, or SSH keys) or unauthorized network transmissions were detected. Data handling is restricted to local file reads for context and local file writes for plan generation.
[PROMPT_INJECTION]: The skill includes no instructions to bypass safety guidelines or override system constraints. It proactively mitigates indirect prompt injection risks by implementing an 'adversarial review' step, where a high-tier model validates generated plans against a checklist of anti-patterns and correctness markers before finalization.

blueprint