steel-session-debugging
Pass
Audited by Gen Agent Trust Hub on Jun 13, 2026
Risk Level: SAFE
Full Analysis
- [COMMAND_EXECUTION]: The skill executes the steel CLI and local helper scripts to retrieve session diagnostics. These operations are scoped to the vendor's own tools and the local project directory, used solely for gathering evidence during a debug session.
- [DATA_EXFILTRATION]: The skill implements strong data protection by using a specialized redaction script (redact-session-debug.mjs). This script identifies and removes sensitive tokens, including Bearer tokens, Steel API keys, cookies, and email addresses, from logs before the agent processes them. Instructions also explicitly forbid sharing raw logs or screenshots externally.
- [PROMPT_INJECTION]: The skill handles browser logs and agent traces which are external, untrusted ingestion points (Cat 8 surface). It mitigates injection risks by requiring deterministic local redaction and structured summarization according to a fixed taxonomy. Ingestion points: scripts/collect-session-debug.mjs. Boundary markers: Local JSON storage and explicit output formatting. Capability inventory: execFile (vendor CLI) and file system writes. Sanitization: Comprehensive regex-based redaction in scripts/redact-session-debug.mjs.
Audit Metadata