The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses highly imperative and restrictive language in forms.md (e.g., "CRITICAL: You MUST complete these steps in order. Do not skip ahead") which attempts to rigidly override the agent's autonomous decision-making process.- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process text from external PDF files, creating a vulnerability surface for indirect prompt injection.
Ingestion points: scripts/extract_form_structure.py, scripts/extract_form_field_info.py, and extraction logic in SKILL.md via pdfplumber and pypdf.
Boundary markers: None. The skill does not provide the agent with delimiters or instructions to ignore potential commands embedded in the processed PDF text.
Capability inventory: The skill allows file reading, file writing, image generation, and the execution of local shell commands for PDF processing.
Sanitization: Extracted content is presented to the agent without validation or filtering for adversarial prompt instructions.- [DYNAMIC_EXECUTION]: scripts/fill_fillable_fields.py implements a runtime monkeypatch of the pypdf.generic.DictionaryObject.get_inherited method. This dynamically modifies the library's behavior at execution time to handle specific form field attribute formats.- [METADATA_POISONING]: There is a discrepancy between the skill author context ('teamily-ai') and the copyright notice in LICENSE.txt ('Anthropic, PBC'), which could lead to confusion regarding the source and support for the skill components.

pdf