The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external data sources such as Excel, CSV, PDF, Word documents, and pasted text. These sources are inherently untrusted and could contain malicious instructions designed to influence the agent's behavior when they are loaded into the LLM context for profiling or recipe drafting.\n
Ingestion points: Data ingestion occurs through the ingest.read_any, ingest.read_large, and ingest.read_paste methods as described in SKILL.md.\n
Boundary markers: The instructions lack explicit requirements for the agent to use boundary markers or specific prompts to ignore embedded instructions within the processed files.\n
Capability inventory: The skill has the ability to write local files (write_xlsx), deploy new script bundles (emit_runner), and execute shell commands (python tidy_[name].py).\n
Sanitization: While the skill provides data formatting and validation for the output, it does not explicitly sanitize raw data content before it is read by the AI agent.\n- [DYNAMIC_EXECUTION]: The skill implements a pattern where it generates a local Python script (a 'runner') containing a data-cleaning recipe and subsequently instructs the agent to execute it.\n
Evidence: SKILL.md describes the emit_runner process which creates a .py file containing a baked recipe and instructs the agent to run it via python tidy_[name].py [file].\n- [COMMAND_EXECUTION]: The agent is instructed to perform shell operations to run cleaning scripts and check the environment.\n
Evidence: Instructions include executing python tidy_[name].py and python ../../scripts/envcheck.py.\n- [EXTERNAL_DOWNLOADS]: The documentation identifies several external Python libraries and a binary tool required for functionality, which are well-known in the industry.\n
Evidence: SKILL.md lists openpyxl, PyMuPDF, pdfplumber, python-docx, and extract_msg as requirements, and mentions the Tesseract OCR binary.

data-tidy