The Agent Skills Directory

[SAFE]: The skill demonstrates security best practices by implementing a bounded execution environment for sub-agents.\n
Access Control: The advisor mode requires explicit user activation via the /advisor command and cannot be enabled automatically by the agent.\n
Data Privacy: Instructions in SKILL.md and references/briefing-template.md explicitly mandate the removal of secrets and unrelated private context from briefings before they are sent to the advisor model.\n
Least Privilege: The advisor sub-agent is strictly prohibited from editing files, running state-changing commands, or spawning further sub-agents, as defined in references/advisor-subagent.md.\n
Trusted Source: The skill is mirrored from the official repository of a well-known technology company, and the instructions focus on native platform capabilities rather than external network calls.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes external data (user requests and tool outputs) in a sub-agent briefing, creating a potential vulnerability surface.\n
Ingestion points: references/briefing-template.md (User's request and Evidence sections).\n
Boundary markers: Absent; data is interpolated into Markdown headings.\n
Capability inventory: The sub-agent has read-only access to repository data and metadata (git log, grep).\n
Sanitization: The skill mitigates risks by requiring manual secret redaction and enforcing read-only permissions for the sub-agent.

cursor-advisor