The Agent Skills Directory

Path Traversal Protection: The skill employs a centralized utility (scripts/_utils.py) to validate all file paths. It uses a safe_resolve function that rejects directory traversal segments (e.g., ..) and ensures that file operations are confined to permitted directories. This effectively mitigates the risk of unauthorized file access or overwrites.
Sensitive Content Scanning: A dedicated script (scripts/scan_sensitive_content.py) is used to scan ADR drafts for PII (Personally Identifiable Information), credentials, and internal URLs before they are written to disk or handed off. This proactively prevents the accidental disclosure of sensitive information.
Safe Data Handling: The skill uses yaml.safe_load for parsing configuration and frontmatter files, protecting against unsafe object instantiation. Additionally, template rendering is performed using a controlled regex-based substitution method rather than potentially unsafe evaluation engines.
Indirect Prompt Injection Surface: The skill processes external data, which represents a theoretical surface for indirect instructions to be processed by an AI agent.
Ingestion points: Data enters the context via scripts/render_template.py (through field arguments) and scripts/normalize_template.py (through user-supplied template files).
Boundary markers: While templates use placeholders, the skill does not currently apply explicit delimiters around interpolated content in the final markdown output.
Capability inventory: The skill can perform filesystem operations, specifically writing and mutating markdown and configuration files.
Sanitization: The potential risk is significantly mitigated by the inclusion of scripts/scan_sensitive_content.py, which performs deterministic scanning for disclosure risks before files are committed to disk.

adr-author