The Agent Skills Directory

[SAFE]: The skill consists entirely of Markdown documentation and YAML configuration files. There are no executable scripts (Python, JavaScript, shell) or automated installation procedures provided within the package.
[SAFE]: All external references, such as the links to the author's (MTEnt) 'Graph Engineering' repository and various engineering blogs, are for provenance and educational purposes only. They do not involve remote code execution (RCE) or automated downloads during the skill's operation.
[DATA_EXPOSURE]: The instructions demonstrate a strong security posture by explicitly forbidding the use of production credentials in tests and warning against the export of private data or credentials during operational handoffs.
[INDIRECT_PROMPT_INJECTION]: Although the skill coordinates work based on external data (issues, design records, release manifests), it mitigates injection risks through explicit process controls:
Ingestion points: Reads authoritative records, issues, and design documents (SKILL.md).
Boundary markers: Instructs the agent to 'distinguish an instruction from a control' and 'identify real enforcement' rather than relying on prompt-based instructions.
Capability inventory: The skill itself contains no subprocess, file-write, or network operations; it defers all actions to the host's existing authorized tools.
Sanitization: Focuses on reconcilement with 'authoritative state' and 'existing engineering policies' to prevent unauthorized actions triggered by external data.

ai-native-sdlc