Capture Eval — Productivity skill for Claude Code
Snapshot the task just completed into the regression corpus so future harness changes can't silently regress it.
How to install Capture Eval
Installs to ~/.claude/skills/ghostlygawd-recursive-harness-capture-eval/SKILL.md
mkdir -p ~/.claude/skills/ghostlygawd-recursive-harness-capture-eval && curl -fsSL https://raw.githubusercontent.com/GhostlyGawd/recursive-harness/HEAD/commands/capture-eval.md -o ~/.claude/skills/ghostlygawd-recursive-harness-capture-eval/SKILL.md Restart Claude Code, or start a new session, for it to be picked up.
What Capture Eval does
description: Snapshot the task just completed into the regression corpus so future harness changes can't silently regress it.
Follow skill: eval-capture. Concretely, for the task just finished (or the one named in $ARGUMENTS):
- Confirm it qualifies: user accepted the result, AND (recurring shape OR correction-born OR articulated taste). If not, say why and stop. Standing qualifier (roadmap item 8, 2026-07-05): ANY Windows papercut that cost a session (cp1252/console encoding
Alternatives in Productivity
- Maestro Update — Detect version, preview changes, apply workflow upgrades 530 ★
- Browser Harness JS — Self-healing browser harness that enables LLMs to complete any task 480 ★
- MCP Tree-sitter Server Roadmap — This document outlines the planned improvements and future features for the MCP Tree-sitter Server project 288 ★
Full documentation available on GitHub
View Source RepositoryRelated Skills
Skills Learn From Task
Extract reusable, de-identified lessons from a completed authorized task — what worked, what failed, false pos
Coder Eval Review
Generate per-task review.json (summary + tags) for a completed run
Repo Scan
Produce a structured snapshot of the target repository for rapid task analysis.
Obsidian Calendar
One calendar command with four modes - agenda (read a snapshot), reconcile (vault vs calendar gaps), meeting (
Learning Summary
You are analyzing a completed sub-agent coding task to produce a structured learning summary.
Gg Complete
Mark the current GitGud task as completed
Related Agents
Hcs Eval Reviewer
Reviews regression-trap quality and eval harness coverage for HCS. Ensures traps capture real failure classes
Impl Reviewer
Use this agent for Phase 6 (Implementation Review) of the claude-forge. Reviews a completed task's implementat
Keystone Writer
Writes regression-gate keystone tests against real Chromium for runtime / actionable() / BrowserDriver changes