edge-pipeline-orchestrator
Pass
Audited by Gen Agent Trust Hub on Oct 6, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The orchestrator uses subprocess.run to execute pipeline stages by calling local Python scripts.
- Evidence: Found in scripts/orchestrate_edge_pipeline.py within the run_stage function.
- Mitigation: The implementation uses a list of arguments and does not use shell=True, which minimizes the risk of command injection through shell metacharacters.
- [INDIRECT_PROMPT_INJECTION]: The skill has attack surfaces for indirect prompt injection as it ingests and processes external data (OHLCV CSV, tickets/drafts YAML) and instructions from the review stage.
- Ingestion points: scripts/orchestrate_edge_pipeline.py reads user-provided file paths for OHLCV data, tickets, concepts, drafts, and review results.
- Boundary markers: Not applicable as this script coordinates data flow between internal scripts rather than constructing LLM prompts directly.
- Capability inventory: subprocess.run is used to execute external scripts (auto_detect, hints, concepts, drafts, review, and export).
- Sanitization: The script uses Path.resolve() for resolving file paths, yaml.safe_load() for secure YAML parsing, and a rigid string-matching whitelist in apply_revisions to process review instructions.
Audit Metadata