The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The script scripts/ensure-html-to-composition.mjs performs a global installation of the html-to-composition package from the npm registry using npm install -g. This behavior introduces external, unverified code into the environment at runtime.
[COMMAND_EXECUTION]: The skill executes multiple subprocesses, including npm for package management and the aippt-html-import CLI for data processing. These operations are performed using child_process.execFile, allowing for arbitrary execution based on the installed package's behavior.
[CREDENTIALS_UNSAFE]: The scripts/publish-template.mjs script manages administrative login credentials, including phone numbers and passwords. It includes a hardcoded default password (123456) and transmits this sensitive data to an external authentication endpoint at https://prism-stone-pre.byering.com.
[DATA_EXFILTRATION]: The skill reads local configuration from .env files and uploads user-provided slide content to an external Aliyun OSS bucket (prism-stone-file.byering.com) via pre-signed URLs obtained from a remote server.
[PROMPT_INJECTION]: The skill processes untrusted HTML data supplied by the user, creating a surface for indirect prompt injection attacks.
Ingestion points: HTML files are ingested via the aippt-html-import CLI as described in SKILL.md.
Boundary markers: There are no markers or instructions to prevent the agent from obeying instructions embedded within the processed HTML content.
Capability inventory: The skill possesses the capability to execute shell commands and perform network operations across several scripts.
Sanitization: While the skill validates asset URLs for specific hostnames, it does not perform sanitization on the text or HTML structure before conversion.

html-to-aippt-composition