mirror of
https://github.com/microsoft/SkillOpt.git
synced 2026-07-03 14:02:58 +08:00
e4ea6a6771
Highlights since v0.1.0: - feat: SkillOpt-Sleep engine — nightly offline self-evolution (harvest -> mine -> replay -> consolidate behind a validation gate), with multi-objective reward, experience replay + dream rollouts, slow-update long-term memory, and secret redaction in cycle diagnostics. Shipped as the `skillopt-sleep` CLI. - feat: cross-tool backends & plugin shells — Claude, Codex (+Desktop harvest), Copilot, Devin, and OpenClaw. - feat: SearchQA split materialization + rollout fail-fast. - fix: Windows robustness for claude/codex backends, hardened JSON fallback, Qwen timeout/thinking gating, Codex failure surfacing. Packaging: - Bump pyproject / skillopt / skillopt_sleep to 0.2.0. - Restore skillopt_webui to the packaged wheel. See CHANGELOG.md for the full changelog and contributor acknowledgements. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
29 lines
728 B
Python
29 lines
728 B
Python
"""ReflACT: Reflective Agent Tuning.
|
|
|
|
A general-purpose framework for iteratively optimizing LLM agent skills
|
|
through structured reflection and self-improvement.
|
|
|
|
Pipeline stages:
|
|
1. Rollout — execute episodes with current skill
|
|
2. Reflect — analyze trajectories, generate patches
|
|
3. Aggregate — hierarchical merge of patches
|
|
4. Select — rank and select top edits
|
|
5. Update — apply edits to skill document
|
|
6. Evaluate — validate candidate skill, accept/reject
|
|
"""
|
|
|
|
__version__ = "0.2.0"
|
|
|
|
from skillopt.types import ( # noqa: F401
|
|
BatchSpec,
|
|
Edit,
|
|
EditOp,
|
|
FailureSummaryEntry,
|
|
GateAction,
|
|
GateResult,
|
|
Patch,
|
|
RawPatch,
|
|
RolloutResult,
|
|
SlowUpdateResult,
|
|
)
|