Skip to Content

Agent

“Agent” means two different things in Glubean. They share evidence and product context, but they live in different places and serve different users.

Two agents

AgentWhere it livesPrimary usersBest at
Your IDE coding agentCursor, Claude Code, Codex, or another agent you chooseDevelopers and power usersReading source, editing repo files, writing contracts/workflows/tests, running CLI/MCP, repairing code
Glubean webapp AgentInside the Glubean webappQA, product, support, writers, developers reviewing evidenceExplaining the current page, summarizing runs, finding performance risks, helping write docs/tests from webapp context, interpreting failures

The IDE agent is yours. You choose the tool, install the Glubean skill, connect MCP, and let it work inside your local repo.

The Glubean webapp Agent is built into Glubean. It is ambient over the current webapp surface: the Target you are viewing, the run detail, the Specification, the Explore request, the API editor, the Portal draft, or the performance page.

IDE coding agent

Use the IDE agent when the task needs source context.

It can:

  • inspect handlers, types, schemas, fixtures, and repo conventions
  • create or update a Glubean verification project
  • write contract.http.with(), contract.browser.with(), workflow(), test(), and load plans
  • run locally through glubean run, glubean load, or MCP
  • inspect structured failures and repair source or test code
  • prepare changes that belong in git

The important setup is skill-first:

npx skills add glubean/skill

Then ask the agent to set up or work inside the verification project. The skill teaches it Glubean’s SDK patterns, project boundaries, MCP flow, and contract semantics.

Glubean webapp Agent

Use the webapp Agent when the task starts from Cloud evidence or a webapp surface.

It should help answer questions like:

  • “Why did this run fail?”
  • “Which contract promise broke?”
  • “What changed compared with the previous run?”
  • “Does this performance chart show a latency or error-rate risk?”
  • “Which endpoint looks flaky?”
  • “Can you summarize this Target for a release review?”
  • “Can you help polish this API description or Portal page?”
  • “Can you draft a contract, workflow, or test idea from this Specification or Explore request?”

The webapp Agent is strongest when the relevant data is already in Glubean: uploaded runs, structured failures, traces, contract projections, Specifications, Explore requests, Library modules, Environments, Data assets, Files, API overlays, Portal content, reports, and performance history.

It can also draft source-side next steps, such as “add a contract for this missing edge case” or “turn this Explore request into a workflow.” Treat that as a handoff, not an automatic source edit: the IDE plane should review, run, and commit the durable change.

Same evidence, different authority

The two agents should not pretend to be the same tool.

The IDE agent has source authority: it can read and edit the repo, run local commands, and make code changes. It is the right place to create durable source truth.

The webapp Agent has product-surface authority: it can explain the current view, connect evidence across Glubean surfaces, help non-developers reason about API behavior, and assist authoring inside the webapp. It should not pretend to know private source code that was never uploaded or synced.

When webapp work needs source changes, the long-term flow is write-back: turn an approved webapp suggestion into a repo change for the IDE plane to review, run, and commit.

Examples

On a Target page, ask the webapp Agent:

What changed in the last five runs, and which failures are connected to contracts?

On a Performance view, ask:

Which endpoint has the highest p95 risk, and is it getting worse?

On a Specification, ask:

Explain this contract in product language and suggest missing edge cases.

In your IDE, ask your coding agent:

Use the Glubean skill and add executable contracts for the checkout workflow. Run them locally and upload the result.

Boundary

If the task requires source truth, use the IDE plane with MCP, CLI, and the Glubean skill. If the task requires interpreting current product evidence, writing webapp-facing copy, comparing runs, reviewing performance, or helping a non-developer understand API behavior, use the Glubean webapp Agent.

Last updated on