The Agent Skills Directory

[PROMPT_INJECTION]: The skill implements a project-wide data persistence mechanism that ingests information from external research and various sub-agents, creating a surface for indirect prompt injection.
Ingestion points: The agent is instructed to record data from component-vuln-intel (web/terminal output) and deliverables from sub-agents using the upsert_project_fact tool.
Boundary markers: There are no instructions to use delimiters or sanitization protocols when storing raw request/response evidence or findings into the blackboard.
Capability inventory: The agent possesses terminal access (curl), browser navigation capabilities, and the ability to modify the project database.
Sanitization: The instructions lack guidance on escaping or validating untrusted data before it is persisted and subsequently retrieved for decision-making.
[COMMAND_EXECUTION]: The skill mandates high-autonomy behaviors including automated horizontal credential spraying and the execution of complex reconnaissance sequences (7-step vulnerability intelligence) using terminal and browser tools whenever new components or credentials are identified.
[DATA_EXFILTRATION]: The skill establishes a centralized repository for highly sensitive penetration testing artifacts, including credentials, exploit evidence, and environment architecture. While the storage is described as a project-bound SQLite database, the systematic aggregation of this data represents a significant exposure risk if the persistence layer or the agent's context is compromised.

pentest-blackboard