Skip to content

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

43 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

StateSeal

English · 简体中文 · 日本語 · 한국어 · Español · Português do Brasil · Deutsch · Français

StateSeal is a read-only runtime intelligence layer for coding agents. It reconstructs what happened across a delivery session—requests, Skills, tools, artifacts, failures, and reported outcomes—without starting, steering, blocking, approving, or applying Agent work.

Observe the execution. Explain the chain. Never become the execution.

StateSeal Runs Console

Why

Agent transcripts contain valuable operational evidence, but it is fragmented across messages, tool calls, files, test output, and sub-agents. Developers need answers to practical questions:

  • Was the expected Skill discovered and actually used?
  • Which tools ran, in what order, and which ones failed?
  • What files and artifacts were produced?
  • Which facts are observed, which attributions are derived, and which conclusions are only inferred?
  • Where did time go, and where should a failed run be debugged?

StateSeal turns those signals into a searchable session list, event timeline, and attribution DAG.

Product invariant

StateSeal is observational infrastructure:

Agent-owned session logs + local artifacts
                    ↓  read only
        normalization and correlation
                    ↓
   panorama · attribution · diagnosis · audit

It does not install lifecycle hooks, proxy model traffic, launch an Agent, modify prompts, return deny/block signals, approve delivery, change branches, or apply code. Agent execution and outcomes remain owned by the Agent and user.

Every analytical statement has an evidence grade:

Grade Meaning
Observed Present directly in an Agent transcript or local runtime event
Derived Deterministically correlated from observed IDs, paths, and time
Inferred A diagnostic hypothesis; useful, but not asserted as fact

Quick start

Install from a source checkout with Go 1.24 or newer:

make install
export PATH="$HOME/.local/bin:$PATH"
seal ui

Or install a published release:

curl --proto '=https' --tlsv1.2 -fsSL \
  https://github.com/hellogxp/stateseal/releases/latest/download/install.sh | sh
export PATH="$HOME/.local/bin:$PATH"
seal ui

The UI listens only on a loopback address and passively discovers supported local Codex session transcripts. It refreshes while the Agent runs. Raw tool arguments and outputs are not exposed by default.

Useful options:

seal ui --no-open
seal ui --address 127.0.0.1:9137
seal workspace list /path/to/workspace

Current scope

  • passive Codex JSONL session discovery;
  • live all-session overview;
  • request → Skill → tool → observed-outcome DAG;
  • explicit Observed / Derived / Inferred semantics;
  • privacy-preserving summaries and secret redaction;
  • historical StateSeal evidence shown as a read-only archive;
  • English, Simplified Chinese, Japanese, Korean, Spanish, Brazilian Portuguese, German, and French UI locales.

Claude Code and other Agent transcript adapters are planned. Compatibility is reported per source format; StateSeal never claims semantic correctness merely because a command completed.

Commands

The default command surface is intentionally observational:

Command Purpose
seal ui Open the local live panorama
seal workspace list [path] Discover repositories without modifying them
seal version Show build identity

Managed execution, verification gates, admission, Apply, restore, lifecycle hook installation, and delivery-control MCP tools are not registered.

Documentation

Development

go test ./...
go vet ./...

Licensed under Apache-2.0.

About

Transactional admission and verified-state management for coding-agent changes.

Topics

Resources

Contributing

Security policy

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages