SPECTRE: A "Waterfall Pipeline" Prompt Workflow for AI Coding Agents

2026-02-18 | Product Hunt | GitHub

Gemini's Take: This is a screenshot of SPECTRE running in the Claude Code terminal. You can see session recovery, structured output for Goals/Summary/Next Steps, and prompts for 19 automatically loaded knowledge skills. The bottom shows the bypass permissions mode—exactly how the founder uses it daily.

30-Second Quick Judgment

What is it?: A prompt workflow plugin for Claude Code that transforms AI coding from "vibe-based prompting" into a standardized pipeline: Scope -> Plan -> Execute -> Clean -> Test -> Rebase -> Evaluate.

Is it worth it?: If you're a heavy Claude Code user, it's worth 10 minutes to install and try. If you don't use Claude Code, this isn't for you. It only has 3 votes on PH and is in the very early stages, but the spec-driven philosophy behind it is a major trend for AI coding in 2026.

Three Key Questions

Is it for me?

• Target Audience: Developers and PMs using Claude Code daily, especially those building mid-to-large features (not just changing a button color).
• Are you the one?: You are if you often face these issues:
  • AI writes spaghetti code where fixing one bug creates three more.
  • You hit the context window limit halfway through a feature and have to re-explain everything in a new session.
  • The AI makes decisions that differ from your intent, wasting time on corrections.
• When to use it:
  • Developing a new feature from scratch -> Start with /spectre:scope.
  • Adding features to a 100k+ line codebase -> /spectre:research then /spectre:plan.
  • Long development cycles needing context across sessions -> /spectre:handoff.

Is it useful?

ROI Judgment: If the feature takes over an hour to build, spending 10 minutes on the scope -> plan flow is almost certainly worth it. For tiny fixes, this process is a burden. Simply put, it applies the old software engineering mantra of "design before you code," but lets the AI handle the design heavy lifting.

Is it enjoyable?

The "Aha!" Moments:

• Guided Next Steps: Every command ends by telling you exactly what to do next, so you don't have to memorize the workflow.
• Knowledge Accumulation: /spectre:evaluate distills what it learned into "Skills" that auto-load next time—it gets smarter the more you use it.
• Session Memory: /spectre:handoff generates a status report that the next session picks up automatically, ending repetitive background explanations.

Real User Feedback:

To be honest, it's so new (3 PH votes) that there aren't many independent public reviews. The founder used it to build a 250K line Tauri/Rust/React desktop app (Subspace), which is the strongest proof point so far.

Industry opinion on the spec-driven direction is polarized: supporters say "one iteration with a spec equals eight blind writes without one"; critics call it "the return of the Waterfall model with a mountain of Markdown files." — via Marmelab and Scott Logic

For Indie Hackers

Tech Stack

• Frontend: N/A (It's a set of prompt files, not traditional software)
• Backend: N/A (Pure Claude Code Skills/Plugins)
• AI/Model: Claude (Anthropic), executed via Claude Code CLI
• Infrastructure: Runs in your local terminal; no cloud dependencies

Core Implementation

SPECTRE is essentially a set of meticulously crafted prompt templates exposed via Claude Code's slash command system. Each command corresponds to a phase in the dev lifecycle:

1. Scope: The AI asks you the right questions (instead of you guessing what the AI needs), turning vague ideas into clear specs.
2. Execute: Uses subagents to break down large tasks. Each subagent generates a completion report for the next, driven by TDD (Test-Driven Development).
3. Evaluate: Performs an architecture review and captures learned knowledge into Skills files that auto-load in new chats.

The founder calls it "Rapid Waterfall"—define requirements first, then execute. The difference is that AI makes the "defining requirements" part extremely fast.

Open Source Status

• Fully Open Source: All code is available on GitHub.
• Similar Projects: claude-code-spec-workflow, cc-sdd, Spec-Flow, Superpowers Plugin.
• Build Difficulty: Low. The core is writing great prompt templates. The real barrier is the 12 months of experience baked into the prompt quality.

Business Model

• Monetization: None currently (purely open source).
• Pricing: Free.
• Speculation: The founder likely uses this as a brand/lead-gen tool for Subspace or the Codename brand.

Giant Risk

High. GitHub has already launched spec-kit, and Claude Code is officially improving its plugin/skills system. By 2026, the Claude Code ecosystem already has 270+ plugins and 739 skills. SPECTRE's moat isn't tech; it's the founder's experience. If official tools become "good enough," SPECTRE's differentiation vanishes.

For Product Managers

Pain Point Analysis

• The Problem: AI Coding Agents produce poor, inconsistent results when they lack clear specifications.
• Severity: High-frequency and critical. Founder's quote: "ambiguity is how you end up with spaghetti code, conflicts, and AI slop." Anyone who has used AI for more than 100 lines of code will relate.

User Personas

• Persona 1: Indie hackers or Tech Leads using Claude Code daily for mid-to-large feature development.
• Persona 2: Product-minded technical managers (like the founder, an ex-TPM) using AI to bridge gaps in coding experience.

