The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest and summarize external resources including webpages, YouTube transcripts, PDFs, and user-provided notes. This creates an inherent attack surface where malicious instructions embedded within these external sources (e.g., hidden text in a webpage or a transcript) could attempt to influence the agent's behavior.
Ingestion points: The references/intake.md file defines comprehensive workflows for fetching and parsing content from various external URLs and file types.
Capability inventory: The skill has the ability to execute local scripts via subprocess (in scripts/test_skill.py) and generate multiple document formats (Markdown, DOCX, PDF, CSV, XLSX, JSON) as detailed in references/output-formats.md.
Boundary markers: Although SKILL.md and references/synthesis.md provide guidance on separating facts from inferences, the skill does not define strict delimiters or "ignore instructions" wrappers when interpolating raw external content into the prompt.
Sanitization: There is no evidence of automated sanitization or filtering of external content before it is processed by the agent.
[COMMAND_EXECUTION]: The script scripts/test_skill.py utilizes subprocess.run() to execute other internal scripts, such as scripts/probe_reading_notes.py, to perform help-text verification and self-tests. While these commands are constructed using static local paths and the current Python executable, they represent a functional capability to launch shell processes.
[DYNAMIC_EXECUTION]: The skill includes a testing framework that dynamically executes local Python scripts to validate the package's integrity. Specifically, scripts/test_skill.py invokes the Python interpreter to run scripts/probe_reading_notes.py with different arguments to verify its classification logic.

reading-notes