Skill Comparator — Development agent for Claude Code
Blind A/B comparator for skill-improver runs.
How to install Skill Comparator
Installs to ~/.claude/agents/air-gapped-skills-skill-comparator.md
mkdir -p ~/.claude/agents && curl -fsSL https://raw.githubusercontent.com/air-gapped/skills/HEAD/.claude/agents/skill-comparator.md -o ~/.claude/agents/air-gapped-skills-skill-comparator.md Restart Claude Code, or start a new session, for it to be picked up.
What Skill Comparator does
name: skill-comparator description: Blind A/B comparator for skill-improver runs. Spawn explicitly with a rubric dir and two unlabelled skill dirs — not for general delegation. tools: Read, Grep, Glob, Bash model: sonnet
Decide which of two Claude Code skill versions is better against the skill-improver quality rubric. This system prompt is the complete, canonical comparison instruction set; the spawn prompt supplies only three paths:
RUBRIC DIR:— skill-improver'sreferences/di
Alternatives in Development
- Engram Assessor — Independent grader of learner productions for the Engram learning plugin 1.4k ★
- Fable Orchestrator — Staged-execution orchestrator for large, multi-part, or multi-session tasks 837 ★
- Finding Checker — Blind adversarial checker — given ONLY a finding artifact and its evidence (never the author's reasoning), tri 348 ★
Full documentation available on GitHub
View Source RepositoryRelated Agents
Blind Scorer
Blind quality scorer for skill-improver runs. Spawn explicitly with a rubric dir and target dir — not for gene
Patch Author
Patch author for one verified vulnerability finding in a patch run. Spawn with a repo path and a nonce-tagged
Comparator
Blind Comparator Agent
Create Worktree Subagent
Use when the user explicitly asks for a SUBAGENT to create a worktree. If the user does not mention "subagent"
Phase Implementer
Invoked by /orchestrate to implement one plan phase end-to-end — read the phase block, write the code, run tar
Mathodology Award Judge
Use for one independent blind judge seat scoring an award-level submission against contest rubric criteria.
Related Skills
05 Verification Panel
<!-- Injected: {{ANGLE}} {{ANGLE_QUESTION}} {{TARGETS_JSON}} {{PAPER_TXT_PATH}} {{STAGED_SOURCES_DIR}} {{QUOTE
Subagents Pydantic AI
Subagent Delegation framework for Pydantic AI, enabling nested subagents that can spawn their own specialists
Delegate Subagent.Skill
Spawn an isolated subagent in a dedicated git worktree by creating a BRIEF.md assignment card and logging the