The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's workflow relies on executing a local Python script (scripts/search_branium.py) to perform searches within the user's filesystem. While the script is part of the skill package, this pattern involves the agent invoking shell commands with arguments derived from user queries and directory context.
Evidence: SKILL.md provides examples such as python .\scripts\search_branium.py --cwd "..." --query "...".
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection. It is designed to read and present snippets from arbitrary markdown files within an Obsidian vault. If these files contain malicious instructions, they could influence the agent's behavior once ingested into its context.
Ingestion points: The agent reads file content via the read_text function in scripts/search_branium.py and displays snippets or full file contents to the user/agent context.
Boundary markers: The skill does not explicitly define markers to segregate searched note content from instructions.
Capability inventory: The agent has the capability to execute the search script and potentially modify the SKILL.md file itself through 'self-healing' instructions.
Sanitization: The script performs tokenization and basic pattern matching but does not sanitize the content of the notes for potential agent instructions.
[COMMAND_EXECUTION]: The test suite (tests/test_search_branium.py) uses subprocess.run to execute the search script during testing. This is a standard practice for functional testing of CLI tools and is considered benign in this context.
Evidence: tests/test_search_branium.py contains subprocess.run([sys.executable, str(SCRIPT), ...]).

search-branium