skills.rest
Lingjie-wang
Lingjie-wangCOMMUNITY
Explore all skills by this author →
WebsiteGitHub

rl-task-clarifier

Clarify vague reinforcement-learning requests into execution-ready AutoRL task cards.

Updated Jul 9, 2026
Search Tags:#requirements gathering#reinforcement learning#task clarification#ambiguity scoring#task card#rl environment#autorl
One-click install
npx skills add https://github.com/Lingjie-wang/autoRL --skill rl-task-clarifier-lingjie-wang
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: rl-task-clarifier
Source: https://github.com/Lingjie-wang/autoRL/tree/main/skills/rl-task-clarifier
Command: npx skills add https://github.com/Lingjie-wang/autoRL --skill rl-task-clarifier-lingjie-wang

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve? Reinforcement-learning requests often arrive underspecified, missing the environment, success metric, runtime permission, or algorithm direction, which makes downstream training or evaluation unsafe to execute. This Skill iteratively asks prioritized questions until ambiguity drops below a defined gate, then emits a structured task card an executor can act on without guessing. ## Core Features & Use Cases - Ambiguity Scoring and Gating: Computes a weighted ambiguity score from a field schema and blocks handoff until the score is 0.15 or lower and all mandatory fields are resolved. - Choice-Style Question Batches: Asks at most 5 high-impact questions per turn using interactive choice prompts or numbered Markdown lists, with recommended options first. - Structured Task Card Handoff: Produces a Markdown task_card.md (plus optional JSON mirror and clarification log) covering environment spec, algorithm direction, success criteria, runtime constraints, assumptions, and blocked fields. - Use Case: A user says "train an RL agent on my robot arm simulator." The Skill asks about task mode, environment source, success metric, and runtime boundary, then outputs a ready task card specifying the simulator adapter, SAC as the algorithm direction, and a dry-run execution boundary. ## Quick Start Use the rl-task-clarifier skill to turn my vague RL training request into an execution-ready task card by asking me clarifying questions.

Frequently Asked Questions about rl-task-clarifier

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I clarify a vague reinforcement learning task before training?▼

Identify the blocking fields first: task mode, environment type and path, RL task behavior, algorithm direction, success metric, and execution boundary. Ask at most 5 high-impact questions per turn, then compute an ambiguity score and iterate until it reaches 0.15 or lower.

What information is required before running an RL training job?▼

Mandatory fields include user goal, task mode, execution boundary, environment type and identifier, reward signal, primary metric, success threshold, evaluation protocol, algorithm direction, compute, and dependency policy. Missing any of these blocks safe downstream execution.

How is the ambiguity score for an RL task calculated?▼

Each intake field carries a weight from the clarification schema, totaling 134. Divide the sum of unresolved field weights by the total weight. Scores of 0.00-0.15 pass the gate, 0.16-0.30 require explicit assumptions, and above 0.30 means keep clarifying.

Can the clarifier choose RL algorithms or write training code?▼

No. The Skill owns only intake and clarification. It must not retrieve papers, select algorithms, write training code, or start experiments; those belong to downstream AutoRL stages that consume the emitted task card.

What happens if the user cannot answer the clarifying questions?▼

After 3 unanswered rounds, the Skill produces a task card with explicit blocked_fields, recorded assumptions, and handoff_status set to blocked_or_assumption_based, rather than silently filling in defaults for high-risk fields.

When should defaults not be used for RL task fields?▼

Never silently default execution boundary, task mode, environment type, RL task, algorithm direction, success metric and threshold, custom reward definitions, offline RL datasets, or real-robot and paid cloud execution. These require explicit user confirmation.

Related Skills

Explore More Education & Research Skills→
gpsandhu23gpsandhu23

cross-tool-action-items

Detect untracked commitments and action items across email, Slack, and issue trackers.

Community
Intermediate
Ritabrata-ChakrabortyRitabrata-Chakraborty

stable-baselines3

Train single-agent reinforcement learning models in Gymnasium environments with PyTorch.

Community
Advanced
yihangchen1205yihangchen1205

mdp-designer

Design and edit MDP terms and wire them into configs and logging.

Community
Advanced
ArogyaReddyArogyaReddy

llm-evaluation

Automates LLM output evaluation with metrics, human feedback, and benchmarking.

Community
Advanced