Tessl · SDD Observatory

Core approach

The bet is the most ambitious rung of SDD: specs, not code, persist as the primary artifact. Specs are Markdown files with a .spec.md extension and YAML frontmatter targeting the files they describe, each requirement tied to a verifying test via [@test] links. The closed-beta Tessl Framework generates code from specs (tessl build) and specs from existing code (tessl document); the open-beta registry distributes versioned, evaluated bundles of skills, rules, and docs that tessl install places in .tessl/ as context for any MCP-compatible agent. Since January 2026 the company brands the platform as an "Agent Enablement Platform" and the registry as "the package manager for agent skills".

Workflow

tessl init — create tessl.json and auto-configure detected coding agents for the Tessl MCP server
tessl search / tessl install — pull skills, rules, and docs from the registry into .tessl/
SDD plugin (tessl-labs/spec-driven-development) — requirement gathering one question at a time, .spec.md specs written into specs/, a stakeholder-approval pause, then implementation with [@test]-linked tests and a drift-catching work review
Framework (closed beta) — tessl create a spec from a prompt, tessl build to generate implementation and tests from it, tessl document to reverse-engineer specs from existing code
tessl skill review / tessl eval run / tessl publish — review, eval, and publish reusable agent context back to the registry

Supported tools

Claude Code
Cursor
Codex
Gemini CLI
GitHub Copilot
Any MCP-compatible agent

Strengths

The registry is a genuine differentiator — no other SDD tool ships a versioned package registry for agent context with review evals, task evals, and per-skill security scores (via a Snyk partnership)
Agent-agnostic by design: MCP-based integration rather than a captive IDE, with plain-Markdown .spec.md files readable outside the ecosystem
Tests are wired into specs — the [@test] link mechanism ties each requirement to a verifying test, a more concrete verification story than most SDD tooling
Best-capitalized dedicated player in the space ($125M from Index, Accel, GV, and boldstart), with heavy ecosystem investment: the AI Native Dev podcast and community, an open tool landscape, and the DevCon conference series

Limitations

The core spec-as-source engine has never publicly shipped — the Framework has been in closed beta since September 2025, and independent hands-on review found the models often ignored the specs
Closed source at the core: the CLI and platform are proprietary; only peripheral plugins and GitHub Actions are open
Strategic drift is real — in under 18 months the pitch moved from AI that writes and maintains code, to spec registry plus framework, to package manager for agent skills
Everything is still beta and pre-1.0, with fast terminology churn (tiles became plugins mid-2026) and credit-based pricing whose unit economics aren't published

Notes

Tessl occupies the most ambitious rung of the SDD ladder: in Birgitta Böckeler’s taxonomy on martinfowler.com it is the only tool of the Kiro / Spec Kit / Tessl trio that aspires to “spec-anchored” and ultimately spec-as-source development, where code is regenerated from persistent specs rather than merely guided by them. The observatory-relevant irony is that this rung remains unproven even at its best-funded proponent: roughly two years after a $125M raise, the Framework is still in closed beta, the shipped public product has pivoted to being a package manager for agent skills, and spec-driven development itself now ships as one MIT-licensed plugin among many on Tessl’s own registry. What to watch is whether the spec-as-source engine ever reaches general availability — and whether the registry’s eval-and-security-score model becomes the distribution standard for agent context regardless.

Added Jul 18, 2026 · Assessment last reviewed Jul 18, 2026 · How we track