The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes arbitrary shell commands defined in the optimization spec's measurement.command field via scripts/measure.sh. This is an intended core feature for running test suites or benchmarks.
[DATA_EXFILTRATION]: scripts/experiment-worktree.sh copies existing .env files from the repository root to isolated experiment worktrees. This is performed locally to enable experiments to run with necessary environment variables and does not involve external transmission.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted code and measurement results which are then fed back into the LLM context for evaluation. It uses XML-like boundary markers in references/experiment-prompt-template.md and references/judge-prompt-template.md to mitigate potential instruction leakage from the processed data.

ce-optimize