Every pull request, checked against the spec it claims to deliver.
Spectrace turns messy docs and conversations into specs with acceptance criteria, puts them in your editor and your AI agent, and checks each pull request against them, criterion by criterion, citing the file and line.
From messy docs to merged code, in three steps.
Messy inputs become structured specs
Start from what you already have. Spectrace turns it into requirements with acceptance criteria you can check code against.
Requirements from the notes you already have
Drop in a PRD, a transcript, a Slack thread, a spreadsheet or a URL. Spectrace pulls out the requirements, writes acceptance criteria for each, and asks clarifying questions before anything is created.
- Docs, spreadsheets, URLs and Slack threads
- Clarifying questions before anything is created
- Vague wording flagged, with a concrete rewrite
Issue sync with Jira and Linear
Already in Jira, Linear or GitHub Issues? New and updated issues flow in as requirements, stage changes flow back to the Linear issue a requirement came from, and Push to Jira turns a requirement into a Jira issue.
Context for humans and agents
The spec goes with the work: into the test plan, into your editor, and into the agent writing the code.
What done looks like, before the editor opens
When a requirement moves to In Dev, Spectrace writes Gherkin test cases from its acceptance criteria, so everyone starts from the same definition of done. Each project can turn this off.
- Gherkin scenarios written from the criteria
- Playwright code in your editor
- Regenerate on demand
The spec in your editor, not another tab
The Spectrace extension puts the backlog, the criteria and the linked files in your sidebar, and hands the whole requirement to the agent you already use.
- VS Code, Cursor, Windsurf and VSCodium
- Hand-off to Copilot, Cursor, Claude Code and Codex
- Branch names and commit messages that link back
An MCP server your agent can drive
Claude Code, Cursor and Codex read the spec, tick criteria as they go, and check a diff against the spec before the branch is pushed, and Claude Desktop connects through a local bridge. 37 tools over the Model Context Protocol.
- Hosted on paid plans, stdio or self-hosted
- Check a diff before you push
- API keys that carry your organization role
Proof at every merge
Every pull request that names a requirement is checked against it, and the evidence lands where reviewers already look.
A verdict for every criterion, with the file and line
When a pull request opens or changes, Spectrace reads the diff against the requirement it names and gives each criterion a verdict, from met to missing to broken by this change. Each verdict cites the file and line it rests on.
- GitHub check run and GitLab commit status
- Flags a change that breaks a criterion
- A nudge on pull requests that name no requirement
Traceability that keeps itself up to date
Files link to requirements as you work, and each verified pull request strengthens the map. When a change touches code tied to another requirement, the pull request says so.
- Files linked while you edit
- Links reinforced by verified pull requests
- Drift called out on the pull request
Cycles
Commit a set of requirements and see, item by item, which shipped with a verified pull request behind them.
Stages, versions and an audit trail
Every edit, stage move and ticked criterion, with who did it and when, including changes made by an agent.
PR #219 adds the order lookup that PR #218 missed.
Drops into the stack you already use.
- GitHub
- GitLab
- Jira
- Linear
- GitHub Issues
- VS Code
- Cursor
- Windsurf
- VSCodium
- Copilot
- Claude Code
- Claude Desktop
- Codex
- MCP server
- Slack
- Markdown
- Word
- Excel
- Confluence
- Azure DevOps
From messy docs to merged code, with proof.
Free for individuals · No card required