guild-evolve-skill

Automate guild skill evolution from snapshot to promotion with non-destructive versioning.

6|Updated Apr 24, 2026
One-click install
npx skills add https://github.com/lookatitude/guild --skill guild-evolve-skill
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: guild-evolve-skill
Source: https://github.com/lookatitude/guild/tree/main/skills/meta/evolve-skill
Command: npx skills add https://github.com/lookatitude/guild --skill guild-evolve-skill

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Automates end-to-end evolution of an existing guild skill, providing a structured, auditable workflow from snapshot to promotion with non-destructive changes.

Core Features & Use Cases

  • Snapshot current skill to a versioned store.
  • Load evals.json and bootstrap from reflections if needed to reach evaluation thresholds.
  • Dispatch paired subagents (A = current, B = proposed) for each skill case.
  • Drafter writes per-case assertions to guide evaluation.
  • Grader evaluates trajectories and assigns per-case results.
  • Benchmark + flip report computes metrics and summarizes differences.
  • Shadow mode runs the proposed change against historical tasks without affecting live routing.
  • Promotion gate enforces policy and records the outcome.
  • On promote: description optimizer runs and commits the change to the live skill, updating versions.
  • On reject: archives the attempt and preserves the original skill.

Quick Start

Trigger the evolve workflow on a target skill by specifying the skill slug and an optional proposed description.

Frequently Asked Questions about guild-evolve-skill

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

FAQPage Schema
How do I automate workflow versioning for skill evaluation and promotion?

Workflow versioning for skill evaluation is automated by running a 10-step pipeline that snapshots current skills, dispatches paired subagents for grading, and only updates live versions after passing a promotion gate. Rejected attempts are archived non-destructively.

What is the best way to test proposed skill changes without affecting live routing?

Testing proposed skill changes without affecting live routing is achieved using shadow mode, which runs the proposed modifications against historical tasks to validate performance before any live traffic is impacted.

How does shadow mode evaluate proposed changes against historical tasks?

Shadow mode evaluates proposed changes by running them against historical task data, computing benchmark metrics and flip reports to summarize performance differences without updating the live skill version.

Can I bootstrap skill evaluations from reflections if evals thresholds are not met?

Skill evaluations can be bootstrapped from reflections to reach required thresholds by loading evals.json, allowing the workflow to proceed with paired subagent dispatching and trajectory grading even with insufficient initial data.

What happens when a skill evolution attempt is rejected at the promotion gate?

When a skill evolution attempt is rejected at the promotion gate, the workflow archives the failed attempt and preserves the original live skill without applying any destructive changes to the active version.

Does the skill evolution pipeline support non-destructive versioning for rejected attempts?

The skill evolution pipeline enforces non-destructive versioning by storing changes in .guild paths, archiving rejected attempts, and only updating the live skill after a proposed change successfully passes the promotion gate.