errors
Pass
Audited by Gen Agent Trust Hub on Oct 23, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from
.claude/logs/errors.jsonl, which includes outputs and inputs from various tools. This creates an attack surface where a malicious process could intentionally trigger an error message containing instructions to influence the agent's troubleshooting behavior. - Ingestion points:
.claude/logs/errors.jsonl(referenced in SKILL.md). - Boundary markers: None specified in the log parsing instructions.
- Capability inventory: The skill uses
Bashand executes local Python scripts (.claude/scripts/analyze_errors.py). - Sanitization: No explicit sanitization or validation of the log content is described.
- [DATA_EXPOSURE]: The error collector hook captures tool inputs and error messages. These logs can inadvertently store sensitive information such as API keys, tokens, or private data if a command containing such information fails and is logged to
.claude/logs/errors.jsonl. - [EXTERNAL_DOWNLOADS]: The skill instructions suggest the global installation of the
portlesspackage from the NPM registry as a resolution for connection errors. - [COMMAND_EXECUTION]: The skill executes local Python scripts (
analyze_errors.py) and utilizes shell commands (Bash) to search through memory and log files for error pattern matching.
Audit Metadata