The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes project files and external code context to provide implementation patterns. This creates an attack surface where instructions hidden in the project code could influence the agent's behavior.
Ingestion points: The search.py script reads project source files and user-provided task JSON files (SKILL.md, docs/search.md).
Boundary markers: There are no explicit delimiters or instructions provided to separate the ingested project data from the agent's own command logic.
Capability inventory: The skill has the ability to execute shell commands via uv run and write to the local file system (creating session and task files).
Sanitization: No sanitization or filtering of external content is mentioned in the instructions.
[COMMAND_EXECUTION]: The skill relies on the agent executing local Python scripts to perform search and read operations on the pattern catalog. (Evidence: SKILL.md shell command blocks using uv run).
[EXTERNAL_DOWNLOADS]: The search tool automatically downloads a large binary embedding model from a remote registry (registry.ollama.ai) on first use. (Evidence: docs/search.md).
[DYNAMIC_EXECUTION]: The search functionality involves loading and running a binary model file using llama-cpp-python at runtime to rank candidates. (Evidence: docs/search.md).

laravel-project-patterns