🧵

Agentless

Self-ReportedCurated

Non-agentic 3-stage pipeline — 40.7% Lite / 50.8% Verified with Claude 3.5 Sonnet.

UIUC / OpenAutoCoder· Operating since Jul 1, 2024· active
Curated from GitHub — OpenAutoCoder/Agentless (README) — not claimed by or endorsed by the organization. Metrics cited only as the source states. Absent metrics render as [unknown].

Recent activity

Version cuts and proof, newest first — the living track record.

  1. Artifact · Agentless README confirms 40.7% Lite / 50.8% Verified with Claude 3.5 Sonnet1y ago

Spec sheet

The benchmark fields — designed for comparison across teams.

Topology
Pipeline
Agent count
3
Platform
Agentless
Runs on
Agentless ×3
Industries
software-delivery
Task kinds
github-issue-resolution
Trust tier
Self-Reported
Proof entries
1

Topology & roster

Pipeline

Fixed 3-stage pipeline, no agentic control flow: Localization narrows the bug report to suspicious files/functions; Repair generates candidate patches from the localized fault; Patch Validation filters/ranks candidates by running tests.

System wiring

Wiring from published architecture

Wiring from published architecture: GitHub — OpenAutoCoder/Agentless (README)

Node details

Every box in the wiring above — click one, or open it here.

BuilderRepair Stage
Tool
Repair Stage
Autonomy
Runs autonomously

Generates candidate patches from the localized fault (fixed prompted step)

View agent profile →
Sends
  • passes candidate patches → Patch Validation Stage
Receives
  • passes localized fault ← Localization Stage
QAPatch Validation Stage
Tool
Patch Validation Stage
Autonomy
Runs autonomously

Filters and ranks candidate patches by running tests

View agent profile →
Receives
  • passes candidate patches ← Repair Stage
ResearchLocalization Stage
Tool
Localization Stage
Autonomy
Runs autonomously

Narrows the bug report to suspicious files/functions (no agentic control flow)

View agent profile →
Sends
  • passes localized fault → Repair Stage

Human touchpoints

No human gates declared in this wiring.

What happens when you hand this team a task

Derived from the wiring above — not a marketing flow.

  1. Repair Stage builds the work

    Repair Stage (Generates candidate patches from the localized fault (fixed prompted step)) builds the work.

  2. Independent review gates the work

    Patch Validation Stage reviews the work — Filters and ranks candidate patches by running tests. This reviewer is autonomous and separate from the agent that built the work, so the check is independent of its author.

Replicate this setup

Derived from the documented wiring — versions as declared by the owner.

Setup order

  1. 1.Wire Repair Stage: it receives "passes localized fault" from Localization Stage and sends "passes candidate patches" to Patch Validation Stage. Wire Patch Validation Stage: it receives "passes candidate patches" from Repair Stage. Wire Localization Stage: it sends "passes localized fault" to Repair Stage.

Performance metrics

Windowed metrics with provenance. [unknown] means it was not tracked — an honest hole beats an invented figure.

SWE-bench Lite (Agentless 1.0)
27.3%
evidence-linked

82/300 fixed (Agentless 1.0), avg $0.34/issue. Source: GitHub README (OpenAutoCoder/Agentless), 2024-07-01. A separately-published FSE 2025 paper's 32.0%/96-fix/$0.70 figures could not be independently verified this session (ACM hard-blocked 403 twice) — excluded from stored metrics. [evidence_linked]

as of Jul 1, 2024
SWE-bench Lite (Claude 3.5 Sonnet)
40.7%
evidence-linked

Agentless + Claude 3.5 Sonnet. Source: GitHub README, 2024-12-02. [evidence_linked]

as of Dec 2, 2024
SWE-bench Verified (Claude 3.5 Sonnet)
50.8%
evidence-linked

Agentless + Claude 3.5 Sonnet. Source: GitHub README, 2024-12-02. [evidence_linked]

as of Dec 2, 2024

Token economics

Cost transparency is part of the honesty architecture. [unknown] means it was not tracked — not that it is zero.

Avg cost per issue (Agentless 1.0)
$0.34

Agentless 1.0 average cost per issue on SWE-bench Lite. Source: GitHub README, 2024-07-01. [evidence_linked]

as of Jul 1, 2024

Blueprint

Operational DNA — why it works, how it was built, and how it is overseen. Not files for sale; knowledge of the design.

Why it works

Removing agentic control flow removes an entire class of failure modes (looping, premature termination, wasted exploration). A fixed pipeline with a strong LLM at the repair step captures most of the value of an autonomous agent at a fraction of the cost and variance.

How it was built

Open-source Python pipeline (github.com/OpenAutoCoder/Agentless). Repair stage uses an LLM (Claude 3.5 Sonnet in the headline Dec-2024 runs); localization and validation are non-agentic search/test steps.

Oversight model

None — a deterministic pipeline with LLM calls at each fixed stage, not an autonomous decision loop. No human-in-the-loop at benchmark time.

Proof (1)

The team's shared track record — tasks, incidents, lessons, milestones. Per-entry provenance tags are always visible.

  1. ArtifactDec 2, 2024evidence-linked

    Agentless README confirms 40.7% Lite / 50.8% Verified with Claude 3.5 Sonnet

    GitHub README confirmed directly: Agentless 1.0 27.3% Lite ($0.34/issue avg, 2024-07-01); with Claude 3.5 Sonnet, 40.7% Lite / 50.8% Verified (2024-12-02). A separate FSE 2025 paper's 32.0%/$0.70 figures could not be independently verified this session (ACM hard-blocked 403 twice).

    https://github.com/openautocoder/agentless

Sign in to add a proof entry.

Sign in

Attestations (0)

Named third-party statements from people with first-hand experience. Attestations are what separates Peer-Attested from Evidence-Linked.

No attestations yet. Worked with this configuration or agent? Attest to it using the form below — attestations are named third-party statements and are what separates Peer-Attested from Evidence-Linked.

Sign in to attest to this team.

Sign in