# Spec Dev > > - **Type:** Skill - **Install:** `agentstack add skill-harshuljain13-agent-skills-spec-dev` - **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior - **Seller:** [harshuljain13](https://agentstack.voostack.com/s/harshuljain13) - **Installs:** 0 - **Category:** [Agent Skills](https://agentstack.voostack.com/c/agent-skills) - **Latest version:** 0.1.0 - **License:** MIT - **Upstream author:** [harshuljain13](https://github.com/harshuljain13) - **Source:** https://github.com/harshuljain13/agent-skills/tree/main/skills/spec-dev ## Install ```sh agentstack add skill-harshuljain13-agent-skills-spec-dev ``` Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent. ## About # Spec-Driven Development Skill ## The Workflow Think before you build. Every project starts with three documents in sequence. No code until tasks.md is finalized. ``` .spec-dev/ ├── requirements.md ← What problem are we solving and for whom? ├── spec.md ← How will we solve it? (decisions, architecture, scope) └── tasks.md ← What exactly needs to be built, in what order? ``` Once all three are done and Harshul signs off, proposals/plans go into Notion. --- ## Step 1: requirements.md The "what and why" — no solutions yet. ```markdown # Requirements: [Project/Feature Name] ## Problem Statement [1-2 sentences: what is broken or missing, and for whom?] [Be specific. "Engineers waste time X" is better than "the process is inefficient."] ## Goals - [Goal 1 — measurable if possible] - [Goal 2] - [Goal 3] ## Non-Goals (explicitly out of scope) - [Thing we are NOT building] - [Edge case we are NOT handling in v1] ## Success Criteria [How will we know this worked? What metrics, behaviors, or outcomes signal success?] ## Constraints - [Technical constraint — e.g., "must work with existing Redis setup"] - [Time constraint — e.g., "needs to ship before X"] - [Resource constraint] ## Open Questions - [Things we don't know yet that affect the design] ``` --- ## Step 2: spec.md The "how" — decisions made, architecture chosen, scope locked. Only write this after requirements.md is stable. ```markdown # Spec: [Project/Feature Name] Status: Draft | In Review | Finalized Last updated: [date] ## Summary [2-3 sentences: what we're building and the core approach] ## Design Decisions ### Decision 1: [Topic] **Options considered**: [A], [B], [C] **Chosen**: [A] **Why**: [Reasoning — tradeoffs made explicit] ### Decision 2: [Topic] ... ## Architecture / Approach [Describe the system, flow, or implementation approach. Use diagrams if helpful (ASCII or mermaid). Be specific enough that someone else could implement it from this doc.] ## Data Model / Schema (if applicable) [Tables, fields, types, relationships] ## API / Interface Design (if applicable) [Endpoints, inputs, outputs, error states] ## Edge Cases and Error Handling - [Edge case 1 → how handled] - [Edge case 2 → how handled] ## What's Explicitly Not in v1 [Repeat non-goals from requirements + any new ones discovered during design] ## Dependencies - [External service or library relied upon] - [Internal system or team dependency] ## Open Questions (resolved) - [Q from requirements.md] → [Answer] ``` --- ## Step 3: tasks.md The "what gets built, in what order" — atomic, actionable, sequenced. Only write this after spec.md is finalized. ```markdown # Tasks: [Project/Feature Name] ## Status [ ] Not started [x] In progress [✓] Done ## Phase 1: [Phase name — e.g., "Foundation"] - [ ] [Task 1 — specific enough to know when it's done] - [ ] [Task 2] - [ ] [Task 3] ## Phase 2: [Phase name — e.g., "Core feature"] - [ ] [Task 4] - [ ] [Task 5] ## Phase 3: [Phase name — e.g., "Testing + cleanup"] - [ ] Write tests for [X] - [ ] Update docs / README - [ ] Code review - [ ] Deploy / ship ## Blocked by - [Task X is blocked by Y] ## Notes - [Any sequencing rationale or implementation notes] ``` **Task quality bar**: Each task should be completable in one focused session (2-4 hours max). If a task would take longer, break it down. --- ## Step 4: Log to Notion (when finalized) Once requirements, spec, and tasks are signed off, the project goes into Notion. Depending on what it is: **If it's a talk/paper proposal**: → Create entry in CFP DB (`collection://1b4e2e4a-acb4-804e-9047-000bf8dade67`) → Tag with conference/event name → Body: paste the abstract + outline from spec.md **If it's a research/learning project**: → Create entry in Notes DB (`collection://9883f279-86ea-4d45-a916-2fe9f0fc32ad`) → Set Status to `In progress` → Link to `Personal projects` if it belongs to a project → Body: paste spec summary and key decisions Use `notion-create-pages` for either. --- ## Workflow Summary ``` 1. requirements.md → Harshul reviews → approved? 2. spec.md → Harshul reviews → approved? 3. tasks.md → Harshul reviews → approved? 4. Build → work through tasks.md phase by phase 5. Log to Notion → when finalized or shipped ``` No step can be skipped. If requirements aren't clear, don't write the spec. If the spec has open decisions, don't write tasks. The three docs force clarity before code. ## Source & license This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code. - **Author:** [harshuljain13](https://github.com/harshuljain13) - **Source:** [harshuljain13/agent-skills](https://github.com/harshuljain13/agent-skills) - **License:** MIT Install and usage instructions live in the source repository linked above. ## Pricing - **Free** — Free ## Security capabilities Automated source analysis of v0.1.0 — what this tool can access: - **Network access:** no - **Filesystem access:** no - **Shell / process execution:** no - **Environment & secrets:** no - **Dynamic code execution:** no *"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.* ## Versions - **0.1.0** — security scan: passed — Imported from the upstream source. ## Links - Listing page: https://agentstack.voostack.com/l/skill-harshuljain13-agent-skills-spec-dev - Seller: https://agentstack.voostack.com/s/harshuljain13 - Browse the marketplace: https://agentstack.voostack.com/browse --- Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.