tursodatabase/turso
10 skills7.5k installs191k starsGitHub
Install
npx skills add https://github.com/tursodatabase/tursoSkills in this repo
1Code QualityCode Quality is a Turso-aligned agent skill that encodes how to write Rust for a production database where correctness beats convenience. It tells agents and solo builders to reject quick hacks, assert invariants, crash on integrity-threatening invalid state, and use enums and exhaustive matching instead of string sentinels. The guide covers comment hygiene, microsecond-level performance awareness, and concrete before/after examples for conditional logic that must not silently ignore impossible branches. Invoke it whenever you are implementing or refactoring Limbo/Turso-related Rust—not as a one-off linter substitute but as procedural knowledge that keeps agents from shipping “shouldn't happen” else blocks. It pairs naturally with test and review workflows across Build and Ship without replacing formal QA.1.2kinstalls2Index Knowledgeindex-knowledge is a Turso-originated agent skill that turns an unfamiliar codebase into a layered AGENTS.md knowledge base: a root file plus optional subdirectory docs where complexity warrants depth. Solo builders and small teams use it when Claude Code, Cursor, or Codex agents keep missing context because READMEs are thin or stale. The workflow is explicit—fire parallel discovery agents, score directories, generate root then children (in parallel where possible), and run a dedupe/trim review—and it insists on TodoWrite tracking across all phases. Create-new mode removes existing AGENTS.md files and rebuilds from scratch; update mode refreshes and adds only where justified. Max depth caps how far nested docs go. The outcome is durable procedural knowledge embedded in the repo so every future agent session starts closer to the truth about layout, boundaries, and hot spots, which pays off across Build, Ship code review, and Operate maintenance passes.761installs3DebuggingThe Turso debugging skill is procedural knowledge for solo and indie builders maintaining Rust-based TursoDB or apps that rely on SQLite-compatible semantics. It walks you through a structured compatibility workflow: compare EXPLAIN bytecode between SQLite and Turso, then branch to codegen versus virtual-machine or storage defects when results diverge. Manual REPL-style inspection uses the tursodb binary on in-memory or file databases. For flaky concurrency, it documents nightly ThreadSanitizer stress harnesses with thread and iteration flags. For hard-to-reproduce failures, deterministic simulators (limbo_sim and concurrent Whopper) accept seeds so you can replay the same failure ordering. Logging guidance warns that trace-level turso_core logs can reach megabytes per test run—useful but noisy. This is not a generic printf-debugging cheat sheet; it assumes you are already in the repo, running cargo and make targets. Pair it with your test suite and issue reproduction notes before asking an agent to hypothesize root cause.739installs4Async Io ModelAsync I/O Model is a contributor-facing agent skill for anyone hacking on tursodb core where cooperative yielding replaces familiar async/await. It explains IOResult as either Done with a value or IO with completions to finish later, and shows how Completion tracks single operations while CompletionGroup batches many—including nested groups and cancel. Solo builders rarely need this unless they embed Turso's Rust core or fork the engine; for that audience it prevents subtle re-entrancy bugs and wrong polling loops. The skill is prescriptive: always apply these patterns in core when touching I/O. It is not a product integration guide for Turso Cloud hosting or SQL APIs—it is internal architecture documentation compressed for agents writing or reviewing engine code.711installs5Storage Formatstorage-format is a Turso agent skill that teaches the SQLite file format used by libSQL/Turso so solo builders debugging edge-case corruption, migration tooling, or custom storage introspection do not guess page layouts. It walks the on-disk structure from the 16-byte magic through header counters, B-tree roles for tables versus indexes, cell overflow mechanics, and freelist recovery patterns—all with offsets and flag tables suitable for agent-assisted code reviews or low-level feature design. Use when implementing backup validators, page-level tools, teaching agents how Turso differs from generic SQL, or reviewing performance issues tied to page size and btree depth. It is reference knowledge, not a migration generator. Intermediate complexity assumes comfort reading hex layouts and database internals. Pairs naturally with Turso deployment and SQL API skills when you move from schema design to why the bytes on disk look the way they