Feature Breakdown

Competitor Comparison

Key Takeaways

1. Auto-suggest Next Steps: Outputting "Next Steps" after every command keeps the user on track and lowers cognitive load.
2. Knowledge Capture: Capturing "Skills" during evaluation so the AI understands your project better over time—a concept applicable to any AI product.
3. Session Handoff: Solves one of the biggest AI pain points—the "amnesia" that happens when starting a new chat.

For Tech Bloggers

Founder Story

• Founder: Joe Fernandez
• Background: Ex-Meta, ex-Amazon Technical Product Manager, ex-Pioneer Square Labs.
• The "Why": As a non-traditional developer (TPM background), he used AI to build products far beyond his previous technical reach. SPECTRE is the result of 12 months of daily Claude Code usage. He claims this workflow allows him to "build and iterate on products 100x more complex than before."
• Products Built: New June (React Native AI + GPS golf rangefinder) and Subspace (250K line Tauri/Rust/React desktop app, positioned as "the new home for coding agents").

Controversies & Discussion Points

• "Waterfall Revival?": Spec-driven development is essentially design-before-code, which irritates Agile purists. Marmelab's article explicitly calls SDD "the return of Waterfall."
• "Markdown Hell?": Critics argue SDD generates too much documentation, forcing devs to spend more time reading long Markdown files than writing code.
• "TPM building Dev Tools?": A PM-background founder building developer tools is a fascinating angle—AI is making it possible for "non-devs to build complex software."

Hype Data

• PH Ranking: 3 votes (Extremely low visibility).
• Twitter/X: Almost no independent discussion yet.
• GitHub: New repo, very few stars/forks.
• Industry Trend: However, Spec-Driven Development as a direction is red hot in 2026—GitHub spec-kit, Amazon Kiro, and several indie tools are all moving this way.

Content Suggestions

• Angle: "In the AI era, can a PM replace a programmer?" — Use the SPECTRE founder as a case study.
• Trend Jacking: The "spec-driven vs. vibe coding" debate is heating up; do a head-to-head comparison.

For Early Adopters

Pricing Analysis

Hidden Cost: SPECTRE's scope/plan steps consume a lot of tokens (generating detailed specs). On a Pro plan, you might hit limits quickly.

Getting Started

• Setup Time: 5 mins to install, 30 mins to learn the flow.
• Learning Curve: Low (if you use Claude Code) / High (if you've never used it).
• Steps:
  1. Ensure Claude Code CLI is installed.
  2. Run marketplace add Codename-Inc/spectre.
  3. Type /spectre:scope in any project.
  4. Follow the AI's guided questions; it will suggest the next step automatically.

Pitfalls & Complaints

1. Anthropic Lock-in: Only works in Claude Code. If you use Cursor, this isn't for you.
2. Manual Session Memory: You must remember to run /spectre:handoff at the right time, or you'll still lose context between sessions.
3. Token Consumption: Scope + Plan docs can be long. The founder suggests disabling auto-compact and manually handing off at 160k tokens.
4. No Community: It's too new. If you hit a bug, you're stuck with the README or filing a GitHub issue; there's no Discord/forum yet.

Security & Privacy

• Data Storage: Entirely local (Claude Code runs in your terminal; SPECTRE is just prompt files).
• Privacy: Code is processed via Anthropic API, following their privacy policies.
• Audit: Open source and auditable, but no formal security audit has been performed.

Alternatives

For Investors

Market Analysis

• AI Coding Agents: Already over $2B. GitHub Copilot ~$800M ARR, Cursor/Anysphere ~$500M ARR.
• Broad AI Agents Market: ~$7-8B in 2025, projected $10-11B in 2026, and $47-183B by 2030.
• Growth Rate: CAGR of 42-50%.
• Driver: Developer demand shifting from "assisted completion" to "autonomous execution."

Competitive Landscape

SPECTRE's "workflow layer" is an interesting but difficult spot to monetize independently—it's an add-on to the Claude Code platform, more like a value-added component than a standalone product.

Timing Analysis

• Why now?: The Claude Code plugin market is maturing in 2026 (270+ plugins), and spec-driven development has been validated by GitHub (spec-kit) and Amazon (Kiro).
• Tech Maturity: Claude Code's skills/plugin system is now robust enough for stable subagent execution.
• Market Readiness: The inflection point from "vibe coding" to "structured agentic coding" is happening now.

Team Background

• Founder: Joe Fernandez (Ex-Meta, Ex-Amazon TPM, Ex-Pioneer Square Labs).
• Core Team: Appears to be a solo project or a very small team.
• Track Record: Built the 250K line Subspace desktop app using SPECTRE.

Funding Status

• Funding: No public information.
• Investment Value: As an open-source prompt workflow, SPECTRE itself isn't a typical VC play. However, the founder's other project, Subspace ("the new home for coding agents"), might be the real commercial play.

Conclusion

One-Sentence Verdict: SPECTRE isn't just a product; it's a set of best practices shared by an experienced Claude Code user. The spec-driven agentic coding direction is correct, but the tool itself is very early, niche, and dependent on a single platform.

Resource Links

2026-02-18 | Trend-Tracker v7.3