Internal Agents Map

Deflaker

Company
Dropbox
Approach type
Agent
Work
Coding
Human involvement
Unknown
Invocation
Event-driven
Deployment stage
Deployed
Evidence strength
Detailed primary
Entry reviewed

Purpose

Deflaker investigates and repairs flaky tests for Dropbox engineers using Nova.

Check the reviewed sources and claim details.

How it works

Representative workflow: Athena flaky-test detection → CI-validated fix or capped attempts. Research details

Collect evidence

Athena identifies flakiness; Deflaker gathers passing and failing logs.

Propose a repair

Nova diagnoses the likely cause and proposes a change.

Retry or finish

Failed validation starts another attempt with new logs and prior notes; the workflow lands a fix or stops after five attempts.

Where people stay involved

Each scope pairs its normal attention boundary with supporting evidence. See the supervision definitions for the level mapping and limits.

Unreported: The source describes an automated fix-and-validate loop and publication outside the agent, but does not state the required human approval or exception-handling policy.

  • Athena flaky-test detection CI-validated fix or capped attempts

    Unknown · Level unknown

Catalog interpretation: Unclassified for Athena flaky-test detection CI-validated fix or capped attempts; human attention boundary: unknown.

Observed in 2026

Implementation details

Model
Not reportedNot documented for this subject in the reviewed source.
Harness
Durable remediation workflow calling Nova; publication and CI remain outside the coding agent.
Sandbox
Not reportedNot documented for this subject in the reviewed source.
Tool access
Not reportedNot documented for this subject in the reviewed source.
Knowledge
Not reportedNot documented for this subject in the reviewed source.
Context management
Prior attempt notes and new test logs accompany retries.
Credentials
Not reportedNot documented for this subject in the reviewed source.
Interfaces
Not reportedNot documented for this subject in the reviewed source.

Validation and failure handling

Validate repeatedly

CI repeats the test at least 100 times, depending on its failure rate.

Reported observations

The catalog records what the sources report, with the scope and the denominator of every figure. A qualification below limits the figure it sits under.

Unreported: The reviewed source does not document this for the named subject.

Lessons

Unreported: The reviewed source does not document this for the named subject.

Sources and research details

Citations link to the original publisher. Each source also keeps a preserved copy in the repository, so a changed or removed page stays checkable.

  1. Introducing Nova, our internal platform for coding agentshttps://dropbox.tech/machine-learning/introducing-nova-our-internal-platform-for-coding-agentsEngineering blog · First party · Last source verification: 2026-08-31
Question coverage and scope
purpose
Reported
workflow
Reported
human involvement
Unreported: The source describes an automated fix-and-validate loop and publication outside the agent, but does not state the required human approval or exception-handling policy.
implementation
Reported
validation
Reported
observations
Unreported: The reviewed source does not document this for the named subject.
lessons
Unreported: The reviewed source does not document this for the named subject.
Research details for every claim on this page
  1. Summary
    Statement type
    Fact
    Provenance
    Reported
    Confidence
    High
    Confidence reason
    A linked first-party source states the claim.
  2. Harness
    Statement type
    Fact
    Provenance
    Reported
    Confidence
    High
    Confidence reason
    A linked first-party source states the claim.
  3. Context management
    Statement type
    Fact
    Provenance
    Reported
    Confidence
    High
    Confidence reason
    A linked first-party source states the claim.
  4. Collect evidence
    Statement type
    Fact
    Provenance
    Reported
    Confidence
    High
    Confidence reason
    A linked first-party source states the claim.
  5. Propose a repair
    Statement type
    Fact
    Provenance
    Reported
    Confidence
    High
    Confidence reason
    A linked first-party source states the claim.
  6. Validate repeatedly
    Statement type
    Fact
    Provenance
    Reported
    Confidence
    High
    Confidence reason
    A linked first-party source states the claim.
  7. Retry or finish
    Statement type
    Fact
    Provenance
    Reported
    Confidence
    High
    Confidence reason
    A linked first-party source states the claim.
  8. Operating model assessment
    Statement type
    Inference
    Provenance
    Catalog judgment
    Confidence
    Unverified
    Confidence reason
    The source describes an automated fix-and-validate loop and publication outside the agent, but does not state the required human approval or exception-handling policy.
    Observation date
    2026
Back to agents
Know an internal agent?

Share a resource or public mention, suggest an addition, or correct an existing entry.