Skip to Content

Agentic QA / Context → Work → Proof

Autonomous QA That Does the Work—and Leaves the Proof.

Sladdis understands your product, plans a focused QA mission, runs approved checks, and follows failures through verification. Testbench keeps the scope, evidence, decisions, and retests connected.

Explore a Real QA ReportWatch the Product FilmRead the Agent API
  • Project-scoped missions
  • Intent saved before execution
  • Evidence linked to results
  • Findings connected to retests

Film 01 / The Complete QA Loop · 48 Sec

From Release Risk to an Evidence-Backed Decision.

Follow one piece of proof from project intelligence and planning through controlled execution, linked evidence, retest, and the next release action.

01 / From Risk to Release00:48
Music included · no spoken dialogue“Horizons” by Scott Buckley · CC BY 4.0
Read the Film Transcript
  1. 00:00Every Release Starts With Risk.The question is whether the release decision can carry proof all the way through.
  2. 00:04See What Could Break.Release health, coverage gaps, and the next best action—before the run begins.
  3. 00:09Map What Matters.Observed pages, APIs, data, and the boundaries that still need review.
  4. 00:14Turn Risk Into a Mission.One shared context. Focused subagents.
  5. 00:20Run the Right Checks in Parallel.Real browser checks. Bounded execution. Versioned test logic.
  6. 00:27Bring Back Proof. Not Just a Score.Steps, traces, logs, and expected-versus-actual stay linked to the result.
  7. 00:33Fix. Retest. Verify.A linked failure becomes one action. The action and evidence close together.
  8. 00:39Know What Is Safe to Ship.What changed. What remains. What happens next.
  9. 00:44From Risk to Release. With Proof.Sladdis does the work. Testbench keeps it connected.

A 48-second product film in which one evolving proof document moves from release intelligence and application mapping through parallel planning, controlled automation, evidence, retesting, and a release decision. The film contains music and on-screen text but no spoken dialogue.

One QA System

Sladdis Acts. Testbench Remembers.

The agent and the workspace have different jobs. Together they turn a QA request into durable, reviewable work.

Sladdis / The Work Engine

Your Autonomous QA Colleague.

Reads the brief, explores the product, chooses a strategy, creates or selects the right checks, uses approved tools, and owns the final run.

  • 01Understand
  • 02Plan
  • 03Execute

Testbench / The Durable Record

The Control Plane and System of Record.

Holds project knowledge, rules, application structure, run plans, results, evidence, issues, retests, notes, and release signals.

  • 01Context
  • 02Policy
  • 03Proof

The Closed Loop

Every Result Has Somewhere to Go.

What the run reveals becomes linked follow-up, durable project context, and a ranked next action—not a lost chat transcript.

Film 02 / Inside Sladdis · 46 Sec

Watch One Mission Become Verified Work.

“The Mission Case” zooms into the work Sladdis owns: one scoped workspace, product research, approved tools, a configured Docker runner, parallel tracks, and a durable handoff to Testbench.

Read the Agent Workflow
02 / Inside the Agent00:46
Music included · no spoken dialogue“Machina” by Scott Buckley · CC BY 4.0
Read the Film Transcript
  1. 00:00A QA Mission Arrives.A scoped assignment—not another dashboard notification.
  2. 00:04One Mission. One Workspace.Product context, risks, rules, and expected evidence stay together.
  3. 00:10Research Before Execution.Walk the product. Read the docs. Inspect the repository or contract when provided.
  4. 00:16Use What the Mission Allows.Browser, web research, terminal, test code, and fixtures—scoped to the job.
  5. 00:22Real Tools. Defined Boundaries.A versioned package enters a configured Docker runner with an exact target and policy.
  6. 00:29Split the Work When It Helps.Focused tracks work in parallel. Sladdis still owns the final run.
  7. 00:35Results Land in Testbench.Run results, linked evidence, findings, and retest work stay connected to the project.
  8. 00:41The Work Moves On. The Proof Stays.Sladdis does the work. Testbench keeps it connected.

A 46-second product film called The Mission Case showing Sladdis entering a scoped workspace, researching the product with approved tools, running controlled checks in Docker, splitting focused work, and saving evidence and follow-up work in Testbench. The film contains music and on-screen text but no spoken dialogue.

  1. 01 / Scope

    Start With a Mission.

    Target, current risk, rules, and expected proof enter one project-scoped workspace.

  2. 02 / Understand

    Research Before Execution.

