The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/extract_codex_resume.py executes local git commands such as git status, git branch, and git log to provide the agent with a briefing on the current state of the workspace.
Evidence: These calls are made within the get_git_state function to help reconcile historical session data with the current file state.- [PROMPT_INJECTION]: The skill implements an indirect prompt injection surface by ingesting historical conversation logs and re-injecting them into the agent's active context.
Ingestion points: Historical session data is read from rollout-*.jsonl files located in the ~/.codex/sessions/ directory.
Boundary markers: The context briefings generated for the agent do not include explicit delimiters or instructions to treat the historical data as untrusted or to ignore embedded commands.
Capability inventory: The skill is designed to guide the agent in continuing previous work, which typically involves file system modifications. The extraction script itself performs local command execution via git.
Sanitization: The parsing logic filters out system preamble noise and truncates long messages, but it does not perform sanitization or validation to prevent the execution of instructions contained within the historical logs.

continue-codex-work