meta-packager

Pass

Audited by Gen Agent Trust Hub on Aug 22, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses local shell utilities including tail, find, and sqlite3 to inspect agent logs and metadata stores for patterns. These are used in a read-only capacity to gather evidence of repetitive work.- [DATA_EXPOSURE]: Instructions direct the agent to read sensitive interaction history from ~/.codex/history.jsonl and ~/.codex/state_5.sqlite. While the skill includes an explicit instruction to avoid credential files, these history logs may contain sensitive information or secrets previously entered by the user during a session.- [INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests untrusted data from previous session logs to generate new executable skills and automations.
  • Ingestion points: Agent history logs (~/.codex/history.jsonl) and session memories.
  • Boundary markers: The skill does not define specific boundary markers for the ingested text.
  • Capability inventory: The skill has the capability to write new files (skills, subagents, automations) to local configuration directories.
  • Sanitization: The skill incorporates design-time safeguards including a two-occurrence threshold for new items and a requirement for the agent to present a shortlist table for user review before file creation.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 22, 2026, 05:39 AM
Security Audit — agent-trust-hub — meta-packager