The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from repository-local files such as AGENTS.md, CLAUDE.md, and project documentation during the planning process. While it lacks explicit boundary markers or sanitization for this content, the risk is mitigated by a structured workflow that requires user review and approval via AskUserQuestion at every phase transition.
Ingestion points: Reads repository-local instructions (AGENTS.md, CLAUDE.md), production settings (production/review-mode.txt), and design rules (design/live-ops/economy-rules.md).
Boundary markers: Absent; instructions do not specify delimiters for external content.
Capability inventory: Access to Write, Edit, Bash, and Task tools for creating documents and spawning sub-agents.
Sanitization: Absent.
[COMMAND_EXECUTION]: The skill uses the Bash tool to perform repository checks and summarize changes. The usage is scoped to standard development workflows and does not include suspicious patterns or privilege escalation.
[DYNAMIC_EXECUTION]: The skill uses the Task tool to spawn sub-agents for specific roles (e.g., economy-designer, analytics-engineer). This is a standard orchestration pattern for multi-agent workflows and operates within the platform's execution model.
[SAFE]: The skill includes a proactive security/safety feature in Phase 7, where it validates economy designs against an ethics-policy.md file to prevent predatory mechanics, blocking completion if violations are found.

team-live-ops