hunt
Pass
Audited by Gen Agent Trust Hub on May 5, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: The skill's core workflow relies on executing local commands and scripts for bug reproduction and bisection. Examples include running deterministic repro scripts, using
git bisect run, and utilizing low-level instrumentation tools likestrace,dtrace, andeBPF(described in SKILL.md). - [EXTERNAL_DOWNLOADS]: To perform root-cause analysis, the skill fetches telemetry data from external platforms including Sentry, Datadog, AWS CloudWatch, and GCP Logging. These operations are performed via MCP servers, official vendor CLIs, or HTTP APIs as detailed in references/telemetry.md.
- [DATA_EXFILTRATION]: While the skill accesses potentially sensitive information such as stack traces, logs, and trace spans, this data is ingested into the agent context for diagnostic purposes. The instructions do not contain patterns for sending this data to unauthorized external endpoints.
- [PROMPT_INJECTION]: The skill implements a set of 'Hard rules' that instruct the agent to resist user requests for shortcuts (e.g., symptom suppression, fixing without a repro). This is a defensive persona design rather than a malicious injection.
Audit Metadata