The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructions frequently demonstrate piping the output of curl requests to remote API endpoints directly into python3. While the Python modules and scripts are provided within the skill, this pattern executes an interpreter on data fetched over the network.
Evidence: curl -s "https://api.openalex.org/works?search=transformer+architecture&per_page=5&mailto=agent@kortix.ai" | python3 -m json.tool in SKILL.md.
[DYNAMIC_EXECUTION]: The skill utilizes python3 -c to execute inline Python scripts for complex data reconstruction tasks, such as rebuilding abstracts from inverted indexes. This involves runtime execution of logic on external JSON payloads.
Evidence: The 'Reconstructing Abstracts' and 'Saving Results to Disk' sections in SKILL.md provide Python snippets designed to process sys.stdin from piped curl commands.
[INDIRECT_PROMPT_INJECTION]: The skill is an ingestion surface for 240M+ scholarly works. Since academic metadata (titles, abstracts) can be influenced by external authors, and the skill processes this data without sanitization or boundary markers, it is vulnerable to indirect prompt injection.
Ingestion points: OpenAlex API responses fetched via curl (multiple locations in SKILL.md).
Boundary markers: Absent; data is processed and printed/stored directly.
Capability inventory: Shell command execution (curl, bash), file system writes, and Python script execution.
Sanitization: Absent; the provided processing scripts perform data transformation but do not filter for malicious instructions.
[COMMAND_EXECUTION]: The skill relies heavily on shell-based workflows, including curl for network operations and redirection operators (>) for writing research data to the local disk.

openalex-paper-search