The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script scripts/search_conf_papers.py dynamically modifies the Python module search path using sys.path.insert(0, _START_MY_DAY_SCRIPTS). The path is computed at runtime based on the script's location (os.path.join(os.path.dirname(os.path.dirname(_SCRIPT_DIR)), 'start-my-day', 'scripts')), which is a technique for dynamic loading from computed paths.
[INDIRECT_PROMPT_INJECTION]: The skill represents a surface for indirect prompt injection by processing untrusted data from external sources.
Ingestion points: Fetches paper metadata (titles, authors, abstracts) from DBLP and Semantic Scholar APIs.
Boundary markers: The instructions in SKILL.md for generating markdown notes do not include explicit delimiters or instructions for the agent to ignore potentially malicious content embedded in the academic metadata.
Capability inventory: The skill performs file system writes to the Obsidian vault, network operations to external APIs, and executes local Python scripts.
Sanitization: While it uses regex to sanitize filenames, the paper titles and abstracts are interpolated directly into markdown templates without specific escaping or validation for prompt injection patterns.
[EXTERNAL_DOWNLOADS]: The skill performs network requests to dblp.org and api.semanticscholar.org. These are well-known academic research services used to retrieve legitimate paper information.
[COMMAND_EXECUTION]: The skill's workflow involves executing shell commands to run local Python scripts (search_conf_papers.py, scan_existing_notes.py, link_keywords.py) to process data and manage the Obsidian note index.

conf-papers