The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted text from Israeli government documents provided by the user or read from local files (Step 1 in SKILL.md). This creates a vulnerability surface where adversarial instructions embedded in the documents could attempt to influence the agent's behavior during summary generation.\n
Ingestion points: User-provided text snippets or file content read via the Read tool.\n
Boundary markers: There are no explicit instructions to use delimiters or ignore embedded commands within the processed document text.\n
Capability inventory: The agent has access to Bash (executing a local analyzer script), Read (file system access), and WebFetch (network access).\n
Sanitization: The input text is processed by a regex-based Python script, but the extracted results are then interpreted by the LLM without specific sanitization against prompt injection.\n- [COMMAND_EXECUTION]: The skill invokes a bundled Python script (scripts/document-analyzer.py) using the Bash tool to extract structured data such as dates, monetary amounts, and legal references from document text. The script uses standard libraries and performs regex-based parsing without any dynamic code execution.\n- [EXTERNAL_DOWNLOADS]: The skill uses WebFetch to retrieve information from well-known and official Israeli services to provide accurate legal context. This includes gov.il (Tax Authority, PIBA, Enforcement Authority), btl.gov.il (National Insurance), and kol-zchut.org.il (a well-known civic rights database). These operations are part of the skill's primary functionality and target established, trusted domains.

israeli-bureaucracy-decoder