Galadriel — Development agent for Claude Code
Fresh-context GATE evaluator for work declared finished.
How to install Galadriel
Installs to ~/.claude/agents/techmefr-mentis-galadriel.md
mkdir -p ~/.claude/agents && curl -fsSL https://raw.githubusercontent.com/techmefr/mentis/HEAD/agents/galadriel.md -o ~/.claude/agents/techmefr-mentis-galadriel.md Restart Claude Code, or start a new session, for it to be picked up.
What Galadriel does
name: galadriel description: Fresh-context GATE evaluator for work declared finished. Binary PASS / NEEDS_WORK, no benefit of the doubt without cited evidence. Invoked as soon as a producer claims it's done. model: opus disallowedTools: Edit, Write, NotebookEdit, Agent effort: xhigh
You are Arbitre, the operator's cold judge. You know nothing of the session that wrote the code: you judge only what you're shown, and nothing is true until it's proven.
**Effort note, 2026-09-07.** `effo
Alternatives in Development
- Claim Verify — Verify that cited claims in a paper accurately represent what the source\ 130 ★
- Change Scoper — Break work into small, trunk-sized tasks with binary DoD and safe rollback 108 ★
- Evaluator Fanout Child — Non-recursive evaluator-fanout leaf — tests one MCP-issued (bug_class × auth) cell, writes durable claim/cover 97 ★
Full documentation available on GitHub
View Source RepositoryRelated Agents
Harness Inbounds Checker
Fresh-context invariant enforcer for the benchmarking climb loop. Diffs a proposed variant against the declare
Vouch Verifier
Read-only verifier one tier below the producer. Checks a worker's CLAIM lines against real evidence and return
Yalla Reviewer
Adversarial code reviewer for Yalla Coding Team. Reviews code in fresh context with no anchoring bias. Uses bi
Loop Judge
Fresh-context cold-read judge for subjective acceptance criteria (readability, API clarity). Runs after mechan
Review Result
Checks a finished quest's or chore's result against its Done when before the creator reads it. Dispatched by t
Paper Accuracy
Review academic papers for factual accuracy. Verifies all numbers, statistics, counts, and quantitative claims
Related Skills
Claude Plugin
Personal skills for Codex and Claude Code, invoked as superb:. craft turns a vague product idea into a decisio
Install Audit
Audit a fresh install — plugin, declared dependencies, and what the human still has to do — before there is an
Weakmap
Priority-ranked weakness report. No arg → fresh report from latest errors per pattern. With concept arg → patc