The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script journal_meta.py dynamically searches for and executes sibling scripts (jabbrv.py and journal_if.py) within user-writable directories such as ~/.claude/skills and ~/.openclaw/skills. This search-and-execute mechanism is used for journal data enrichment but could be exploited if a malicious actor can place files in these search paths.
[COMMAND_EXECUTION]: The tool invokes other Python scripts via subprocess.run. It correctly avoids shell=True and passes arguments as a list, which effectively mitigates common shell injection vulnerabilities when passing journal names or identifiers to external processes.
[EXTERNAL_DOWNLOADS]: The skill connects to well-known academic services (OpenAlex, Crossref, and AbbrevISO) to resolve metadata. These connections are made via standard HTTP requests and include a configured User-Agent. This is consistent with the skill's primary function of academic research.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from academic APIs that could potentially contain malicious instructions intended for the agent.
Ingestion points: openalex_fetch and crossref_fetch functions in journal_meta.py ingest titles and journal names from external API responses.
Boundary markers: None present; the skill does not wrap API-provided text in delimiters or provide specific instructions to the agent to ignore embedded commands in the metadata.
Capability inventory: The skill has the capability to write to stdout (read by the agent) and execute local sibling scripts via subprocess.run.
Sanitization: The skill uses urllib.parse.quote for URL construction and list-based argument passing in subprocess.run to ensure data is handled safely by the operating system.

journal-meta