CLI FOR AI-WRITTEN PRS

Your agent writes the code.Scriptonia keeps it on product.

Turn customer signal into a sourced PLAN.md. Then verify the agent's diff against what customers asked for, what your team decided, and what was explicitly out of scope.

$npx scriptonia loginbrowser sign-in · no key copying

Works with any agent that reads PLAN.md or AGENTS.md — including Claude Code, Codex, Cursor, OpenClaw, and Hermes.

~/sample-productexample CLI output
$ npx scriptonia plan "filter audit log by actor"
✓ PLAN.md written · 5 cited signals · 1 decision
$ codex exec "implement PLAN.md"
✓ captured run_01J · child exit 0
$ npx scriptonia verify
SCRIPTONIA VERIFY · PASS
PASSacceptance criteria
PASSnon-goals
PASSproduct decisions
PASSunresolved constraints
PASStest coverage
evidence, not vibesexit 0
Local-firstrepository verification
Deterministiccore checks
Human-approveddecision overrides
Versionedplans, rules, and evals

Coding agents know your codebase. They do not know your customers.

A technically correct PR can still be the wrong product decision. The customer request lived in a call. The constraint lived in Slack. The non-goal was agreed three months ago.

Scriptonia gives those facts a durable path into the plan — and a testable path into review.

WITHOUT PRODUCT MEMORY
issueagentplausible PRhope
WITH SCRIPTONIA
signalPLAN.mdagentverified PR

One loop from customer words to merge confidence.

Each stage produces an artifact you can inspect, review, and keep. There is no dashboard required to understand what happened.

  1. 01

    Signal

    Sourced customer evidence
    npx scriptonia add feedback.txt call.vtt

    Add the raw customer words your agent would otherwise never see. Every item keeps a source and a stable signal ID.

  2. 02

    Plan

    Executable PLAN.md
    npx scriptonia plan "filter the audit log by actor"

    Turn one issue into PLAN.md with cited evidence, non-goals, acceptance criteria, real file paths, and contradiction gates.

  3. 03

    Run

    Traceable agent run
    npx scriptonia run -- codex exec "implement PLAN.md"

    Let your coding agent work from the reviewed target while Scriptonia captures the exact command and exit result.

  4. 04

    Verify

    CI-safe gate status
    npx scriptonia verify

    Check the diff against product decisions, unresolved constraints, scope, tests, and declared CI evidence before merge.

  5. 05

    Learn

    Draft regression case
    npx scriptonia learn finding_<id>

    Convert a confirmed miss into a reviewable, versioned draft eval. Once reviewed and activated, it can catch the same failure next time.

Trust the artifacts, not a black box.

Scriptonia is deliberately explicit about what runs locally, what leaves your machine, and who gets the final say.

01

Deterministic by default

The core verification path works without model credentials. Semantic evaluation is advisory and explicitly enabled.

02

Your repository stays local-first

Verification reads Git, PLAN.md, reviewable YAML, and declared evidence. It does not run commands found in repository config.

03

Humans resolve contradictions

A prior decision cannot be silently reasoned away. It stays UNRESOLVED until a person records the override.

04

Every gate is inspectable

Decisions, evals, evidence references, evaluator versions, and findings remain readable instead of disappearing into a score.

Local where it should be. Hosted where it helps.

We will not tell you “nothing leaves your machine” when that is not true. Customer signal is hosted so you can retrieve it across projects and time. Repository policy, deterministic checks, and operational memory stay local-first.

ON YOUR MACHINE
  • Git diff and repository snapshot
  • PLAN.md and AGENTS.md
  • Decisions and regression evals
  • Deterministic verification state
SCRIPTONIA SERVICE
  • Customer signal you choose to add
  • Search and plan synthesis
  • Account and credit metering
  • Bounded semantic evidence, when enabled

Prove it on one real issue.

Sign in, add one piece of customer evidence, and ask for the issue your team is already considering. Keep Scriptonia only if the resulting plan is better than the brief you would have written.

Run the quickstart
  1. 1
    npx scriptonia login

    Sign in and install the Claude skill.

  2. 2
    npx scriptonia add feedback.txt

    Add customer evidence for free.

  3. 3
    npx scriptonia plan "the issue"

    Review the sourced PLAN.md.

Start free. Pay when the loop earns its place.

Adding customer signal is always free. Credits meter synthesis and verification. Every plan has a hard cap, so there are no surprise usage bills.

Free
$0

Try the complete workflow

  • 25 credits each month
  • Planning and verification
  • Signal ingestion stays free
  • Hard monthly usage cap
Start free
Pro
$29 / month

For founders shipping every week

  • 400 credits each month
  • Planning and verification
  • Signal ingestion stays free
  • Optional credit top-ups
Choose Pro
Team
$49 / seat / month

For teams with higher usage

  • 600 credits per seat, pooled
  • Planning and verification
  • Signal ingestion stays free
  • Hard monthly usage cap
Choose Team

Use the same email for Scriptonia login and checkout so paid access activates automatically. Need a custom plan or credit pool? Talk to the founder.

Reasonable questions. Direct answers.

Still unsure? Email sathwik07@scriptonia.dev. A human reads it.

Does Scriptonia write the code?

No. Claude Code, Codex, Cursor, OpenClaw, Hermes, or another coding agent writes the code. Scriptonia gives that agent a grounded target and checks the result against product memory.

Where does my data go?

Customer signal is stored by Scriptonia so it can be searched across your history. Repository verification is local-first. If you explicitly enable semantic evaluation, only the bounded evidence bundle is sent through the authenticated Scriptonia gateway.

Does verification replace CI?

No. Scriptonia complements CI. It consumes declared CI and JUnit results, then adds the product checks ordinary test suites cannot know: customer requirements, non-goals, prior decisions, and approved scope.

What happens when the evidence is incomplete?

Scriptonia says so. Individual checks return PASS, FAIL, INCONCLUSIVE, ERROR, or SKIPPED. The aggregate gate is PASS, PASS_WITH_WARNINGS, ACTION_REQUIRED, BLOCKED, or INCOMPLETE. Missing proof is not converted into a confident success.

Can a team review the rules?

Yes. PLAN.md, decisions, evals, and enforcement live in readable files designed for code review. Deterministic definitions are versioned with the repository.

Give your coding agent more than code.

Give it the customer evidence, decisions, scope, and checks that make the code worth shipping.

npx scriptonia login