For Claude Code & Codex
Spectre is a battle-tested 6 stage agentic engineering workflow for building complex, production-ready features. You stay in the loop for the decisions that matter. Your agents handle the the rest.
Think like a product builder. Ship like an engineering org.
Free · Open source · MIT licensed
/spectre:scopeWhy Spectre
Direct prompting works when the change is obvious. When the details are fuzzy—or the work is big enough that planning would prevent rework—every missing decision becomes another correction loop. You become the orchestration layer.
The more ambiguity a change carries, the more expensive improvisation becomes.
Spectre moves the feedback loops forward—to scope, UX, and technical decisions, where your judgment has the most leverage.
How we solve it with Spectre
Spectre scales the workflow to the work. It keeps focused changes compact and adds structure as uncertainty, risk, or coordination grows.
01 / Right-sized from XS to XL
Spectre creates the smallest sufficient plan for the job. Focused work stays compact. As uncertainty, risk, or coordination grows, Spectre adds the research, review, and task breakdown the work actually needs.
/spectre:scopeIntent → architecture → work → evidence
02 / Less rework, Less fix loops
Get much closer to right the first time. Your agent is not improvising from a paragraph-long prompt. Before implementation, it has explicit scope, UX decisions, codebase research, an approved technical design, a stress-tested implementation plan, and a defined verification path. When it is time to prove the work, that original intent becomes the acceptance criteria.
03 / Foccus where it matters
Replace correction loops with product work. You focus on definition, user flows, high-level technical decisions, and last-mile adjustments to a working experience. Spectre handles the coordination so agents can work longer without asking you to babysit every step.
04 / From Idea to PR, Step by Step
For feature work worth planning, start with /spectre:scope in Claude Code or $spectre-scope in Codex. Spectre clarifies the boundaries, evaluates the work, and chooses the smallest sufficient path—from a compact XS plan to a reviewed, structured XL plan. No command map to memorize. No documentation to keep open.
Feature work that starts with Scope
/spectre:scopeSet boundaries and acceptance. Spectre then guides you to UX or Plan.
Research the current surface, isolate the change, and preserve its contracts.
Define the smallest complete product, then follow Spectre’s recommended next step.
Set the boundary, then move into /spectre:ux for flows and behavior.
One command → next step included
05 / Regularly updated & improved
Every skill, handoff, and workflow has been shaped through 18 months of daily use and iteration across Subspace, New June, Spectre itself, and projects built by early users. When reality exposes a gap, that lesson goes back into the system. Spectre will keep getting faster, more efficient, easier to use, and more effective.
“My time is now spent 90% on scope/UX planning, big thorny complex problems, dogfooding, and building the machine that builds the machine.”
The right-sized plan. Less rework. Less back-and-forth. One clear next step at a time.
How it works
Each phase turns decisions into durable context for the next. You stay at the judgment points, where AI helps you shape the right product. Spectre handles the research, coordination, review, and verification between them.
01 / 06
Decide what you’re building
Define what is in, out, future, and explicitly not part of the feature. If the experience is still fuzzy, work through user flows, interactions, states, and UI prototypes before planning.
02 / 06
Turn intent into architecture
Subagents fan out across the codebase first. You review, refine, and approve the high-level design. Spectre then creates the implementation plan, challenges it with an opposing agent, and prepares detailed tasks, subtasks, and a ready-to-use goal prompt.
03 / 06
Build without becoming the coordinator
Your primary agent dispatches dependency-ready work, keeps each workstream test-first and committed, independently verifies the affected system after every wave, and routes compounding risk through review and repair. Once the plan is complete, an opposing agent performs one final adversarial review before Prove.
04 / 06
Show that the experience works
Spectre turns the scope and UX you approved into acceptance criteria, then gathers real evidence by using the product like a user would. It inspects the resulting tests, screenshots, video, output, state, and logs and creates a visual proof artifact you can review.
05 / 06
Finish the last five percent
Spectre cleans the working set, closes risk-weighted test gaps, safely rebases, observes the full suite once, and opens a reviewer-ready draft pull request.
06 / 06
Make the next feature smarter
Capture the durable decisions, patterns, and gotchas this build uncovered so future agents can load the right context when it matters.
Scope → Plan → Execute → Prove → Ship → Learn
/spectre:scopeClaude Code$spectre-scopeCodexThe principles
When scope, UX, plans, or acceptance are vague, the model fills in the blanks. Sometimes you get lucky. On real product work, those guesses compound into tangled rework, brittle code, and poor experiences.
01 · explicit inputs
Spectre makes rigorous scoping, UX thinking, and technical planning easier. You bring the vision; AI helps shape it, challenge it, and explore its edges before planning and implementation begin.
02 · canonical decisions
Canonical artifacts preserve the intent behind the work across phases, agents, and sessions.
03 · adversarial review
Plans and code are challenged by independent reviewers—Codex reviews Claude Code and vice versa—before mistakes can compound.
04 · observable proof
Passing tests and confident agent summaries are not enough. Your agent uses the finished experience as a user would and checks it against the original intent with reviewable evidence.
05 · durable knowledge
Each meaningful build can teach future agents the patterns, decisions, and gotchas your project has already earned.
Do not use AI to skip thinking. Use AI to make rigorous thinking cheap, then give agents enough clarity to execute it well.
Use the right tool for the work
Use Spectre when the details need alignment, existing behavior needs protection, or a plan would reduce rework—whether the work is focused or complex.
If the edit is exact, obvious, and low-risk, use your agent directly. When ambiguity, boundaries, or verification make planning worthwhile, Spectre scales the workflow to fit. For bugs, use /spectre:fix.
Completely free and open source
Choose your agent and install the plugin. For feature work worth planning, Scope is the only command you need to remember.
claude plugin marketplace add joenandez/spectreclaude plugin install spectre@spectrecodex plugin marketplace add joenandez/spectrecodex plugin add spectre@spectreRestart or open a new Codex session after installation.
Your first feature
Tell Spectre what you want to build. It researches what it can, asks questions until the boundaries are clear, writes the scope, and recommends the right next action. UX-heavy feature? It guides you to UX. Experience already clear? It moves you to Plan. You never have to remember what comes next.
One command to start. One clear next step from there.
Read the documentation on GitHub/spectre:scope$spectre-scope