The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and follow instructions from external files located at .qp/EMI.md and ~/.qp/EMI.md. This creates a vector where content from these files can steer the agent's behavior. Ingestion points: Files .qp/EMI.md and ~/.qp/EMI.md (SKILL.md). Boundary markers: Instructions state that EMI files are subordinate to system safety and developer authority. Capability inventory: Writing HTML reports to ~/.qp/report/, updating wisdom indices in ~/.qp/OGBON.md, and invoking other skills like html-artifact and Iwadi (SKILL.md, cross-session-learning.md). Sanitization: The skill explicitly instructs the agent to ignore prompt-like commands within the OGBON.md evidence files.
[PERSISTENCE_MECHANISMS]: The file references/global-activation.md contains instructions for the agent to modify the host's global instruction file to ensure the skill activates in every session. This functions as a persistence mechanism for agent behavior across the entire workspace.
[DATA_EXPOSURE_AND_EXFILTRATION]: The skill aggregates session activity, including links to local and remote artifacts, into a single HTML file stored in ~/.qp/report/. While the instructions forbid including credentials or secrets, the aggregation of session metadata in a persistent format increases the risk associated with local data exposure.

olofofo