    Walk the product and inspect supplied documentation, repository context, or contracts when provided.

  3. 03 / Use Tools

    Match the Tool to the Boundary.

    Browser, web research, terminal, test code, and fixtures stay scoped to the job.

  4. 04 / Run

    Execute Against an Exact Target.

    Configured runners apply origin, package, test-data, logging, and mutation policy.

  5. 05 / Parallelize

    Split Focus, Keep One Owner.

    Focused subagents can investigate separate tracks while Sladdis retains mission ownership.

  6. 06 / Record

    Return Durable Work.

    Run results, linked evidence, findings, and retest work land in Testbench with project and run context.

Autonomy With a Contract

Move Fast Inside Explicit Boundaries.

Each project can define what Sladdis must do, may do, must never do, and what requires human approval. The runtime applies those rules to targets, tools, test data, mutations, devices, and evidence.

  • 01Must
  • 02May
  • 03Never
  • 04Requires Approval

01 / Project Boundary

Project-Scoped Access

The agent exchanges a one-time request secret for a scoped token. Your normal Testbench session is never handed to the agent.

Scoped token · Expiring access

02 / Execution Boundary

Controlled Execution

Exact target origins, configured runners, safe Android devices, package allowlists, bounded logs, and mutation gates keep real tools inside the mission.

Exact target · Approval required

03 / Evidence Boundary

Private by Default

Detailed evidence remains linked to its project, run, and result. Public reports expose recipient-safe conclusions—not secrets or private working context.

Project-linked · Redacted when shared

Real QA Work

Research Like a Tester. Build Like an Engineer. Hand Off Like a Teammate.

Move from an assignment to executable coverage and verified follow-up without turning every step into a manual handoff.

  1. 01 / Understand

    Start With the Product, Not a Blank Prompt.

    Use the project brief, a product walkthrough, and supplied documentation, repository context, or API contracts to understand the critical journey and rank current risk.

    • Project brief
    • Walkthrough
    • Application Map

  2. 02 / Build & Execute

    Create and Run the Checks the Mission Needs.

    Draft Playwright coverage, API checks, fixtures, and small scripts. Run approved browser, Docker, terminal, API, and safe Android workflows against configured targets.

    • Playwright
    • Docker
    • API
    • Android

  3. 03 / Persist & Follow Through

    Return Structured Work, Not a Chat Transcript.

    Save normalized results, screenshots, traces, redacted excerpts, findings, and retests in Testbench. Export reviewed automation packages to GitHub when the team is ready.

    • Evidence
    • Work Board
    • Retests
    • GitHub

Testbench / Durable QA Memory

The QA Memory Shared by Your Agent and Your Team.

Sladdis reads structured context before it tests and writes back after it runs. Quality decisions stay consistent across missions, people, and time.

Open Testbench

Testbench / Run 042

Persisted

Release signal

Hold release.

One covered failure breaks payment confirmation when the session expires.

Passed
11
Failed
1
Blocked
2

Finding / 01

Expired session breaks payment confirmation.

Expected vs actual, evidence, run context, and release impact stay linked.

Retest / 01

Open after fix.

The next check remains visible until it is verified or explicitly deferred.

01

Screenshot

02

Trace

03

Console

04

Network

01

Project Intelligence

Release health, coverage gaps, failures, active work, retests, memory, and the next best action in one brief.

02

Application Map

Connect pages, APIs, data, boundaries, and durable tests so coverage reflects the product.

03

Run Plans

Save scope, risks, selected tests, expected evidence, confidence, and focused tracks before execution.

04

Plan vs Actual

Compare selected and executed checks, surface scope drift, and keep every artifact linked to its result.

05

Work Board & Retests

Route one finding through ownership, fix, retest, history, and verification without duplicating the issue.

06

Agent Notes

Preserve observations, durable memory, next-test ideas, and human replies beside linked runs and tickets.

Real Output / Recipient-Safe Proof

A QA Report Stakeholders Can Act On.

See what was tested, what failed or stayed blocked, which evidence supports the decision, and what should happen next. The films are illustrative. The report is the product output.

  • Release verdict and confidence
  • Tested scope, gaps, and latest run
  • Issues linked to evidence and retests
  • Published findings without private working context
Explore the Real QA Report

Sladdis + Testbench

The Work Moves On. The Proof Stays.

See autonomous QA where every mission leaves a recorded outcome, traceable context, and a clear next action.

Explore a Real QA ReportOpen Testbench