The Agent Skills Directory

[PERSISTENCE]: The skill is designed to run as a scheduled background process on a local machine using macOS launchd. The installation instructions in references/scheduling.md and the runner script scripts/run_ops_weekly.sh direct the user to install a LaunchAgent plist to maintain execution across sessions.
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection because it ingests untrusted data from Google Chat messages and processes that data to perform actions like writing to a Google Doc and sending emails.
Ingestion points: Data enters via the mcp__google-workspace__get_messages tool as documented in SKILL.md and handled by scripts/parse_reports.py.
Boundary markers: While the skill uses marked blocks (〔自動彙整 …〕) to isolate its output in the Google Doc, it lacks explicit prompt delimiters or instructions to ignore embedded commands within the ingested chat messages themselves.
Capability inventory: The skill possesses capabilities to write to documents (batch_update_doc), post to chat (send_message), and send emails (send_gmail_message).
Sanitization: scripts/parse_reports.py provides a layer of sanitization by using regular expressions to extract specific report fields before the data is processed by the AI agent.
[COMMAND_EXECUTION]: The installation instructions in SKILL.md suggest running an npx command to add the skill, which executes remote code from the author's repository.

zynkr-ops-weekly