harness-orchestration

Pass

Audited by Gen Agent Trust Hub on Jun 21, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes local scripts such as orchestration-scorecard.sh and render-html.sh using the Bash tool for orchestration data processing.
  • [COMMAND_EXECUTION]: The skill instructions direct the agent to write data to files using shell redirection (e.g., > .claude/state/orchestration-scorecard.html), which bypasses the 'Write' tool restriction defined in the frontmatter.
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through its data ingestion process. * Ingestion points: The skill reads orchestration logs and totals from the .claude/state/ directory. * Boundary markers: The instructions do not define markers to separate data from instructions or warn the agent about untrusted content in the logs. * Capability inventory: The skill utilizes Bash for script execution and file system access to process data. * Sanitization: No validation or filtering is mentioned for the data processed from the local JSONL and JSON files before rendering into HTML or terminal output.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 21, 2026, 01:01 PM
Security Audit — agent-trust-hub — harness-orchestration