Research report · data-report
Debugs and fixes: a census of 126 listings
35 of 126 listings answer yes for “Debugs and fixes”. Counted from our own datasheet, with the query that reproduces every figure.
On this page
Debugs and fixes, across 126 settled listings
Of the 126 Published listings whose datasheet settles “Debugs and fixes”, the largest group is “Does not apply” at 53 (42%). Measured 2026-09-16.
Explore the matching agents
Debugs and fixes · Datasheet criterion debugs_code · Report counted .
Current listings as of ; this set may differ from the report's original sample. Includes “Yes” and “Qualified” answers only. Read the conditions on qualified answers before treating them as a match. “No”, “Not applicable” and “Not established” answers are excluded.
View 63 current matching records
- Agentix Labs Qualified
It diagnoses and hardens an existing agent deployment during a teardown; no page offers debugging of a client's general application code.
Source retrieved 2026-09-09
- AgentsKit.js Qualified
The documented self-debug pattern feeds a failed run's own trace back to an agent to diagnose and retry; nothing describes diagnosing bugs in your application code.
Source retrieved 2026-09-09
- Agno Yes
Bug fixing is a documented pattern: reproduce, patch, rerun the failing tests.
- Aider Yes
Runs your linter and test command after each edit and attempts the repair itself whenever the command exits non-zero, so failures it introduced get diagnosed and fixed in the same session.
Source retrieved 2026-09-09
- Aisera Agent Studio Qualified
The code-generation line claims it identifies and fixes errors using engineering-domain models, but nothing in the vendor's description lands that fix anywhere: there is no repository, branch or commit in the flow.
Source retrieved 2026-09-03
- Antigravity CLI Yes
A dedicated /boost command escalates hard defects to multi-agent reasoning, and the docs push you to give the agent a test command so it iterates on failures itself.
Source retrieved 2026-09-10
- Augment Code Yes
PR Fixer repairs CI failures and review findings; a CI Failure Investigator expert diagnoses failed runs.
Source retrieved 2026-09-02
- AutoGen Qualified
Within a run the agent retries failed executions and reflects on the result; nothing diagnoses a failure in an existing repository or lands a fix there.
Source retrieved 2026-09-02
- Bitsclan IT Solutions Qualified
Defect triage and repair is a contracted human service, and the only published turnaround commitment sits in the white-label SLA rather than in the AI agent engagement itself.
Source retrieved 2026-09-11
- Bland Qualified
Norm reproduces a failing call, finds the root cause and applies the fix, but the fix lands in a Bland pathway, not in your codebase.
- Bolt Yes
The agent troubleshoots as it builds and can apply fixes, though for runtime bugs the docs tell you to paste browser console logs back in.
Source retrieved 2026-09-11
- Botpress Qualified
Agent(0), the assistant in the ADK Dev Console, edits actions and workflows and debugs build and runtime errors in your own ADK project, not an arbitrary repository or a CI failure.
Source retrieved 2026-09-02
- Browserbase Qualified
Documents a CI recipe where a coding agent inspects the live DOM on Browserbase and fixes broken selectors; the fix comes from Claude Code, not Browserbase.
- Claude Agent SDK Yes
The vendor's own quickstart is an agent that locates crash bugs in a file and edits the fix in.
Source retrieved 2026-09-02
- Claude Code Yes
Takes an error message or symptom, traces it through the repository, and implements the fix.
Source retrieved 2026-09-02
- Cline Yes
Diagnosis and repair are advertised as one loop: it reads stack traces or failing tests, then edits the files and re-runs commands to confirm the fix.
Source retrieved 2026-09-12
- CodeRabbit Yes
It reads failing CI logs on the pull request and posts inline fixes on the offending lines, and a Fix CI finishing touch can land those fixes as a commit or stacked PR.
Source retrieved 2026-09-15
- Codex CLI Yes
Documented first-task workflow includes debugging an issue.
Source retrieved 2026-09-02
- Crescendo Qualified
Debugging is scoped to the integrations it builds: failures in a tool run are inspected, the logic updated and rerun in the same workspace, rather than diagnosing bugs in customer software.
Source retrieved 2026-09-12
- Cresta Qualified
Conductor diagnoses production failures and makes the change once a human approves, but the fix lands in the Cresta agent configuration, not in your repository.
Source retrieved 2026-09-02
- CrewAI Qualified
A code-executing agent receives the exception from its own generated code and retries, capped by max_retry_limit; nothing documents diagnosing or fixing a failure in an existing repository.
