Ticket Crusher

AI engineering agent for Jira

From ticket to merged PR,
with evidence at every step.

Ticket Crusher runs AI coding agents — Kiro, Claude, Gemini or Codex — through the whole life of a Jira ticket. It investigates before it guesses, cites the query behind every claim, and waits for your approval before it writes a line of code.

See how it works Why it's trustworthy
  • Kiro
  • Claude
  • Gemini
  • Codex

One label starts it. You approve what matters.

The workflow is driven by Jira labels, so your teammates see the same state in Jira that you see on the dashboard.

  1. Validate

    Checks the ticket has enough context to work on, and says exactly what is missing if it doesn't.

  2. Investigate

    A budgeted, read-only session queries your environments through a guarded route and builds an evidence bundle.

  3. Analyse & Plan

    Posts a short root-cause comment. The full code path, impact and test cases go on the ticket as attachments.

  4. You approve

    Read the RCA, approve it, or request changes. Nothing is implemented before this.

  5. Implement

    Works in an isolated git worktree, runs the tests and raises a pull request.

  6. Verify

    Exercises the fix against a testing environment and posts the verdict, with its evidence, to the ticket.

  7. Review & Merge

    Addresses review comments, merges once a teammate approves, and writes the Dev Notes.

It investigates first, so it doesn't guess.

Before analysis, Ticket Crusher gathers facts: database queries, server logs, metrics and error traces from the environments you configure. Every claim in its evidence bundle cites the query that produced it.

If a later verification contradicts the fix, the ticket goes back to investigation carrying that contradiction, bounded by an iteration count and a per-ticket cost ceiling.

Attached to the ticket
  • investigation-evidence.mdclaims, each citing its query
  • investigation-queries.mdevery read-only query, in run order
  • investigation-logs.txtthe log lines the evidence cites
  • analysis-detail.mdcode path, impact, testing strategy
  • test-cases.mdwhat the fix must satisfy

More than tickets

Every module shares the same base, and the Home screen answers one question: what needs me right now?

PR Crusher

A review queue for pull requests waiting on you, created by you, or going stale.

Slack inbox

Unread conversations ranked by what Slack already knows needs you, with reply, mark-read and ticket-from-message built in.

Unit Test Crusher

Measures coverage with your own build and plans the tests worth writing.

Monitoring

Monitors run queries on a schedule, group results into signals, and file or update one ticket per signal.

Data Explorer

A read-only SQL console with schema suggestions and natural-language queries.

Learnings

What the loop learned from finished sessions, held for your review. Nothing changes until you accept it.

Built to be trusted with real systems

Read-only where it explores

Investigation runs through guarded, validated routes. Writes are refused outside testing environments.

A human at every gate

Plans need your approval, and merges need a teammate's review. The agent cannot approve its own work.

Bounded cost

Every session runs on a budget, in your AI tool's own unit, with a per-ticket ceiling.

Configured on day one

A shared company profile means a new install arrives with your applications, rules and queries already in place.