The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill operates on existing markdown pages to identify and resolve broken citations. There is a risk of indirect prompt injection where malicious text embedded in a brain page or returned by the citation lookup could influence the agent's behavior during the audit or update phase.
Ingestion points: The skill reads local files using list_pages and get_page to analyze content.
Boundary markers: There are no explicit markers or instructions defined in the skill to isolate or distinguish untrusted citation text from the agent's core instructions.
Capability inventory: The skill uses put_page to modify brain pages and executes API calls (via host integration) based on extracted content.
Sanitization: The instructions do not specify any validation or sanitization requirements for the fragments of text extracted and used to query external services.
[COMMAND_EXECUTION]: The skill includes a shell script for batch processing that utilizes common CLI tools to traverse the workspace and filter files. While used for routine operations, the presence of these instructions establishes a local command execution surface.
Evidence: A bash script snippet in the 'Batch mode' section uses find and grep to identify candidate markdown files based on frequency of specific keywords.

citation-fixer