The Agent Skills Directory

[SAFE]: A thorough review of the skill's instructions and referenced templates confirmed that the operations are confined to the local project environment. The skill uses standard markdown templates and configuration files, with no evidence of obfuscation or unauthorized network access.
[COMMAND_EXECUTION]: The skill includes functionality to stage and commit ADR files using the Bash tool. This is a legitimate feature for document versioning and is explicitly gated by user confirmation in the post-creation steps.
[PROMPT_INJECTION]: The skill demonstrates a surface for indirect prompt injection because it reads content from existing ADRs and project documentation to inform its generation process.
Ingestion points: Existing ADR files in project directories (e.g., docs/adr/*.md) and documentation sources defined in config.yaml.
Boundary markers: The instructions do not define specific delimiters to wrap ingested content.
Capability inventory: The agent possesses Write and Edit tools for file management, alongside Bash for repository commits.
Sanitization: Ingested data is processed via structured markdown templates, which limits the potential for external instructions to alter the final document's structure.

adr