The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from various external, untrusted sources such as web search results (via Firecrawl) and scientific or regulatory databases (PubMed, CFSA, USP, etc.).
Ingestion points: Data enters the agent's context through external retrieval interfaces documented in external-search.md and food-knowledge.md.
Boundary markers: The instructions do not specify any delimiters or safety prompts to isolate external content or instruct the agent to ignore potentially malicious instructions embedded in the retrieved data.
Capability inventory: The skill possesses significant capabilities, including executing shell commands for updates (SKILL.md) and performing file system writes and edits (import.md).
Sanitization: No explicit instructions for sanitizing, escaping, or filtering the external content are provided in the skill files.
[DYNAMIC_EXECUTION]: The skill implements an automated upgrade logic that executes shell commands dynamically based on state signals from the backend server.
Evidence: SKILL.md instructs the agent to monitor for an HTTP 426 Upgrade Required response. Upon receipt, the agent is directed to execute npx --yes skills@latest update deepknow to modify the skill's code.
[REMOTE_CODE_EXECUTION]: The automated update mechanism involves downloading and installing packages from a remote source.
Evidence: The skill uses npx to fetch and execute the latest version of the management tool and potentially re-install the skill from the SBTbiotech/deepknow-skill repository.
[COMMAND_EXECUTION]: The skill provides the agent with shell command templates using curl and npx. These are used for authenticated API requests and system-level maintenance tasks, involving direct execution in the host environment.

deepknow