The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data exported from a third-party service (web.letterly.app) and uses it as context for agent-driven metadata generation. This creates a surface where malicious instructions embedded in transcriptions could influence the agent's behavior during the metadata generation phase.
Ingestion points: The script scripts/workflow.py reads Letterly-export*.csv files from the unprocessed/ directory.
Boundary markers: No explicit boundary markers or delimiters for the external content were identified in the orchestration script.
Capability inventory: The skill can execute sub-scripts via subprocess.run, create directories, and move files using shutil.move within the Obsidian vault.
Sanitization: No sanitization of the transcription content is performed before it is passed to subsequent steps.
[COMMAND_EXECUTION]: The script scripts/workflow.py utilizes subprocess.run to orchestrate a multi-stage pipeline by calling internal scripts such as export.py, process.py, and link_notes.py. While these are internal components, the script relies on a hardcoded absolute path (/Users/rami/Documents/life-os/ai-agents-config/skills) to locate the Python executable and sub-skills.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of Playwright browser binaries (playwright install chromium) to facilitate the export process from the Letterly web application. Playwright is a well-known automation library from a trusted organization.

letterly-automation