The Agent Skills Directory

[PROMPT_INJECTION]: The skill manages a surface for indirect prompt injection by reading output from other coding agents. * Ingestion points: Untrusted data enters the context through paneflow read (scrollback) and paneflow status --json (last_result). * Boundary markers: The skill explicitly instructs agents to wrap peer scrollback in <untrusted_terminal_output> fences. * Capability inventory: The conductor can dispatch prompts with auto-submission (--submit), spawn new agents (up), and execute multi-step pipelines (flow run). * Sanitization: Instructions explicitly warn the agent to treat peer output as data to analyze, never as instructions to follow, and to watch for injection attempts.
[COMMAND_EXECUTION]: The skill relies on local shell command execution via the paneflow CLI and utility tools. * Evidence: Uses paneflow ps, paneflow ls, paneflow up, mktemp, and jq for fleet management and state parsing. * Mitigation: Employs pre-flight checks to ensure the IPC socket is available and requires human-in-the-loop confirmation for destructive or ambiguous actions unless explicit free-access mode is enabled.

paneflow-conductor