The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements a persistent memory system that injects cross-session user modeling, AI identity cards, and session summaries directly into the agent's system prompt.
Ingestion points: User-authored conclusions via the honcho_conclude tool and automatic observation of user messages.
Boundary markers: The skill uses specific context blocks (Session summary, User representation, AI peer card) and documentation notes that delimiter sequences are escaped to prevent prompt structure corruption.
Capability inventory: The skill provides tools for searching (honcho_search), reasoning (honcho_reasoning), and writing persistent facts (honcho_conclude) across the memory workspace.
Sanitization: A dedicated 'Memory-context sanitization' mechanism is documented which strips XML/HTML tags from user conclusions, normalizes whitespace, and escapes delimiters before injection into the system prompt.
[CREDENTIALS_UNSAFE]: The skill manages an apiKey and baseUrl for the Honcho service. It correctly instructs users to configure these via an interactive setup wizard (hermes memory setup honcho) rather than hardcoding credentials, following best practices for secret management.
[EXTERNAL_DOWNLOADS]: The skill depends on the official honcho-ai Python package and facilitates connections to the service's dashboard (app.honcho.dev). These references are consistent with the skill's primary purpose and target well-known service domains.

honcho