The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted markdown and canvas notes from the Obsidian vault, which serves as a potential vector for indirect prompt injection.\n- Ingestion points: Files within the Research/{project-slug}/ directory are parsed by scripts such as kb_common.py, kb_link_check.py, and kb_lint.py to update registries and indexes.\n- Boundary markers: The instructions do not define explicit delimiters or include safety prompts to ignore embedded instructions in the notes being processed.\n- Capability inventory: The skill scripts perform file system operations including directory deletion (shutil.rmtree in project_kb.py), moving files (shutil.move), and script execution (subprocess.check_output).\n- Sanitization: While the skill normalizes paths and slugs, it does not sanitize the text content of notes before ingestion by the AI agent.\n- [COMMAND_EXECUTION]: The skill uses subprocess calls to execute local commands and orchestrate helper scripts.\n- Git Integration: The find_repo_root function in kb_common.py calls git rev-parse --show-toplevel.\n- Script Orchestration: The kb_lint.py script uses subprocess.check_output to run additional Python scripts located in the same directory (e.g., kb_registry_check.py).\n- Implementation: Commands use absolute paths to scripts within the skill's directory and do not interpolate unsanitized user input into the shell command strings.

obsidian-project-kb-core