The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill requires the installation of standard third-party Python libraries (pdfplumber, PyPDF2, python-docx) from a registry during setup. These dependencies are consistent with the skill's documented document-parsing functionality.
[COMMAND_EXECUTION]: The skill instructs the agent to execute a local Python script (scripts/main.py) to process documents. This script accesses the file system to read input files and write output artifacts, which is the primary intended behavior.
[INDIRECT_PROMPT_INJECTION]: The skill parses untrusted Investigator's Brochure documents (PDF, DOCX, TXT) provided by the user. The extracted text is then reviewed by the AI agent for summarization, creating a surface where malicious instructions embedded in the documents could attempt to influence agent behavior.
Ingestion points: The scripts/main.py script reads external files passed via the input_file command-line argument.
Boundary markers: Absent. The instructions do not provide the agent with specific delimiters or guidelines to distinguish extracted document text from system instructions.
Capability inventory: File system read and write operations are present in scripts/main.py to handle document ingestion and output generation.
Sanitization: Absent. The script performs direct text extraction and regular expression matching without sanitizing the output for potential instruction injection.
[METADATA_POISONING]: The skill package includes a file ib-summarizer_audit_result_v2.json containing self-authored security 'PASS' verdicts and high safety scores. This metadata is potentially misleading as it attempts to establish authoritative safety claims that could cause users or agents to disregard standard security precautions based on unverified assertions.

ib-summarizer