Supakit kitjanabumrungsak

Turn corrections and proven techniques into durable, versioned improvements to your existing skills.

Keep API keys, tokens and passwords out of repos and logs, add CI guardrails, and rotate first when one leaks.

Implement Clean Architecture and hexagonal patterns to decouple business logic from frameworks and databases.

Build and structure MCP servers with correct transport, tool schemas, and host-specific configurations.

Detect, reproduce, and fix race conditions using atomic operations, idempotency, and proper locking strategies.

Technical framework for choosing local LLMs, sizing quantization to VRAM, and weighing QLoRA vs RAG.

Enforce UTC-based storage and IANA-compliant timezone handling to eliminate DST bugs and production offset errors.

Enforce codebase and API state verification using grep and data probes to eliminate assumptions and hallucinations.

Deploy zero-downtime database migrations using the expand-contract pattern for safe production schema changes.

Enforce Conventional Commits 1.0 standards for git history, PR titles, and automated changelogs.

A proactive task register that prevents deferred work and abandoned branches from being silently dropped.

Transform flat plans into orchestrated task graphs with mandatory test-driven implementation and root-cause debugging.

Enforce a single source of truth by preventing duplicate tracking, TODO, and checklist documents in your repository.

Enforce additive-only changes and dormant code annotations to prevent regressions during refactors and migrations.

Tune reusable prompts and workflows using systematic held-out evaluation to improve accuracy, latency, and cost.

Navigate complex solution spaces using branch-and-explore, tree search, and debate patterns to find the best output.

Implement adaptive planning loops that decompose goals, handle mid-execution changes, and satisfy complex constraints.

Implements durable cross-session learning through structured recall, reflection, and lesson compression loops.

Iterative quality enforcement for LLM outputs using structured critique-rewrite patterns and numeric scoring gates.

A task router that selects the optimal iterative pattern and generates the required loop configuration.

A data-driven prompt optimization loop that scores, versions, and evolves prompts against test sets.

Enforce quality gates using single-critic, multi-lens, adversarial, or ensemble review patterns via loop-ledger.

bantamkit memory

by Supakit kitjanabumrungsak

Free

Your coding agent stops forgetting: local, offline memory it recalls before it acts. Free, MIT.