The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input from external URLs and local files.
Ingestion points: External content is ingested via file reading or URL fetching (SKILL.md).
Boundary markers: Includes instructions to strip sidebars and verify numbers, providing a basic sanity check for the processed content.
Capability inventory: The skill environment allows for file writing and shell execution.
Sanitization: Structural verification is performed by a local validation script (verify.sh).
[DYNAMIC_EXECUTION]: The skill utilizes Python and Bash scripts for document assembly.
Evidence: SKILL.md Step 4 provides a Python snippet and a Bash command to be executed by the agent.
Pattern: These scripts perform local file manipulation based on provided templates. The execution is limited to specific utility tasks defined in the instructions.
[COMMAND_EXECUTION]: The skill requires the agent to execute local command-line tools.
Evidence: Instructions include executing python3 for string replacement and bash for structural verification (SKILL.md).
[EXTERNAL_DOWNLOADS]: The skill loads functional resources from external services.
Evidence: The MathJax library is loaded from jsdelivr.net (skeleton.html).
Context: This is a well-known service used for mathematical formula rendering and is documented as a standard dependency.

ape-present