The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and summarizes data from multiple shared sources, including Google Docs (sources.ops_weekly) and Google Sheets (sources.main_tracker), which are subject to edits by various team members. This creates a surface for indirect prompt injection where instructions hidden in these logs or trackers could potentially influence the agent's behavior.
Ingestion points: The skill reads from sources.ops_weekly (Weekly Ops Log) and sources.main_tracker (Project Tracker).
Boundary markers: While the routine-prompt.tmpl provides structural instructions, there are no specific delimiters or "ignore" instructions applied to the ingested external content during the summarization process.
Capability inventory: The skill has access to tools for sending emails (mcp__google-workspace__send_message) and writing to local cache directories (limits.doc_dump_dir).
Sanitization: Basic markdown link stripping is performed in the parsing scripts, but no semantic sanitization for embedded instructions is present.
[COMMAND_EXECUTION]: The skill utilizes several Python scripts included in the scripts/ directory (derive_state.py, extract_newest_block.py, kpi_locate.py, render_routine_prompt.py, tracker_diff.py) to process data and prepare the final report. These scripts are invoked via the python3 command and rely exclusively on the Python standard library.
[DATA_EXFILTRATION]: As part of its primary functionality, the skill collects internal company metrics, project statuses, and financial runway data from various sources (Supabase, Google Workspace, CRM) and sends this aggregated information to specific recipients via Gmail.
[DYNAMIC_EXECUTION]: The skill uses a template rendering script (render_routine_prompt.py) to dynamically generate a self-contained prompt from a template and a local configuration file. This rendered prompt is used to configure a cloud-based agent routine. This represents a low-risk use of dynamic generation for standard automation purposes.

zynkr-gm