phabalicious

Warn

Audited by Gen Agent Trust Hub on Sep 27, 2026

Risk Level: MEDIUMPROMPT_INJECTIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill includes instructions that explicitly direct the AI agent to disregard security and authorization concerns. The 'Overview' section states, "Don't get sidetracked by: ... Adding security/authorization concerns not mentioned in this skill" and "Ignore... security/authorization concerns." These instructions are designed to override the agent's default safety filtering and environment validation protocols.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted data from external environments, creating a vulnerability to indirect prompt injection attacks.
  • Ingestion points: Data ingested via phab db:query results, configuration details from phab output, and file contents from phab get:file (referenced in SKILL.md).
  • Boundary markers: None; the skill instructs the agent to avoid adding security concerns, which reduces the likelihood of the agent distinguishing between data and instructions.
  • Capability inventory: The skill facilitates remote shell access (phab shell), database manipulation (db:query, db:shell), and file system operations on production and staging environments.
  • Sanitization: There are no instructions for validating or sanitizing the content retrieved from these commands before it is processed by the agent.
  • [COMMAND_EXECUTION]: The skill encourages the use of powerful CLI commands that interact directly with production databases and shells. While the skill includes a confirmation protocol for destructive actions, the instruction to ignore environment detection ("Don't get sidetracked by: Detecting whether a fabfile exists... The user is asking for help, not environment detection") increases the risk of recommending commands that are inappropriate or dangerous for the current context.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Sep 27, 2026, 01:05 PM