The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes a local validation script scripts/validate-findings.py via bash to ensure that the findings produced by the agents conform to a machine-readable JSON contract. This is a controlled execution of a script distributed with the skill.
[INDIRECT_PROMPT_INJECTION]: The skill defines a workflow that ingests and compares external user-provided files (Beamer LaTeX/PDF and Quarto QMD/HTML) using sub-agents.
Ingestion points: Document files located during the Phase 0 Pre-flight check (specifically .tex, .pdf, .qmd, and .html extensions).
Boundary markers: The skill references a validation schema (finding-schema.json) and requires a validation script to exit with code 0, providing structured output verification, although it does not explicitly mention input delimiters for the slide content itself.
Capability inventory: The workflow utilizes powerful tools including Bash, Write, Edit, Agent, and Task, which could be targeted if an ingested file contains adversarial instructions.
Sanitization: The skill employs a Python-based validation harness to check the final report structure, which helps prevent schema confusion but does not necessarily sanitize the content of the slides being analyzed.

qa-quarto