Sambhav Surana
Community@Sambhav242005 · India
Sambhav Surana publishes 39 skills spanning LLM context engineering, multi-agent system design, token compression modes, code minimalism, and Prisma ORM v7 migration and operations.
Agent Skills by Sambhav Surana
Showing 39 vetted skills indexed across 1 GitHub repositories.
Sambhav242005caveman-help
Displays a quick-reference card of caveman modes, skills, and configuration options.
Sambhav242005caveman-review
Generates one-line code review comments with location, severity, problem, and fix.
Sambhav242005caveman-compress
Compress natural language memory files into terse caveman format to reduce input tokens.
Sambhav242005cavecrew
Guides delegation to compressed-output subagents for code location, editing, and review tasks.
Sambhav242005caveman-stats
Reports actual session token usage and estimated savings from the Claude Code session log.
Sambhav242005caveman
Compresses AI responses into terse caveman-style prose while preserving technical accuracy.
Sambhav242005caveman-commit
Generates terse Conventional Commits messages from staged changes with imperative subjects and optional bodies.
Sambhav242005multi-agent-patterns
Design multi-agent systems with supervisor, swarm, and hierarchical coordination patterns.
Sambhav242005bdi-mental-states
Transform RDF context into BDI beliefs, desires, and intentions with ontology patterns.
Sambhav242005context-degradation
Diagnose and mitigate context degradation patterns in LLM agent systems.
Sambhav242005context-compression
Compress long agent conversation histories into structured summaries preserving files, decisions, and next steps.
Sambhav242005memory-systems
Design persistent semantic memory architectures for agents using vector stores, knowledge graphs, and temporal validity.
Sambhav242005advanced-evaluation
Build LLM-as-judge evaluation systems with bias mitigation, rubrics, and calibrated confidence scoring.
Sambhav242005ponytail
Enforces minimal, standard-library-first solutions for coding tasks with adjustable intensity levels.
Sambhav242005ponytail-audit
Audits an entire codebase for over-engineering and produces a ranked list of deletions and simplifications.
Sambhav242005harness-engineering
Design control surfaces, feedback loops, and governance boundaries for autonomous agent workflows.
Sambhav242005ponytail-help
Displays a quick-reference card of ponytail modes, skills, and commands.
Sambhav242005self-improvement-loops
Designs recursive self-improvement loops where agents mine failures and edit their own harnesses.
Sambhav242005context-fundamentals
Explains context engineering fundamentals including attention mechanics, token budgets, and progressive disclosure.
Sambhav242005latent-briefing
Compact orchestrator trajectories into worker KV caches using Attention Matching for multi-agent memory sharing.
Sambhav242005ponytail-review
Reviews code diffs for over-engineering and lists what to delete or simplify.
Sambhav242005project-development
Plan LLM project architectures, staged pipelines, and cost estimates before writing code.
Sambhav242005evaluation
Build evaluation frameworks with multi-dimensional rubrics, test sets, and production monitoring for agent systems.
Sambhav242005context-optimization
Reduce LLM context token usage through masking, compaction, caching, and partitioning.
Frequently Asked Questions About Sambhav Surana
FAQPage SchemaWhat tasks can I accomplish with Sambhav Surana's skills?▼
You can compress agent context and memory files to save tokens, generate ultra-compact code reviews and commit messages, design multi-agent systems with BDI mental states, build evaluation harnesses with quality gates, audit codebases for over-engineering, and migrate or operate Prisma ORM v7 with Postgres, MongoDB, and Compute deployments.
Who are these skills designed for?▼
They target engineers building LLM-powered agent systems: context engineers managing token budgets and degradation, architects designing multi-agent coordination and harnesses, and backend developers working with Prisma ORM v7, driver adapters, and Prisma Postgres provisioning.
How do the caveman and ponytail modes work in practice?▼
Caveman triggers via /caveman or phrases like 'be brief', compressing output tokens by a measured 65% across lite, full, ultra, and wenyan intensity levels. Ponytail activates on coding tasks to force minimal stdlib-first solutions, with companion skills for repo audits, debt ledgers, and impact scoreboards.
Are these skills open source and what do they cost?▼
The ponytail skill and all nine Prisma skills carry explicit MIT licenses in their frontmatter, making them free to use and modify. Other skills in the manifest do not declare a license, so usage terms for those should be confirmed in the repository before redistribution.
What prerequisites do the Prisma skills require?▼
Prisma skills assume an existing Prisma ORM project, with version-specific guidance for v6-to-v7 upgrades. Prisma Postgres setup uses the Management API with service tokens or OAuth, driver adapter work requires SqlDriverAdapter knowledge, and Compute deployment supports Hono, Next.js, Nuxt, Svelte, and similar frameworks.