The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from a JSON manifest and local image assets, which represents a surface for indirect prompt injection.
Ingestion points: The skill reads poster.json and local PNG/JPEG files (referenced in generate_poster.py and inventory_images.py).
Boundary markers: The skill uses strict schema validation, SHA-256 hashing of all assets, and a canonical content hash that must be approved by the author (see _manifest.py). It also contains explicit instructions to never fabricate data or research scientific claims.
Capability inventory: The skill can write .pptx files, read/write JSON reports, and parse ZIP/XML structures across its utility scripts.
Sanitization: The skill implements defensive XML parsing that rejects DTDs and entities and performs ZIP preflight checks to prevent traversal attacks and expansion bombs.
[EXTERNAL_DOWNLOADS]: The skill instructions include the installation of external Python dependencies from a public registry.
Evidence: SKILL.md contains instructions to run uv pip install "python-pptx==1.0.2" "Pillow==12.3.0" "lxml==6.1.3".
Context: These are official versions of well-known, trusted libraries (python-pptx, Pillow, lxml). The skill enforces exact version pins to ensure a reproducible and safe generation environment.

pptx-posters