Ready to supercharge your AI agents?

Install any skill in seconds. Free to start, no credit card required.

Spec-Driven Development (SDD) — measured AI agent skill | DecimalAI

brycewang-stanford/sdd

SkillScore50Maintained

Implements the Spec-Driven Development lifecycle (Intent, Requirements, Design, Tasks, Build) for structured feature development. Use when the user wants to scaffold a new feature spec, generate EARS requirements, create a technical design, break work into tasks, or check spec status. Trigger on keywords: sdd, spec-driven, ears requirements, feature spec.

by brycewang-stanford

Links this skill to your agents and keeps it current with the author’s updates.

Automatic

Coding pack

decimalai skills pull brycewang-stanford-sdd

Get this skill — curl · raw file

curl -fsSL https://app.decimal.ai/s/brycewang-stanford-sdd/SKILL.md

Raw file: /s/brycewang-stanford-sdd/SKILL.md ↗ · machine-readable summary at /s/brycewang-stanford-sdd.json ↗

In productionGlobal

how this skill behaves on real traffic · last 30 days

No production data for this skill yet — usage numbers appear here once agents start running it.

Anonymized aggregate across every org using this skill — proof from real production use, not a demo.

Does this skill help? · Benchmark: with vs. withoutsign in to run this on your agents

22 test cases from the publisher’s eval.yaml · each run twice on gemini-3.6-flash — with the skill injected, then without · last run 6h ago

—+3% tokens0% turns50% of whole cases pass with skill (11/22)

Test case	Without → With	Effect	Δ tokens	Δ turns
case-10	✗→✓	▲ Improved	-30%	0%
case-15	✗→✓	▲ Improved	1%	0%
case-16	✗→✓	▲ Improved	-33%	0%
case-19	✗→✓	▲ Improved	-37%	0%
case-05	✓→✓	＝ Same ✓	67%	0%

Showing 5 of 22 cases

SkillSafety

✓ 3 of 3 passed

Code scan15 of 15 checks clean

0 critical · 0 warning · 0 info — scanner v4 — 8/2/2026

live_secretremote_code_executionreverse_shelldata_exfiltrationssrf_metadataagent_snoopinganti_refusaldestructive_commandsobfuscated_unicodeunicode_homoglyphopaque_blobsuspicious_urltool_overreachtrigger_abuseprompt_injection_phrasing

IntentSafe

LLM judge · does the skill do what it claims

ContentClean

LLM judge · policy & harmful-content review

Source

Imported

brycewang-stanford / Auto-Empirical-Research-Skills

skills/09-meleantonio-awesome-econ-ai-stuff/_skills/engineering/sdd/SKILL.md

View on GitHub ↗

License NOASSERTIONImported 8/2/2026Version v1

Skill contents

SKILL.md

Spec-Driven Development (SDD)

Core Philosophy

Clarity before Code: Never generate code until requirements and design are approved.
Iterative Refinement: Loop through Req → Design → Tasks until solid.
Code via Docs: The truth is in the markdown files, not the chat.

Commands

Execute in the project root (where spec/ and steering/ live).

`init`

Scaffold the SDD folder structure and template files.

If spec/ already exists, skip or ask before overwriting.
Create spec/ and steering/.
Create spec/intent.md (blank or minimal placeholder).
Copy templates from this skill’s templates/ directory (in the same folder as SKILL.md, or from your install path under ~/.cursor/skills/sdd/ after copying the skill there) into the project:

templates/spec/requirements.md → spec/requirements.md
templates/spec/design.md → spec/design.md
templates/spec/tasks.md → spec/tasks.md
templates/steering/coding-standards.md → steering/coding-standards.md

`reqs`

Generate EARS requirements from intent.

Read spec/intent.md and steering/*.md.
Convert the intent into EARS requirements (see EARS Quick Reference below). Add a Properties (invariants) section.
Write to spec/requirements.md.
Ask for user approval before proceeding.

`design`

Generate technical design from requirements.

Read spec/requirements.md and steering/*.md.
Create a technical design: architecture, data models, component interfaces, error handling, security. Apply the Design Checklist below.
Write to spec/design.md.
Ask for user approval before proceeding.

`tasks`

Generate implementation tasks from design.

Read spec/design.md and spec/requirements.md.
Create a sequential task list: max two levels (Task > Subtask). Link each task to requirement IDs (e.g. REQ-001). Follow Task Rules below.
Write to spec/tasks.md.
Ask for user approval before proceeding.

`status`

Report current state of the spec.

List files in spec/ (and optionally steering/).
If spec/tasks.md exists, count unchecked [ ] vs checked [x] and summarize.

EARS Quick Reference

Ubiquitous: <system> shall <response>
Event-Driven: WHEN <trigger> [precondition] the <system> shall <response>
Unwanted: IF <unwanted condition> THEN the <system> shall <response>
State-Driven: WHILE <system state>, the <system> shall <response>
Optional: WHERE <feature is included>, the <system> shall <response>

Use IDs like [REQ-001]; add a Properties (Invariants) section for universal correctness statements.

Design Checklist

Edit ruthlessly (remove over-engineering).
Check for circular dependencies; fix via interface extraction, layering, or events.
Ensure alignment with steering documents.

Task Rules

Two-level hierarchy maximum (Task > Subtask).
Sequential order (each task builds on previous).
Traceability: each task or subtask links back to requirement IDs (e.g. Traceability: Implements REQ-001).

Additional Resources

For full framework detail (workflow, refinement, iteration triggers), see reference.md.

Want to see if brycewang-stanford/sdd would activate on your phrasing? Try it before you install — no LLM call, no signup.

Other measured skills in the registry, with their headline benchmark lift.