The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the codebase and markdown files, which are then used to influence agent behavior through term definitions and conflict detection.
Ingestion points: scripts/migrate.py parses UBIQUITOUS_LANGUAGE.md files; scripts/query.py reads the YAML dictionary; SKILL.md instructs the agent to scan project structure and content.
Boundary markers: Absent. Extracted terms and definitions are interpolated into the session context without protective delimiters or instructions to ignore embedded commands.
Capability inventory: scripts/write.py and scripts/migrate.py perform file system writes; SKILL.md modifies project configuration files (AGENTS.md/CLAUDE.md).
Sanitization: None detected in the provided scripts.
[PERSISTENCE]: The skill includes instructions to modify AGENTS.md or CLAUDE.md to ensure it is automatically invoked in future sessions. While it requests user permission, this establishes a persistent instruction set that overrides default agent behavior for the project workspace.

with-ubiquitous-language