The Agent Skills Directory

[DATA_EXFILTRATION]: The collect_plan_evidence.py script implements strict validation for file access, including file size limits, extension filtering, and prevention of symbolic link resolution or remote URL fetching.
[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to treat all ingested document content as untrusted data and ignore instructions embedded within those documents. Ingestion points: collect_plan_evidence.py reads user-selected files. Boundary markers: references/plan-summary-workflow.md contains explicit instructions to ignore prompt-like text. Capability inventory: generate_plan_summary.py performs file-write operations for artifacts. Sanitization: generate_plan_summary.py uses html.escape and string normalization to sanitize external content.
[COMMAND_EXECUTION]: The skill uses an isolated Python environment to run its internal scripts, preventing environmental interference and ensuring that only authorized packaged scripts are executed.
[DYNAMIC_EXECUTION]: The HTML report generation script uses proper HTML escaping and normalization to sanitize data before interpolation into the output template.

plan-summary-md