debug-mode

Warn

Audited by Gen Agent Trust Hub on Mar 8, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: Path traversal vulnerability in file operations. The sessionId argument is used to construct file paths without sanitization, potentially allowing the creation or deletion of files outside the intended directory.
  • Evidence: scripts/cli.ts uses path.join(logDir, ...) with unvalidated sessionId input in the /log endpoint and cleanup command.
  • [DATA_EXFILTRATION]: The local server enables Cross-Origin Resource Sharing (CORS) by default, allowing any website visited by the user to potentially interact with the log server endpoints.
  • Evidence: scripts/cli.ts uses HttpMiddleware.cors() without origin restrictions.
  • [EXTERNAL_DOWNLOADS]: Fetches and installs dependencies using Bun and NPX from the author's repository.
  • Evidence: README.md and package.json reference andrueandersoncs/debug-skill and several @effect packages.
  • [PROMPT_INJECTION]: The skill presents raw runtime logs to the agent for analysis, creating an attack surface for indirect prompt injection if the instrumented application logs untrusted data.
  • Ingestion points: Log files read during Phase 5 (Analyze Logs) as described in SKILL.md.
  • Boundary markers: Absent; logs are processed as raw NDJSON without delimiters or instructions to ignore content.
  • Capability inventory: Local server startup, file system modification (write/truncate/delete), and reading of arbitrary local log files.
  • Sanitization: Log content is not sanitized or escaped before being presented to the agent's context.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Mar 8, 2026, 04:13 AM