skills.rest
LinkDry
LinkDryCOMMUNITY
Explore all skills by this author →
GitHub

drift-detector

Detect drift between experiment evidence and locked anchors using analysis-report.json and config-snapshot.json.

1|Updated Mar 23, 2026
Search Tags:#analysis#evaluation#drift-detection#drift#experiment#anchor
One-click install
npx skills add https://github.com/LinkDry/Link-Research --skill drift-detector
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: drift-detector
Source: https://github.com/LinkDry/Link-Research/tree/main/skills/drift-detector
Command: npx skills add https://github.com/LinkDry/Link-Research --skill drift-detector

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Checks whether the latest experiment evidence still matches the locked anchor, preventing the system from drifting between analysis results and judge verdict.

Core Features & Use Cases

  • Guards the transition from analysis to judgment by validating anchor consistency against current results.
  • Reads canonical artifacts such as anchor files, experiment-memory.md, results.tsv, analysis-report.json, and config-snapshot.json to assess drift.
  • Provides clear next steps when drift is detected, including review, correction, or archival actions.

Quick Start

Run drift-detector after binding an anchor and recording results to confirm alignment before moving to judgment.

Frequently Asked Questions about drift-detector

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

FAQPage Schema
What is experiment drift detection and why do I need it before judging results?▼

Run drift detection after binding an anchor and recording results to confirm alignment before judgment. It evaluates experiment-memory.md and results.tsv against the anchor to verify evidence consistency before proceeding to the verdict phase.

How do I check if my experiment evidence still tests the locked anchor?▼

Run drift detection after binding an anchor and recording results to confirm alignment before judgment. It evaluates experiment-memory.md and results.tsv against the anchor to verify evidence consistency before proceeding to the verdict phase.

What files do I need to set up before running drift detection on my experiment?▼

Yes, drift detection handles cases where evidence diverges from the anchor by providing clear next steps. It guides you through review, correction, or archival actions, and updates STATE.md and experiment-memory.md to maintain canonical steering.

What happens when drift is detected between my analysis report and anchor?▼

Yes, drift detection handles cases where evidence diverges from the anchor by providing clear next steps. It guides you through review, correction, or archival actions, and updates STATE.md and experiment-memory.md to maintain canonical steering.

Can I use drift detection if my workflow only has a config snapshot and no analysis report?▼

Drift detection differs from standard evaluation by validating anchor consistency against current results rather than just analyzing data. It guards the transition from analysis to judgment, preventing the system from drifting between analysis results and judge verdicts.

Related Skills

Explore More Data & Analytics Skills→
LinkDryLinkDry
1

judge

Issue a Phase 1 verdict from experimental evidence and state files.

Community
Advanced
LinkDryLinkDry
1

analyze-results

Generate analysis-report.json from the latest result group for drift review.

Community
Intermediate
DNYoussefDNYoussef
4

when-reviewing-code-comprehensively-use-code-review-assistant

Coordinate multi-agent pull request reviews across security, performance, style, testing, and documentation.

Community
Advanced
skyllwtskyllwt
1.6k

/exp-pilot-run

Execute pilot experiments from Pilot Spec YAML and return raw metrics.

Community
Advanced