optimize-trace-observability
Pass
Audited by Gen Agent Trust Hub on Sep 23, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes external trace data and local agent source code, which are potential vectors for indirect prompt injection.
- Ingestion points: External trace data retrieved via the Coval CLI/API and local project files analyzed through code search (e.g., SKILL.md Phase 1 and Phase 1b).
- Boundary markers: The skill does not mandate specific delimiters or "ignore" instructions for the content of the traces it processes.
- Capability inventory: The skill utilizes network access to communicate with the Coval API and file system read access for project analysis.
- Sanitization: Phase 4 explicitly instructs the removal or bounding of sensitive data, including API keys, tokens, and PII, before committing enrichment.
- [COMMAND_EXECUTION]: The skill instructions suggest using the
grepcommand to search the local codebase for business logic identifiers (e.g., 'cart', 'order', 'payment') to identify instrumentation candidates. - [DATA_EXFILTRATION]: Trace data and metrics are exported to the vendor's platform using the Coval CLI and API. This functionality is intended for observability, and the skill includes steps to ensure that credentials and sensitive secrets are not included in the telemetry data.
Audit Metadata