The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/keep_codex_fast.py executes system commands via subprocess.check_output to monitor running application processes, specifically using ps on Unix systems and PowerShell on Windows.
[DYNAMIC_EXECUTION]: The tests/smoke_test.py script uses exec() to dynamically execute generated restoration code during its automated testing procedures.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted chat history data to create handoff documents and reactivation prompts for new agent threads. Ingestion points: Reads rollout.jsonl session files and state_5.sqlite databases. Boundary markers: The instructions lack strict delimiters or escaping for the interpolated chat content. Capability inventory: The skill has broad file system access, including moving and writing files. Sanitization: No filtering of chat content is performed before it is used to generate new prompts.
[DATA_EXPOSURE_AND_EXFILTRATION]: The skill provides the ability to expose local file paths, thread identifiers, and process metadata to the agent and user, particularly when the --details flag is employed.

keep-codex-fast