do.709installs6TestingTurso’s testing skill is a maintainer-focused guide for anyone contributing to the database codebase and needing a single source of truth on where tests live and which format to choose. It positions `.sqltest` under `testing/sqltests/tests/` as the default for new SQL compatibility work, while TCL `.test` files remain as legacy material being phased out. Rust integration tests cover regressions and complex scenarios; fuzz targets edge cases. The readme spells out everyday commands—full suite via `make test`, single TCL files, the sqltests runner, and workspace-wide `cargo test`—so solo builders and small teams can reproduce CI locally without guessing paths. For agent-assisted development, it gives concrete templates for expectations and row formatting, which reduces bad tests that pass on one backend only. Install it when you are adding features, fixing SQL compat, or migrating TCL suites—not when you only need Turso as a hosted DB without touching this repo.703installs7Pr Workflowpr-workflow is an agent skill that encodes Turso’s general pull-request and commit playbook for solo and indie builders using Claude Code, Cursor, Codex, or similar tools. It tells the agent how to structure atomic commits with clear intent, avoid mixing unrelated changes, and use interactive rebase when cleaning history. On the PR side, it pushes for focused, test-backed submissions where commits tell a coherent story. For teams running Claude in GitHub Actions, it sets expectations around turn limits, acceptable WIP commits and PRs, and staying on the main task. Security and compliance sections block secret leakage and require a two-step dependency ritual—license file plus NOTICE.md—before new third-party code ships. External API and CLI usage must follow official documentation rather than guessed parameters. Install it when you want repeatable git and PR behavior across build and ship work, not one-off chat advice.697installs8Transaction CorrectnessTransaction Correctness is a Turso-focused reference skill that walks solo builders through Write-Ahead Logging mechanics as implemented in TursoDB—not generic SQLite trivia. It explains how writes append frames to the WAL, how COMMIT is signaled, how readers pin snapshots with read marks, and when checkpoints move data back into the main database file. That matters when you are debugging “lost” writes, unexpected rollbacks, or performance spikes from WAL growth on laptops, CI runners, or small SaaS backends. The guide also clarifies Turso’s deliberate departure from multi-process `-shm` indexing, which changes how you reason about concurrency in agent-assisted refactors. Use it while designing migrations, batch jobs, or multi-connection access patterns, and again in Ship and Operate when incidents smell like race conditions or checkpoint stalls. It does not replace Turso product docs for limits or APIs; it gives you the mental model to ask your coding agent the right questions about transactional safety.696installs9Differential FuzzerDifferential-fuzzer is an agent skill for contributors and maintainers working on Turso who need a repeatable way to hunt SQL correctness bugs. It describes the differential oracle fuzzer under `testing/differential-oracle/fuzzer/`, which executes generated statements against Turso and SQLite and flags divergent results. Solo builders shipping embedded or edge databases benefit when they extend the engine or chase regressions: you get single-run recipes with deterministic seeds, high-volume runs, loop mode for soak testing, and options to retain `.db` artifacts for postmortems. The skill explicitly tells the agent to load the related Debugging skill so failures are interpreted with the project’s triage conventions, not ad-hoc guesses. It is narrowly scoped to operating that tool— not general product QA— and assumes a Rust toolchain and local clone of the Turso repository.686installs10MvccMVCC is a reference skill for solo builders and contributors who need accurate mental models of Turso’s in-progress Multi-Version Concurrency Control mode without mistaking it for stable WAL semantics. It explains how to enable MVCC at runtime with a pragma, how row-level snapshot isolation differs from page-level WAL snapshots, and how logical logging replaces classic WAL files. The guide maps core structures—versioned rows with begin/end timestamps, shared `MvStore`, and checkpoint machinery—so agents do not hallucinate storage paths or isolation guarantees. It also sets guardrails: treat the feature as experimental, and do not drag MVCC into unrelated debugging threads. Use it when reviewing concurrency PRs, designing tests around transactional visibility, or answering whether a reported bug could stem from MVCC-specific versioning rather than generic SQL logic.686installs