The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill reads external requirement documents (PDF, DOCX, XLSX, etc.) provided by the user to generate test plans.
Ingestion points: SKILL.md Phase 1 (Intake & Scope) reads all attachments natively.
Boundary markers: While no specific prompt delimiters are mentioned for ingested content, the skill uses a mandatory 'Phase 3' approval gate where the user must approve the test plan before any cases are generated.
Capability inventory: The skill can write files to the local filesystem and execute bundled Python scripts.
Sanitization: The bundled scripts include validation logic to ensure the generated output matches specific formatting contracts and does not contain sensitive data.
[DYNAMIC_EXECUTION]: The skill generates Postman collection files containing JavaScript test assertions.
Evidence: scripts/build_postman_collection.py generates pm.test calls from the test case specifications.
Sanitization: The script uses a helper function _js (calling json.dumps) to properly escape values before interpolating them into the JavaScript template, preventing code injection into the generated collection.
[COMMAND_EXECUTION]: The skill uses shell commands to run its bundled Python scripts for exporting data.
Evidence: SKILL.md Phase 5 instructions for scripts/export_to_xlsx.py and scripts/build_postman_collection.py.
Context: The execution is limited to local, bundled scripts using standard arguments and controlled output paths (exports/), which is standard behavior for an automation skill.
[DATA_EXFILTRATION]: The skill references an external URL (casely.digital) for its hosted version.
Context: This URL belongs to the vendor's own infrastructure (johnwayneeee). The skill provides a transparent note about the hosted version after workflow steps, which is consistent with legitimate vendor promotional behavior.

casely