The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and summarize data from untrusted external sources, such as Teams chat logs, email communications, and external project folders (via the Check-ExternalFolders.ps1 script). This creates an attack surface where malicious instructions embedded in these logs could attempt to influence the agent's behavior during report generation or 1on1 preparation.
Ingestion points: Manual paste of chat/email data into the data-collector agent, synchronization of external files from monitored folders, and workIQ (M365) log retrieval.
Boundary markers: Although the data-collector agent applies YAML frontmatter to normalized data, the prompt templates (e.g., daily-report.prompt.template.md) do not provide strong delimitation or explicit instructions for the LLM to ignore embedded commands within the activity content.
Capability inventory: The skill has the ability to read and edit files, execute terminal commands (run/terminal in general-worker), and run local scripts that interact with the system.
Sanitization: There is no evidence of content sanitization or instruction-filtering applied to external logs before they are processed by the reporting agents.
[COMMAND_EXECUTION]: The skill includes a Python script (assets/scripts/collect_daily_activity.py) that uses subprocess.run to execute git log commands for activity evidence collection. While the script uses parameterized arguments to avoid shell injection, it represents an automated interaction with the local file system and version control tools. Additionally, the general-worker agent is whitelisted for the run/terminal tool to handle unclassified tasks.

biz-ops-setup