메뉴
DSPy optimization workflows — teleprompters, metrics, evaluation, and compilation strategies. Use when optimizing DSPy programs with BootstrapFewShot, MIPROv2, or custom metrics.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Execute batched or parallel RLM work using Daytona sandboxes. Use when running multiple sub-tasks, parameter sweeps, or batch recursive RLM calls in the daytona_pilot runtime.
Debug fleet-rlm runtime issues from Claude Code. Use when diagnosing daytona_pilot failures, API and websocket contract problems, sandbox persistence bugs, or runtime readiness drift.
Execute Python code in Daytona sandboxes with durable volume persistence. Use when running code in a Daytona sandbox, processing data with stateful execution, or persisting results across sessions.
(EXPERIMENTAL) Research implementation for RLM long-context processing using standalone Python scripts. For production use, prefer the rlm skill which uses the fleet-rlm package with Daytona sandboxes. This skill is for experimentation, evaluation, and alternative implementation patterns.
Long-term memory persistence for RLM using Daytona durable volume storage. Use when storing, recalling, listing, or searching data that persists across sandbox sessions in the daytona_pilot runtime.
Run fleet-rlm through its current public entrypoints. Use when you need the right command for the Web UI, API server, MCP server, terminal chat, or Daytona smoke validation from a Claude Code workflow.
| name | dspy-optimization |
| description | DSPy optimization workflows — teleprompters, metrics, evaluation, and compilation strategies. Use when optimizing DSPy programs with BootstrapFewShot, MIPROv2, or custom metrics. |
Guidance for optimizing DSPy programs: teleprompter selection, metric definition, evaluation workflows, and save/load patterns.
This skill requires uv as the Python package manager.
uv venv
source .venv/bin/activate
uv pip install dspy
Rule: Never use raw
pip. Always useuv pipfor installs anduv runfor script execution.
# Use the optimize-dspy script
uv run scripts/optimize-dspy.py --module my_module --metric my_metric --examples examples.jsonl
from dspy.teleprompt import BootstrapFewShot
optimizer = BootstrapFewShot(metric=my_metric, max_bootstrapped_demos=4)
compiled = optimizer.compile(program, trainset=trainset)
compiled.save("./compiled/program.json")
Use this skill when:
See optimization.md for:
.save() / .load(), never raw pickle)uv: uv pip for installs, uv run for scripts.save() / .load(): Never use raw pickle for DSPy programsdspy.LM: Never use deprecated dspy.OpenAI or dspy.settings.configure.with_inputs() on dspy.Example