The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill integrates with external data sources and application runtimes, creating a surface where the agent could ingest untrusted data (such as application logs or user-provided messages).
Ingestion points: Ingests application logs and file contents via the reflexive debugging tool, and user-provided inputs via the HustleChat interface. It also references project metadata from external platforms like Migrate.fun.
Boundary markers: The skill explicitly documents a "review-first" model for all financial operations and requires explicit user confirmation before executing value-moving actions. Built-in moderation and approval prompts are mentioned for plugin data access.
Capability inventory: The reflexive tool provides capabilities for read_file, list_directory, and search_logs. The hustle-incognito SDK allows for tool execution via a custom plugin system.
Sanitization: Documentation in references/plugins.md provides examples for implementing PII protection hooks to redact sensitive information before it reaches the AI context.
[SAFE]: All external infrastructure referenced (e.g., api.emblemvault.ai) and packages are owned or maintained by the author (EmblemCompany). The skill emphasizes secure session management and warns against hard-coding sensitive credentials in prompts or configuration files.

emblem-ai