SPECTRE - An agentic coding workflow for product builders
by•
SPECTRE is an agentic Coding Workflow - /Scope, /Plan, /Execute, /Clean, /Test, /Rebase, /Evaluate - that uses simply step by step product development workflow to generate high quality results from your AI Coding Agents. - Codename-Inc/spectre
Replies
SPECTRE
@joenandez Congrats on your launch! What happens when execution diverges from plan mid-workflow?
SPECTRE
@kimberly_ross the execution process is specifically "adaptive". The agent that is dispatching subagents (and the subagents themselves) are specifically instructed to flag when their plan needs to change, or when they had to switch gears and try something different. The primary agent's job is to adapt the Tasks and when they dispatch the next wave of subagents to share the new information. So it is very adaptive to changes to the plan mid workflow.
@joenandez Hi Joe - love the clarity behind SPECTRE. “Ambiguity is death” really captures why most agent workflows fail. Treating inputs as a first class system instead of just prompting casually makes a huge difference.
A repeatable workflow across any codebase is powerful.
I’m building Ahsk, a macOS AI assistant focused on structured, in flow AI use. Would love to connect and exchange feedback.
Stellify
I like the objective Joe, best of luck with the launch!
SPECTRE
@stellify_software I appreciate it Matt!
SPECTRE
@vouchy oh man all the time. I actually experimented with the first version of our app Subspace and built two versions - one with Spectre (an early version) and one without. The without version looked better, but once you got down one layer it was a complete untangle-able mess. The Spectre version is what exists now.
Agentic coding has completely changed how I ship. Built a 118k-line SwiftUI Mac app solo using Claude Code — the key insight was that it's not just autocomplete, it's having something that actually understands the full codebase. What's your approach to context management across longer sessions? That's where I've found the biggest gains.
SPECTRE
@matthias_stralman 100%. Our app Subspace has a Rust backend. I’ve never written a single line of Rust before we started building it. LLM’s are a universal translater and the language is system architecture!
SPECTRE uses a manual /handoff that auto inserts a status report into the next session - complete continuity until you run /forget. And the learning Skill documents features that get auto loaded when relevant. It’s really simple but it works really well.
@joenandez The /handoff pattern is smart — that's essentially what I've been doing manually with context files committed to git. Skill documents that auto-load is the missing piece though. Do you expose that as part of SPECTRE or is it internal tooling?
SPECTRE
@matthias_stralman yeah once you create 1 skill with /spectre:learn, it auto-injects the "spectre:apply" skill directly into the context window, as well as a registry of skills created by SPECTRE. The Apply Skill's main job is to force the agent to consider if it should use a Skill based on the user's request, and then provides the keywords and location of the Skill.
I have a library now of 25 (and growing) skills for various aspects of my project/codebase created via this method. Claude Code is really good at pulling the right one up right when I need it. I used to have to @ mention relevant files to load the context now I just say what I want and it almost always loads the right Skill.
@joenandez It's part of SPECTRE? Or internal tooling you've built on top? Asking because the auto-load piece is exactly what I've been missing — my CLAUDE.md approach works but it's manual.
Congrats on the launch, @joenandez !
SPECTRE
@neilverma thanks Neil!
The slash command naming is clever - makes it feel like invoking a protocol, not just prompting an AI. In my experience the /Scope phase is always the first thing that gets skipped when you're trying to ship fast, and then you end up paying for it later. Curious how /Rebase handles things when the context window fills up - that's usually where I see things go sideways.
SPECTRE
@soumikmahato hey soumik! it is not a fork of claude code! its a plugin that works with claude code and takes advantage of some of the features Claude Code offers to developers.
Congratulations on the launch! You should offer this as a technology for Cursor to acquire. I think they would be interested.
I've been using AI coding for 12 months and I've spent many, many days of that 12 months fighting AI when performing high-level development. It takes an awful lot of patience, dedication, and commitment.
I started to use Claude code in my terminal recently (to avoid tools like Replit and Lovable and the rest) and came across Spectre from a product hunt email (I think it was). I thought I'd give it a try. I performed my first action with it yesterday and scoped out an incredibly high-level task for me and for the platform. I went through all of the steps and it was built and implemented today with minimal revisions. I would confidently say it would have taken me three or four days (if not more) to do what happened with Spectre in a few hours. It's that much of an increase in productivity and output and the lack of stress is amazing.