The Agent Skills Directory

EXTERNAL_DOWNLOADS (MEDIUM): The skill's pre_execution and post_execution hooks use npx agentic-flow@alpha. This downloads a package from the npm registry at runtime. Since agentic-flow is not a verified trusted source and the skill specifies the @alpha tag, the executed code is unverifiable and could change at any time.
COMMAND_EXECUTION (LOW): The skill executes various shell commands including jq, cat, and the GitHub CLI (gh). This is used for local data processing and environment checks.
PROMPT_INJECTION (LOW): Indirect Prompt Injection Surface (Category 8). The skill is designed for 'GitHub issue management,' which requires processing untrusted external data.
Ingestion points: Content from GitHub issues.
Boundary markers: No delimiters or 'ignore' instructions are used in the shell scripts to isolate external content.
Capability inventory: Subprocess execution via npx and gh, and temporary file writes.
Sanitization: There is no evidence of sanitization or validation of the data retrieved from GitHub before it is processed by the agent.
DATA_EXFILTRATION (LOW): The post_execution hook sends session metadata (task name, agent ID, and status) to the agentic-flow command. While this appears to be for telemetry or state management, it involves transferring internal agent context to an external package.

agent-v3-queen-coordinator