orbit-prompt

https://github.com/user-attachments/assets/747913ce-ed00-4622-82fb-29b8e8dd4028

Orbit Prompt

A Claude Code plugin that turns rough ideas into structured prompts.

Most AI correction turns are caused by vague prompts, not by the model.

orbit-prompt catches weak prompts before they cost you turns. Two surfaces: a /orbit-prompt command you call explicitly, and automatic session diagnostics that stay silent when the session is healthy.

---

Example: turning a vague request into an actionable prompt

Example:

/orbit-prompt "Fix the login flow"

Output:

[Orbit Prompt — Prompt Refinement]

ORIGINAL PROMPT:
"Fix the login flow"

ANALYSIS:
- No failure mode identified (what exactly is broken?)
- No scope boundary (which part of the login flow?)
- No success criterion (how do we know it is fixed?)

IMPROVED PROMPT:
Investigate and fix the login redirect issue in the web app.

Context:
Users can submit valid credentials, but after login they are sometimes redirected back to the login page instead of reaching the dashboard.

Scope:
- Focus only on the web login flow.
- Start by inspecting the auth route, session handling, and dashboard redirect guard.
- Do not refactor the full authentication module.
- Do not change password validation, OAuth behavior, or API contracts unless a failing test proves they are involved.
- Prefer the smallest safe change that fixes the redirect behavior.

Expected work:
1. Reproduce or identify the redirect failure path.
2. Find the smallest safe code change.
3. Add or update one regression test that proves a valid logged-in user reaches the dashboard.
4. Keep existing tests passing.

Acceptance criteria:
- Valid login redirects to the dashboard.
- Invalid login keeps the current error behavior.
- Existing session behavior remains unchanged.
- At least one regression test covers the redirect behavior.
- No unrelated authentication files are modified.
- All existing tests pass.

Recommended first step:
Search for the login submit handler, session persistence logic, and dashboard redirect guard before editing code.

KEY IMPROVEMENTS:
- Added specific failure mode (redirect to login page after valid credentials)
- Scoped to web login flow only — excludes password validation, OAuth, and unrelated auth code
- Defined explicit acceptance criteria with a required regression test

READY TO SEND:
Yes — the request now has a specific failure mode, clear scope boundaries, and measurable acceptance criteria.

This turns a vague request into a scoped engineering task with a clear target, safe boundaries, acceptance criteria, and a regression test.

---

What it is

A Claude Code plugin that turns rough ideas into structured prompts. It exposes two surfaces:

1. /orbit-prompt (primary) — You request an analysis of a prompt before sending it. The plugin identifies gaps and suggests a structured version. You decide whether to use it.
2. Session diagnostics (automatic, silent by default) — It observes the session and only surfaces a diagnosis when it detects one of the 8 inefficiency patterns. When the session is healthy, it stays silent.

The plugin does not modify anything. It reads, analyzes, and displays. You act.

---

Main Command: /orbit-prompt

Use when you want to review a prompt before sending it.

Syntax:

/orbit-prompt "your prompt here"

What happens:

The plugin reads the prompt, identifies what is missing (scope, acceptance criteria, boundaries) and displays a suggested version. You read it and decide whether to use it, adapt it, or discard it.

Example:

/orbit-prompt "Refactor auth module"

Output:

[Orbit Prompt — Prompt Analysis]

ORIGINAL:
"Refactor auth module"

GAPS:
- No file target (which file?)
- No scope boundary (password? OAuth? both?)
- No acceptance criteria (what does done look like?)
- Risk: high speculation — will require corrections

SUGGESTED:
"Extract password validation from src/auth/auth.ts into
src/validators/password.ts. Keep existing function signatures.
Do not touch OAuth flow or routes. Done = all existing tests pass."

CHANGES MADE:
- Added specific file (auth.ts → validators/password.ts)
- Scoped to password only (not OAuth)
- Preserved contract (function signatures unchanged)
- Added verifiable success criterion

You copy what you want. Or ignore it. The plugin sends nothing.

---

Session Diagnostics

The plugin watches the session and surfaces a diagnosis only when it detects one of the 8 patterns. If the session is healthy, it stays silent. You do not trigger anything — automatic detection is the default.

Advanced triggers (optional): if you want to force a diagnosis on demand instead of waiting for automatic detection, these natural-language phrases work:

analyze cost
/analyze-cost
how efficient is this?
is this optimal?