The Agent Skills Directory

[PROMPT_INJECTION]: The mandate defined in assets/mandate.md uses high-urgency markers like <EXTREMELY_IMPORTANT> to instruct the agent to prioritize the patina-mode skill as the primary entry point for all non-trivial engineering tasks, effectively overriding the agent's default task-routing logic.
[COMMAND_EXECUTION]: The skill provides a shell script scripts/install-machinery.sh intended for user execution. This script performs file system manipulations, including directory creation and configuration file modification, using standard shell utilities like awk and cat.
[PERSISTENCE]: The setup process modifies the user's global agent configuration file (~/.claude/CLAUDE.md) to include a reference to the engineering model override sheet. This mechanism ensures that the skill's specific instructions and configurations are loaded automatically in every future session for that user.
[INDIRECT_PROMPT_INJECTION]: The skill establishes a workflow where the agent automatically ingests content from external files, creating a potential surface for instruction injection.
Ingestion points: The global ~/.claude/CLAUDE.md file (via the @ inclusion of engineering-models.md).
Boundary markers: The mandate block in CLAUDE.md is delimited by HTML comments, but the included model override file lacks specific sanitization boundaries.
Capability inventory: The skill possesses file system write capabilities and script execution via the provided bash script.
Sanitization: The skill generates the configuration file based on user prompts but does not perform rigorous validation or escaping of the written content.

setup-engineering