audit-activation
Pass
Audited by Gen Agent Trust Hub on Oct 30, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes data from
.claude/logs/subagent-spawns.jsonland~/.claude/analytics/agent-usage.jsonl. These files could contain untrusted data that is later presented to the agent, potentially influencing its behavior. - Ingestion points: The
scripts/collect-activation-audit.mjsscript reads and parses JSONL log files from user-specified telemetry roots and the global analytics path. - Boundary markers: The skill instructions do not explicitly mandate delimiters when the agent reads files, but the script acts as a processing layer that filters content.
- Capability inventory: The skill uses
Read,Grep,Bash, and Claude Code task tools (TaskCreate,TaskUpdate,TaskList). - Sanitization: The collector script uses
JSON.parseand extracts specific fields (subagent_type,agent, etc.), which provides a layer of sanitization by ignoring unrelated or potentially malicious keys in the log objects.
Audit Metadata