The Agent Skills Directory

[COMMAND_EXECUTION]: The skill requires the execution of a local Python script (scripts/build_output.py) to concatenate and format the final worksheet. This provides a functional surface for file system interactions.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8). It ingests untrusted text from user-provided articles (article.md) and processes them through several reasoning steps without using delimiters or instructions to ignore embedded commands. An attacker could embed instructions within an article to divert the agent's behavior.
Ingestion points: The article.md file and user-provided reading passages are read directly into the agent's context.
Boundary markers: None. The prompts in assets/day_test_template_v2.md do not include 'ignore embedded instructions' warnings or data delimiters.
Capability inventory: The skill can execute Python scripts and perform file read/write operations via scripts/build_output.py.
Sanitization: There is no evidence of input validation or sanitization for the content of the source article before it is processed by the AI or the script.
[DATA_EXFILTRATION]: While no network exfiltration is present, the script scripts/build_output.py reads file paths provided via command-line arguments. If the agent is manipulated via prompt injection, it could be coerced into reading sensitive local files (e.g., .env or configuration files) and writing their content into the public-facing completed.md output.

reading-cloze-builder