The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process an 'evidence horizon' that includes external sources provided or named by the user, existing work artefacts, and official sources for products or systems. This creates a surface for indirect prompt injection where malicious instructions embedded in these documents could influence the agent's behavior.
Ingestion Points: SKILL.md identifies sources the user provides, work artefacts, and official sources as the primary inputs for building the decision tree.
Boundary Markers: The instructions do not define clear delimiters or instruction-isolation techniques for when the agent processes these external materials.
Capability Inventory: The instructions explicitly tell the agent that 'Finding facts is your job' and to 'Use the available environment, tools, and authoritative sources,' which implies the use of file-system or network tools to retrieve data from the evidence horizon.
Sanitization: There are no instructions for sanitizing, escaping, or filtering content retrieved from the evidence horizon before it is incorporated into the agent's reasoning process.
[COMMAND_EXECUTION]: The skill mandates that the agent autonomously use environment tools to resolve facts rather than delegating to the user. While this is part of the intended functionality, it grants the agent broad autonomy to execute tools (such as file readers or web scrapers) based on potentially untrusted references found in the user-provided evidence horizon.

torture-gently