read-codex-session
Pass
Audited by Gen Agent Trust Hub on Nov 12, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data by reading and exporting local Codex session transcripts (rollout JSONL files).
- Ingestion points:
scripts/session_export.pyreads data from~/.codex/sessions/and~/.codex/archived_sessions/. - Capability inventory: The skill can list session metadata, read file contents, and write exported transcripts to a local path specified by the user or the agent.
- Boundary markers:
SKILL.mdincludes instructions for the agent to "distinguish facts stated by the user from conclusions made by the agent" and "Treat transcripts as potentially sensitive," which helps mitigate but does not eliminate the risk of the agent following instructions found within the logs. - Sanitization: While the script filters out reasoning, tool calls, and injected context, it does not sanitize the actual message content for prompt injection patterns.
- [COMMAND_EXECUTION]: The skill requires the agent to execute a local Python script (
session_export.py) to list and export session data. It also suggests using shell utilities likewc,rg, andmktempto manage the resulting Markdown files.
Audit Metadata