field-handler

Pass

Audited by Gen Agent Trust Hub on Sep 29, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDYNAMIC_EXECUTIONPERSISTENCEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill instructs the installation of an entire multi-agent tool suite using npx skills add gregbarbosa/skills. This fetches scripts and configurations from the author's repository to provide the core functionality for the field agent system.
  • [COMMAND_EXECUTION]: The instructions rely heavily on the herdr command-line interface to create workspaces, split terminal panes, manage agent identities, and control terminal input/output. These commands are integral to the skill's ability to orchestrate multiple agents in parallel.
  • [DYNAMIC_EXECUTION]: The skill uses python3 -c snippets to dynamically parse JSON output from the herdr CLI at runtime. This allows the agent to extract specific data like pane IDs and agent lists from structured command output.
  • [PERSISTENCE]: A persistent monitoring mechanism is established using a Monitor tool with the persistent: true flag. This loop runs in the background to track agent lifecycle events (done, idle, blocked) and ensures the handler is notified when a sub-agent completes its task.
  • [INDIRECT_PROMPT_INJECTION]: As an orchestrator, this skill processes reports generated by other agents. These sub-agents may be acting on untrusted external data (such as code in a repository), creating a potential surface for indirect injection via report callbacks. The skill mitigates this by instructing the handler to manually verify all claims and files before acknowledging a task as complete.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 29, 2026, 08:51 PM