tinyspec
tinyspec is a small framework for writing specs that language models implement. It integrates with Claude Code and gives a structured loop: create a spec, refine it with Claude, then work through the implementation plan. It sits at the lightweight end of spec-driven-development: one Markdown file per feature, one directory added to the project, one binary.
Repository snapshot (GitHub API, 2026-11-20)
| Item | Detail |
|---|---|
| Repo | nmcdaines/tinyspec, Rust, MIT |
| Created | 2026-02-16 |
| Activity | last push 2026-03-18; latest release v0.0.9 (2026-02-19) |
| Size | 7 stars, 2 forks |
| Install | cargo install tinyspec (crates.io) or build from source |
A very young, very small project. Treat it as a pattern to borrow or try, not an established ecosystem.
How it works (README and tinyspec.dev)
- Init:
tinyspec initinstalls three Claude Code skills in.claude/skills/and sets up shell completions:/tinyspec:refine,/tinyspec:doand/tinyspec:task.tinyspec init --forceupdates them after an upgrade. - Create:
tinyspec new my-featurewrites a timestamped Markdown file in.specs/with sections for Background, Proposal, Implementation Plan and Test Plan.group/name(for examplev1/my-feature) puts it in a one-level group; names are unique across groups. - Refine:
/tinyspec:refine my-featurehas Claude read the spec, ask clarifying questions and build the plan as task groups (A, B, C) with subtasks (A.1, A.2). - Implement:
/tinyspec:do my-featureworks through each group in order, ticks subtasks off and commits after each group./tinyspec:task my-feature A.1does one task. - Track:
tinyspec statusandtinyspec dashboard, a terminal dashboard that watches.specs/and shows specs sorted by status with completion percentages. - Amend: re-run refine to change the plan as requirements move, then run do again.
Other features: multi-repo specs (application names in the spec front matter resolve to folders through ~/.tinyspec/config.yaml, managed with tinyspec config set|list|remove), and Markdown templates with variable substitution (a repo-level .specs/templates/ takes precedence over the user-level one). Specs stay readable and editable without AI.
Comparison with other spec-driven tools
| tinyspec | Spec Kit | OpenSpec | BMad Method | GSD Core | Kiro | |
|---|---|---|---|---|---|---|
| Licence | MIT | MIT | MIT | MIT | MIT | Product from AWS |
| Written in | Rust, single binary | Python 3.11+ and uv | TypeScript, Node.js 22.20+ | Python and Node.js, installed as skills | JavaScript, Node.js installer | IDE, CLI and web |
| Stars (2026-11-20) | 7 | about 139.6k (2026-09-30) | 75,315 | 55,568 | 11,720 (original repo 64,126) | n/a |
| Artefacts per feature | One Markdown file | Several: constitution, spec, plan, tasks | Four per change: proposal.md, specs/, design.md, tasks.md | Briefs, specifications and architecture; the README names no files | STATE.md, CONTEXT.md and per-phase plans | Three: requirements.md (or bugfix.md), design.md, tasks.md |
| Flow | Refine, then do | Constitution, specify, plan, tasks, implement, converge | Explore, propose, apply, archive | Clarify, plan, build and verify, learn and adjust | Discuss, plan, execute, verify, ship | Requirements, design, tasks |
| Governance or roles | None | Constitution of project principles | Specs archived and merged per change | Specialised product, architecture, UX, dev and test perspectives | Fresh-context subagents per step | Not covered here |
| Agent support | Claude Code skills | Many agents; Copilot is the default | 30+ tools | Tools that support skills | Claude Code, Codex, Copilot, Cursor, OpenCode and more | Kiro itself |
| Scope | One feature, one checklist | Formal stages per feature | Per-change specs with an archive and a living spec set | Whole delivery loop, with modules for testing, games and more | Milestones and phases with context management | IDE-integrated specs |
| Human-first | Plain Markdown, usable without AI | Not stated | Plain Markdown scenarios you review before coding | Decisions kept explicit for the human | Verify step walks through what was built | IDE-driven |
How to read the table: figures come from each project’s own README and GitHub API on 2026-11-20, except Spec Kit and Kiro, which come from this vault’s notes (Spec Kit, Kiro, spec-driven-development). Stars measure popularity, not quality. tinyspec.dev describes the other tools as heavier than itself (OpenSpec’s four files, Spec Kit’s governance and Python dependency, BMad’s many files); the table records what their own pages say instead.
Where it fits
- Choose tinyspec for a solo developer or small team on Claude Code who wants a readable checklist per feature with little ceremony and a progress view.
- Choose Spec Kit when a team wants shared principles, several agents and formal stages.
- Choose OpenSpec for per-change specs with an archive step and wide tool support.
- Choose BMad Method for a whole-delivery method with roles and modules, or GSD Core when context rot across long sessions is the main problem.
- Choose Kiro when the workflow should live inside an IDE.
- No benchmark or independent review compares these tools, so which one gives better results is not established.
Same name, different project
A Spec Kit community extension is also called “TinySpec” (Quratulain-bilal/spec-kit-tinyspec). It adds a lighter workflow for small changes inside Spec Kit and is a different project from nmcdaines/tinyspec. Its page was not opened.
Sources
- nmcdaines/tinyspec README and GitHub API (2026-11-20)
- tinyspec.dev (2026-11-20)
- Spec Kit community: TinySpec extension listing (seen as a search result)