Source retrieved 2026-09-13
- Cursor Yes
Debug Mode instruments the code and reasons from runtime evidence before editing, rather than guessing at a fix.
Source retrieved 2026-09-02
- Devin Yes
Reproducing and fixing bugs is named as a core task, including bugs arriving as tickets or bug reports.
Source retrieved 2026-09-02
- Devin Desktop Qualified
End-to-end debugging is claimed for the delegated cloud Devin agent, which is bundled with Pro, Max and Teams but absent from Free.
Source retrieved 2026-09-03
- Dust Qualified
Marketed for debugging by surfacing code context, docs and past issues to the engineer; nothing documents the agent landing the fix itself.
Source retrieved 2026-09-09
- ElevenLabs Agents Qualified
Bug fixing is delegated: the agent starts and steers a Cursor background agent, which diagnoses and lands the change; ElevenLabs itself never touches the repository.
Source retrieved 2026-09-02
- Factory Yes
Documented flows include fixing a named bug and repairing failing tests under autonomy levels.
Source retrieved 2026-09-02
- Gemini Code Assist Yes
Agent mode and the Gemini CLI run a reason-and-act loop over local tools to diagnose and fix; the IDE chat is documented for debugging help.
Source retrieved 2026-09-02
- Genspark Qualified
The vendor claims the agent iterates through its own plan-code-test loop and can start a project from existing code, but no page documents diagnosing and fixing a defect in a repository you bring.
Source retrieved 2026-09-15
- GitHub Copilot Yes
The CLI and cloud agent both diagnose failures and land the fix as a commit or pull request; the cloud agent's own task list names bug fixing.
Source retrieved 2026-09-02
- Glean Yes
Vendor names bug fixes as a target task; the Resolve Jira ticket agent turns an issue into a review-ready PR.
Source retrieved 2026-09-02
- Google Agent Development Kit (ADK) Qualified
The bundled web UI and event/trace inspection let a developer diagnose why an agent failed, but landing the code fix is left to the developer or to a separate AI coding tool; ADK ships no repair agent.
Source retrieved 2026-09-15
- goose Yes
The vendor documents goose reading failed CI runs through the GitHub CLI and then applying and staging fixes itself; that walkthrough is flagged as written against a beta build, but the same shell and edit tools ship today.
Source retrieved 2026-09-16
- Juggler Yes
The documented one-shot example is diagnosing and fixing a failing test.
- Junie Yes
Debug mode drives a live debugger in a connected JetBrains IDE: breakpoints, runtime state, expression evaluation.
Source retrieved 2026-09-02
- Kilo Code Yes
Debug mode reads errors, traces the issue and proposes the fix.
Source retrieved 2026-09-02
- LangChain Qualified
LangSmith Engine diagnoses failures and proposes fixes, but the pricing table marks Engine N/A on the Developer plan.
Source retrieved 2026-09-02
- LangGraph Qualified
LangGraph itself does not diagnose code; the separate LangSmith Engine product analyses LangGraph agent traces, proposes fixes and can open a pull request with one.
Source retrieved 2026-09-02
- Letta Yes
Vendor states coding agents debug failures and run test suites.
Source retrieved 2026-09-02
- Lindy Yes
Lindy Build runs QA agents that exercise the app and repair what they break; no equivalent claim covers code you wrote yourself.
Source retrieved 2026-09-02
- Manus Yes
Runs what it builds in its own browser, finds the failure and fixes it without being asked.
- MindStudio Qualified
The Debugger diagnoses a failing agent workflow block by block with logs and runtime variables, but the human applies the fix, and it never looks at source code in a repository.
Source retrieved 2026-09-02
- Moveworks Qualified
It diagnoses misconfigurations in the agents you build in Agent Studio, not failures in your own codebase.
- Muse Code Yes
The vendor names debugging as a core job, and a background verification observer checks the agent actually ran the work it claims it finished.
Source retrieved 2026-09-02
- n8n Qualified
It diagnoses and repairs the workflows it builds and troubleshoots node execution errors; the failure it fixes is an execution, not a bug in your repository.
Source retrieved 2026-09-02
- OpenAI Agents SDK (Python) Yes
The documented sandbox use case is orchestrating automated fixes for GitHub issue reports and running targeted tests.
Source retrieved 2026-09-02
- OpenCode Yes
Asked to fix an issue, it works in a new branch and submits a PR with the changes.
Source retrieved 2026-09-02
- OpenHands Yes
Investigates failing builds, tests and production errors and can implement the fix, not just describe it.
Source retrieved 2026-09-02
- Perplexity Yes
Computer is sold as fixing code from a plain-language instruction; the sandbox docs list reproducing an error as a use case.
