The Agent Skills Directory

[COMMAND_EXECUTION]: The skill requests permission to use Bash(rm*), which allows for file and directory deletion. This directly contradicts the stated 'Read-only' purpose of the skill in SKILL.md.
[REMOTE_CODE_EXECUTION]: The skill executes Python code via uv run that depends on external or local scripts not provided in the skill package, specifically .scripts/openalex/openalex_client.py and the cli_council module in packages/cli-council. This represents execution of unverifiable code.
[PROMPT_INJECTION]: The skill has an Indirect Prompt Injection surface as it ingests content from .tex and .bib files and passes this data to sub-agents or external processes without specified sanitization.
Ingestion points: LaTeX and BibTeX source files.
Boundary markers: No specific delimiters or safety instructions are defined for passing this content to sub-agents in 'Deep Verification Mode'.
Capability inventory: The skill possesses Write, Bash, and Task capabilities which could be abused if the agent follows instructions embedded in the files.
Sanitization: There is no evidence of sanitization for the citations or metadata extracted from files before they are used in commands or prompts.

bib-validate