The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides an orchestration layer for processing user-supplied text and files, which constitutes an indirect prompt injection attack surface.
Ingestion points: Untrusted content enters the workflow via the pasted_text and named_file fields in the handoff envelope defined in references/handoff-contract.md.
Boundary markers: The skill implements a structured YAML handoff envelope to maintain context boundaries between different processing stages.
Capability inventory: The orchestration graph includes routes to skills with file system write capabilities (file-edit-in-place).
Sanitization: No specific sanitization logic is present in the provided validation scripts; security relies on the guardrails and safety protocols of the downstream specialized skills.
[COMMAND_EXECUTION]: The project includes a test script that executes local Python code to validate the orchestration graph.
Evidence: The file scripts/validate_connections.test.py utilizes subprocess.run to execute the sibling script scripts/validate_connections.py.
Context: This is a standard development tool for verifying graph integrity and does not interact with untrusted remote sources or pass unvalidated user input to the shell.

avoid-ai-writing-router