Records evidence-backed go, stop, branch, and finalize decisions for research quests.
npx skills add https://github.com/OpenLAIR/dr-claw --skill ds-decision
Please help me install this Agent Skill. Skill: ds-decision Source: https://github.com/OpenLAIR/dr-claw/tree/main/skills/ds-decision Command: npx skills add https://github.com/OpenLAIR/dr-claw --skill ds-decision
Long-running research quests often stall or drift when the next step is unclear. This Skill enforces a disciplined decision process so every continuation, branch, baseline reuse, or stop is backed by durable evidence rather than momentum.
Ask the agent to evaluate the current experiment evidence and record a decision on whether to continue, branch, or stop the research line.
High-intent search queries and answers about installing and using this skill.
State the question explicitly, collect decision-relevant evidence such as run artifacts and baseline state, then choose a verdict (good, bad, neutral, blocked) mapped to an action like continue, iterate, or stop. Record the decision durably with reasons and evidence paths.
Compare candidates against explicit criteria such as evidence quality, feasibility, and expected information gain. Record the winner, the rejected alternatives, implementation priority, and what you expect to learn from the chosen package.
Block only when multiple options are plausible and the choice depends on user preference, cost, or scope, and the runtime decision policy is not autonomous. Otherwise decide from durable evidence, record the reasoning, and continue.
Decisions should rely on durable sources: recent run artifacts, report artifacts, baseline state, and quest documents. Memory serves only as supporting context, and major decisions should never be made from vibe or momentum.
Choose baseline reuse when blocked states or existing results make a new run redundant. The decision is complete only after the baseline is attached and confirmed, or an explicit blocker or waiver is recorded.
ResearAIAudit existing baselines, results, and drafts to classify trusted assets and gaps.
louzhengshuaiAutomate follow-up research analysis campaigns with bounded resource allocation.
louzhengshuaiManage optimization frontiers and candidate briefs in algorithm-first research quests.
dogekikiEvaluate experimental results against research claims using deterministic and generative pipelines.