Source retrieved 2026-09-02
- PolyAI Qualified
Wren diagnoses production conversations (tool calls, latency, drop-offs) and can implement the fix on a branch; the subject is agent behaviour, not application code.
Source retrieved 2026-09-02
- Pydantic AI Yes
The harness is written for exactly this: an agent with files, shell and planning turned loose to fix a codebase over a long session.
Source retrieved 2026-09-02
- Replit Agent Yes
Agent runs its own tests, reports what it found and fixes the issues itself.
Source retrieved 2026-09-02
- SafeNet Creations Qualified
Defect correction is a written commitment inside the accepted scope only; diagnosing failures in systems SafeNet did not build is nowhere offered as a service.
Source retrieved 2026-09-04
- Salesforce Agentforce Yes
Vibes runs the test or deployment, reads the failure and iterates toward a fix rather than stopping at the error.
Source retrieved 2026-09-03
- Sierra Qualified
Ghostwriter diagnoses its own failed simulations and implements the fix; the failure it repairs is agent behaviour, never a defect in your software.
Source retrieved 2026-09-03
- Sim Qualified
Babysit Mode reads failing checks and review threads and runs bounded fixing rounds; no general 'diagnose a production failure' claim.
- TecAdRise Qualified
They diagnose and repair the automations they built and operate, fixing broken integration nodes after upstream API changes, but no service is offered for debugging a client's own application code.
Source retrieved 2026-09-12
- Vecbase Yes
Takes a stack trace or error log and returns a root-cause report with a proposed patch and verification checklist.
Source retrieved 2026-09-03
- Vellum Qualified
Diagnosis is in-house (read-only researcher subagents do root-cause analysis); landing a fix in a real repo is handed to an external coding agent you connect through ACP.
Source retrieved 2026-09-03
- Warp Yes
Local agents diagnose failures in a real terminal; in CI the agent pulls failure logs, attempts a root-cause fix and opens a PR with it.
Source retrieved 2026-09-03
- Zed Yes
Debugging is one of the uses Zed lists for the Agent Panel, alongside generation and refactoring.
Source retrieved 2026-09-03
- Zencoder Yes
A dedicated Fix a Bug workflow investigates root cause, writes a solution document for review, then implements the fix and confirms the regression is gone.
Source retrieved 2026-09-09
- zot Qualified
No dedicated debugging feature; the vendor names failure investigation only as an example of a task you can hand a background sub-agent, which then edits the same working directory.
The population
126 listings in this index are Published. This is the population as of 2026-09-16. 126 of the 126 published listings in this index carry a settled answer for “Debugs and fixes”. That row asks: Can it diagnose a failure and land the fix? Every listing in the index is settled on this row, so nothing is left out.
Findings
-
35 of the 126 settled listings answer “Yes” for “Debugs and fixes”. That is 28% of the settled set.
-
28 of the 126 settled listings answer “Qualified” for “Debugs and fixes”. That is 22% of the settled set.
-
6 of the 126 settled listings answer “No” for “Debugs and fixes”. That is 5% of the settled set.
-
53 of the 126 settled listings answer “Does not apply” for “Debugs and fixes”. That is 42% of the settled set.
-
4 of the 126 settled listings answer “Not established” for “Debugs and fixes”. That is 3% of the settled set.
1 of those yes answers is goose, which still answers yes for “Debugs and fixes”. Its listing carries the stored answer and links to the source recorded for it. This is one worked example, not an independent audit of every source in the census.
What we counted, and how
Each figure above is a count over the “Debugs and fixes” row of the listing datasheet, taken from the same stored answer the listing page renders. The denominator is the 126 listings whose answer is settled, meaning one of yes, qualified, no, does not apply, not established. On this row that is the whole index, because every listing has a recognised stored answer. Every number here is stored with the read-only query that reproduces it and re-run every sixty seconds against the live corpus, so a figure that stops reproducing surfaces as drift rather than as a stale sentence nobody notices. Our full method covers how a datasheet row is settled in the first place.
Limitations
This counts stored datasheet answers, not independently tested capabilities. A sourced answer can record a vendor statement or our reading of published evidence; this census does not re-fetch those sources. “Not established” means we have not established an answer. That can reflect vendor nondisclosure, blocked evidence, or unfinished research, not a no. An absent or unrecognised answer is excluded rather than treated as a researched finding. The figures are restated when the stored corpus changes; the date above is the count used for this published version, not a new verification of the vendors.
Get the next report
New agents rankings and fresh data reports. One short email, one-click unsubscribe.