The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data (PDF, EPUB, Docx, HTML) to generate summaries. This creates a surface area for indirect prompt injection if the source documents contain malicious instructions designed to influence the subagents during the fan-out phase.
Ingestion points: scripts/extract_text.py reads various external file formats.
Boundary markers: The skill uses SOURCE_DEMARC ("========== rabbit-reads source: %s ==========") to separate concatenated sources.
Capability inventory: The skill utilizes subprocess.run to call external converters and a sibling security scanner (scan.py). It also performs file read/write operations.
Sanitization: The skill proactively mitigates this risk by running rwlib.injection.scan (from the sibling rabbit-writes engine) over the raw markup/text of all sources before processing, specifically looking for concealed directives.
[COMMAND_EXECUTION]: The skill uses subprocess.run to invoke external system utilities and sibling scripts.
scripts/extract_text.py executes pdftotext and textutil to normalize document formats.
scripts/check_notes.py executes a sibling script scan.py to perform security audits on the generated notes.
tests/helpers.py executes the skill's own scripts as subprocesses for validation.
These calls use explicit argument lists rather than shell strings, reducing command injection risks.
[DYNAMIC_EXECUTION]: The test suite employs dynamic loading techniques to discover and execute tests.
tests/run.py uses the __import__ function for test discovery.
tests/conftest.py uses importlib.util to dynamically load helpers.py.
These instances are limited to the testing environment and are standard for specialized test runners.

rabbit-reads