microsoft/SkillOpt

ALFWorld Embodied Agent guides agents in the ALFWorld text environment to complete household tasks by navigating rooms, interacting with objects, and using appliances. Use it when an agent needs to choose admissible actions for pick-and-place, clean, heat, cool, or examine tasks.

DocVQA guides an agent to extract exact, evidence-backed answers from document images, tables, forms, and handwriting. Use it when answering visual document questions where precision and exact text spans matter.

Document Rewriter rewrites full AI agent skill documents to incorporate selected revise suggestions from trajectory analysis while preserving scope and protected blocks. Use it when refining agent skills with consolidated feedback and clarity improvements.

Live Mathematical MCQ Heuristics guides models through mathematical multiple-choice questions by comparing options, checking theorem-level precision, and verifying hypotheses. Use it when answering rigorous math MCQs where distractors differ by strength, quantifiers, or exact conditions.

OfficeQA retrieves and verifies numerical answers from official tables and time series, applying strict calculation, unit, and evidence checks before final answers.

Question Answering normalizes short-answer trivia responses by applying rules for concise answer form, evidence matching, and clue interpretation. Use it when an agent must produce accurate, well-formed answers for QA or crossword-style questions.

Skill Optimization Coach writes compact optimizer-side meta skills that guide future agent skill edits using evidence from adjacent-epoch comparisons. It identifies which editing directions help or hurt in self-evolving agent environments.

SkillOpt Sleep runs a periodic self-improvement cycle for Claude agents, harvesting past sessions, replaying recurring tasks, and consolidating improvements into memory and skills behind a held-out gate.

SkillOpt Sleep gives a local Codex agent a sleep cycle to review past sessions, mine recurring tasks, replay them through a backend, and consolidate validated improvements into skills and memory. Trigger it for nightly/offline self-improvement, status/harvest/dry-run/run/adopt actions, or scheduling background optimization.

SkillOpt Sleep reviews recent local Cursor sessions, mines recurring tasks, replays them, and proposes bounded improvements to a project Cursor skill. Use it when asked for an offline sleep/dream cycle, skill consolidation, or SkillOpt-Sleep status, harvest, dry-run, run, scheduling, review, or adoption.

SkillOpt Sleep is a reference-only OpenClaw adaptation of the shared SkillOpt-Sleep cycle, contributed to study or port a DeepSeek backend wrapper. It is not a ready-to-run installation; use it as source material for a future integration.

Spreadsheet Manipulation guides agents in editing Excel .xlsx files with Python, using openpyxl and pandas to preserve formulas and formatting. Use it when an agent needs to read, transform, or write spreadsheet cells without breaking existing structure.