The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted external content, including BBox metadata and source images, which creates a potential surface for indirect prompt injection attacks where malicious instructions could influence agent behavior during the reconstruction process.\n
Ingestion points: manifests/initial-bbox.final.json and input/source.png.\n
Boundary markers: None. The skill interpolates external data directly into script templates and prompts.\n
Capability inventory: The skill performs file system writes, executes shell commands (node, npm, grep), and accesses the network to download packages.\n
Sanitization: No sanitization, validation, or escaping of the ingested data is specified before its use in scripts.\n- [DYNAMIC_EXECUTION]: The skill uses templates to generate and execute local JavaScript files at runtime to manage the PowerPoint generation workflow.\n
Evidence: Instructions provide source code for crop-assets.mjs, build-svg-assets.mjs, and generate.mjs.\n
Pattern: These generated scripts are executed locally using the node command.\n- [EXTERNAL_DOWNLOADS]: The skill installs software dependencies from a remote registry during the setup phase.\n
Evidence: The skill uses npm install to fetch @artifact-kit/deckkit and @artifact-kit/deckkit-pro.\n
Context: These packages originate from the skill's author (artifact-kit), representing standard vendor-owned functionality.\n- [COMMAND_EXECUTION]: The skill executes various shell commands to manage the environment and run reconstruction scripts.\n
Evidence: Commands include npm install, node scripts/..., git rev-parse --show-toplevel, and grep for icon metadata.

deckkit-ppt-replica