The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external sources including email bodies and meeting transcripts through spark search, spark thread, and spark meeting commands. This creates a surface for indirect prompt injection where malicious content in an email could attempt to influence the agent's behavior.
Ingestion points: Email search results, thread details, and meeting notes accessed via the spark CLI (SKILL.md).
Boundary markers: No specific delimiters or safety instructions are provided to the agent to treat this data as untrusted.
Capability inventory: The agent can draft emails (spark draft), share information with teams (spark action shareInTeam), and assign tasks (spark action assign).
Sanitization: No explicit sanitization or filtering of the ingested content is specified.
Mitigation: The skill mitigates risks by explicitly instructing the agent to always confirm email drafts with the user before creation, ensuring human-in-the-loop oversight for high-impact actions.
[COMMAND_EXECUTION]: The skill uses the spark command-line interface to manage project communications. These commands are restricted to the functionality provided by the use-spark base skill and operate within the expected scope of a project management persona.

spark-persona-project-manager