The Agent Skills Directory

[DATA_EXFILTRATION]: The skill encourages a risky data handling practice by instructing users to capture and provide "full unredacted dumper / analyzer / visualizer output" for documentation and debugging purposes. These outputs from the doca_flow_tune tool can contain sensitive network telemetry, hardware counter states, and internal pipeline configurations that could be exposed to the agent or third parties. Evidence: Found in SKILL.md (line 234) and TASKS.md (line 235).
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection as it is designed to ingest and analyze untrusted external data in the form of CSV, JSON, and Mermaid files.
Ingestion points: The agent reads dumper CSV, analyze JSON, and visualize mermaid files provided by the user or generated by tools (SKILL.md, TASKS.md).
Boundary markers: The instructions do not specify the use of clear delimiters or warnings to ignore instructions embedded within these data files.
Capability inventory: The skill allows the agent to propose and draft C/C++ code modifications (minimum-diff changes) to the user's DOCA Flow application based on the analyzed data (TASKS.md).
Sanitization: There is no mention of sanitizing or validating the contents of these external files before they are processed by the agent.

doca-flow-tune