AgentSkillsCN

optimize

利用CodeEvolve加速Python函数的运行速度。当用户希望优化、提速或提升Python函数性能时使用。

SKILL.md
--- frontmatter
name: optimize
description: Evolve a Python function to be faster using CodeEvolve. Use when the user wants to optimize, speed up, or improve the performance of a Python function.
disable-model-invocation: true
argument-hint: <function-file> <test-file> [--iterations N] [--function <name>] [--yes]

Optimize

Run CodeEvolve to evolve a Python function to be faster.

The user's message contains the arguments after /codeevolve:optimize. Parse them:

  • •First path argument = function file (Python file containing the function to optimize)
  • •Second path argument = test file (pytest-compatible test file)
  • •--objective time (only supported value in v1.0; if user passes space or balanced, tell them these are deferred to v2.0 and proceed with time)
  • •--iterations N (default: 20)
  • •--function <name> (only needed if the file has multiple top-level functions)
  • •--yes (skip the confirmation prompt)

If the user did not provide both a function file and a test file, ask for the missing ones before proceeding.

Step 1: Preflight check

Run with the arguments the user provided (substitute actual values for FUNCTION_FILE, TEST_FILE, etc.):

bash
python codeevolve/scripts/preflight.py FUNCTION_FILE TEST_FILE --objective OBJECTIVE --iterations ITERATIONS

Show the full output to the user. If the script exits non-zero, stop and report the error. Do not proceed to Step 2.

Extract the JSON blob printed between the lines __PREFLIGHT_JSON__ and __PREFLIGHT_JSON_END__ — you will pass it to the next step.

If the output contains "WARNING: No test cases extracted", warn the user that the correctness gate will be disabled and ask whether to proceed or edit the evaluator first.

Step 2: Build the run harness

Generate a run ID using the current date and time in YYYYMMDD_HHMMSS format. Then run:

bash
python codeevolve/scripts/build_harness.py FUNCTION_FILE TEST_FILE --run-id RUN_ID --objective OBJECTIVE --iterations ITERATIONS --preflight-json 'PREFLIGHT_JSON'

Show the output. Extract the run directory path printed between __RUN_DIR__ and __RUN_DIR_END__.

Step 3: Run evolution

bash
python codeevolve/scripts/run_evolution.py --run-dir RUN_DIR

Tell the user evolution is running and will take ~2–6 minutes. Show the output as it arrives. Extract the result JSON printed between __RESULT_JSON__ and __RESULT_JSON_END__.

Step 4: Present results

If avg_speedup > 1.1 and correctness == 1.0:

Show a clear summary (speedup, iterations, time, cost). Read RUN_DIR/results/best_program.py and show it. Show a diff against the original function. Ask the user:

  • •Accept — replace the function in the original file in-place (use AST to edit only the function, not the whole file)
  • •Reject — discard, keep original
  • •Iterate — run more iterations (ask how many, then repeat Step 3)

If avg_speedup <= 1.1:

Report no significant improvement found. Suggest trying --iterations 50.

If status is "failed" or "no_valid_programs":

Read the last 20 lines of RUN_DIR/results/evolution.log and report the error. Common causes: empty TEST_CASES, evaluator timeout, import errors in evolved code.