Awesome

Patterns and techniques for evaluating and improving AI agent outputs. Use this skill when: - Implementing self-critique and reflection loops - Building evaluator-optimizer pipelines for quality-critical generation - Creating test-driven code refinement workflows - Designing rubric-based or LLM-as-judge evaluation systems - Adding iterative improvement to agent outputs (code, reports, analysis) - Measuring and improving agent response quality

Review legacy SQL and mapper queries for correctness, performance, binding safety, pagination issues, and maintainability.

Maintain domain-specific timeline, entity, release, and version views so operators can understand work across software, narrative, and knowledge domains.

Find localization or message resource IDs, trace where they are used, and check for reuse before introducing new strings.

Trace service identifiers, URLs, controllers, services, mappers, and call chains across layered or legacy systems.

Run the final quality gate after implementation. Executes multi-expert inspection, remediation, verification, and governance closure before a task is considered done.

Create, repair, and refresh governance artifacts for AI-assisted delivery. Use for harness manifests, context ledgers, review notes, execution plans, handovers, and UTF-8-safe documentation updates.

Safely upgrades legacy or older initialized workspaces to the latest managed harness structure with dry-run, backup, restore, and review discipline.

Run a structured multi-expert review with evaluation criteria, context isolation, three review rounds, convergence, and governance-ready outputs.

Run a disciplined code review pipeline after each implementation chunk. Includes criteria extraction, automated checks, third-party review, test coverage verification, and remediation loops.

Analyze session memories, work logs, governance traces, and repeated user requests to identify stable recurring workflows. Use when a harness workspace should turn repeated work into a reusable skill, agent, checklist, or governance playbook without overfitting to one session.

Create durable work logs, final records, AS-IS vs TO-BE summaries, and flow documents that keep long-running AI work resumable and auditable.

Maintain the Harness Dashboard 4.6 Hypertext Project World Model, ledger-backed projections, briefs, dictionary, VCS evidence, and listener state without using a database.

Prepares governed handoff bundles and portable launch guidance for CLI-based agents across local, sandboxed, and vendor-specific runtimes.

Orchestrate long-running implementation with governance-first planning, three plan-review cycles, resumable chunking, targeted delegation, verification, and handover discipline.