The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external, potentially untrusted documents which could contain embedded instructions intended to influence the agent's behavior once parsed into text.
Ingestion points: Processes images (PNG, JPG, BMP, WEBP, GIF) and PDFs provided by the user, including scans and screenshots.
Boundary markers: Absent; there are no specific instructions or delimiters provided to the agent to treat the OCR results as untrusted data or to ignore instructions found within the parsed text.
Capability inventory: Reads local files, performs network requests to the Zhipu cloud API, writes output files (JSON, Markdown, and images) to the local disk, and can output data to stdout for further processing.
Sanitization: Absent; the skill returns the raw content extracted from the documents to the agent without filtering.
[EXTERNAL_DOWNLOADS]: The skill instructs users to install the glmocr Python package. This is a vendor-owned SDK resource necessary for the skill's operation.
[COMMAND_EXECUTION]: The skill documentation includes examples of using the glmocr CLI and piping results into other utilities like jq. Agents should be aware that these commands involve shell execution.
[DATA_EXFILTRATION]: The skill transmits the content of local documents to Zhipu's cloud API for OCR processing. This is the primary intended function of the skill and requires a user-provided ZHIPU_API_KEY for authentication.

glmocr-sdk