The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external content which creates a potential attack surface for indirect prompt injection.
Ingestion points: In Phase 0, the skill reads source code files and JSON logs (e.g., planner_state.json, world_model.json) to gather material for slides.
Boundary markers: There are no explicit instructions to use delimiters or ignore embedded instructions when reading these external files, which could allow malicious content in the files to influence the agent's behavior.
Capability inventory: The skill uses the Bash tool to execute pdflatex, git, gh, and hf commands. It also has Write and Edit permissions to create and modify files based on the ingested data.
Sanitization: The skill lacks explicit sanitization or escaping mechanisms for the data retrieved from external logs and code before it is interpolated into LaTeX templates or markdown outlines.
[COMMAND_EXECUTION]: The skill relies on executing local scripts and system binaries to perform its core functions.
Evidence: The skill executes scripts/render_and_check.sh and scripts/md2pdf.py using the Bash tool.
Evidence: It invokes pdflatex for document rendering and pdftoppm for visual verification of the generated slides.
Evidence: Phase 4 involves executing git and hf (Hugging Face) CLI tools to push the final documents to remote repositories, provided the user grants permission.

academic-slides-iterative