sworn v0.1 - building in the open

AI agents write code.
sworn.sh makes them answer for it.

One binary. Plan, implement, verify, merge - fail-closed at every step. Bring your own model. Zero runtime dependencies.

sworn run
$ sworn run --task "add rate limiting to the API"
Star on GitHub Get early access →
The fidelity chain
Accountability at every layer

Most tools verify that code compiles. sworn verifies that what shipped is what was needed - at every seam from requirement to production.

01 / requirements
Requirements Fidelity
Is the spec the right spec?
PASS

Quality gate over ISO/IEC/IEEE 29148 characteristics. Traceability matrix: need → acceptance criteria → proof. Nothing drops silently between intake and spec.

02 / design
Design Fidelity
Is this the right solution overall?
PASS

Stakes-calibrated, human-owned design-fit gate. Reversibility × blast-radius sets the bar. Design-system conformance audit for UI-bearing projects.

03 / delivery
Delivery Fidelity
Does the code match the spec?
PASS

Fresh-context adversarial verifier reads only spec + diff. Exit 0 only on PASS. Silence, errors, and unparseable verdicts all block the merge.

04 / ship
Journey Validation
Does the assembled system work?
ATTEST

Fail-closed human walkthrough of critical journeys. Mocks off, real infra. sworn ship requires your attestation before any release goes out.

Command surface
Five commands.
End to end.

Each command is a gate. Each gate is fail-closed. The loop generates its own inputs - you never wire spec plumbing by hand.

01
sworn init
One API key, zero other config. Sets a safe-hosted default model. Turnkey from a clean machine in under a minute.
02
sworn plan
AI drafts the spec - EARS-structured acceptance criteria, traceability to need, benefit hypothesis. You ratify. The requirements gate is satisfied before a line of code is written.
03
sworn run
Implement → fresh-context verify → retry / escalate → gated merge. Fully automated. The loop generates the spec, drives a coding model, produces the proof bundle, and verifies - so you never assemble the plumbing.
04
sworn top
Real-time Bubble Tea TUI evidence board. Live runs, verdicts, costs, journey-validation status. The green board - or the kill list. Read-only; sworn produces evidence, it doesn't own your workflow.
05
sworn ship
Human-walked critical journeys, mocks off, against real infra. You attest. Only then does the release ship. Not before. This is the oath.
Open protocol
Built on the open Baton protocol

Baton defines the fidelity chain - rules, role prompts, verdict contracts, phase gates. SwornAgent embeds it. Two repos, one install. The protocol is open for any team to implement; the product is the turnkey binary that ships it ready to run.

GitHub → Baton protocol →
install 
# Homebrew
$ brew install swornagent/tap/sworn

# Go
$ go install github.com/swornagent/sworn/cmd/sworn@latest

# Then run
$ sworn init
$ sworn run --task "your task here"
Building with AI agents?

We're building in the open. Get notified when sworn reaches your use case.

No spam. Just signal. We'll reach out when it's your turn.