axiom-analyze-triage
Pass
Audited by Gen Agent Trust Hub on Aug 19, 2026
Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes the xcsym triage command, interpolating arguments such as version strings and user counts. This is the primary mechanism for classifying the crash corpus.
- [DATA_EXPOSURE_AND_EXFILTRATION]: The agent is instructed to look for Sentry authentication tokens in ~/.sentryclirc and project-specific configuration files. This is a standard method for credential management in development tools, and the skill explicitly forbids logging or requesting the token from the user.
- [EXTERNAL_DOWNLOADS]: The skill fetches data from Sentry and App Store Connect APIs. These are well-known production monitoring services required for the skill's triage functionality.
- [INDIRECT_PROMPT_INJECTION]: 1. Ingestion points: Sentry API and App Store Connect diagnostic logs. 2. Boundary markers: None explicitly defined for data written to /tmp/corpus.jsonl or interpolated into reports. 3. Capability inventory: Execution of xcsym triage via shell. 4. Sanitization: The skill normalizes data to JSONL before processing but does not specify sanitization of fields like culprit or metadata before semantic analysis.
Audit Metadata