Datasets:
instance_id stringclasses 2
values | repo stringclasses 1
value | messages listlengths 18 40 | trajectory_format stringclasses 1
value | exit_status stringclasses 2
values | duration_sec float64 0 0 |
|---|---|---|---|---|---|
tilt-dev__tilt-6246 | tilt-dev/tilt | [
{
"role": "system",
"content": "You are an autonomous software-engineering assistant operating a shell to resolve a repository issue.\n\nRespond with a short THOUGHT explaining your reasoning, then EXACTLY ONE bash command in a fenced\ncode block. Keep all prose in the THOUGHT; never add `#` comments inside... | mini-swe-agent-1 | incomplete | 0 |
tilt-dev__tilt-6246 | tilt-dev/tilt | [
{
"role": "system",
"content": "You are an autonomous software-engineering assistant operating a shell to resolve a repository issue.\n\nRespond with a short THOUGHT explaining your reasoning, then EXACTLY ONE bash command in a fenced\ncode block. Keep all prose in the THOUGHT; never add `#` comments inside... | mini-swe-agent-1 | Submitted | 0 |
tilt-dev__tilt-6246 | tilt-dev/tilt | [
{
"role": "system",
"content": "You are an autonomous software-engineering assistant operating a shell to resolve a repository issue.\n\nRespond with a short THOUGHT explaining your reasoning, then EXACTLY ONE bash command in a fenced\ncode block. Keep all prose in the THOUGHT; never add `#` comments inside... | mini-swe-agent-1 | incomplete | 0 |
tilt-dev__tilt-6246 | tilt-dev/tilt | [
{
"role": "system",
"content": "You are an autonomous software-engineering assistant operating a shell to resolve a repository issue.\n\nRespond with a short THOUGHT explaining your reasoning, then EXACTLY ONE bash command in a fenced\ncode block. Keep all prose in the THOUGHT; never add `#` comments inside... | mini-swe-agent-1 | incomplete | 0 |
tilt-dev__tilt-6246 | tilt-dev/tilt | [
{
"role": "system",
"content": "You are an autonomous software-engineering assistant operating a shell to resolve a repository issue.\n\nRespond with a short THOUGHT explaining your reasoning, then EXACTLY ONE bash command in a fenced\ncode block. Keep all prose in the THOUGHT; never add `#` comments inside... | mini-swe-agent-1 | incomplete | 0 |
tilt-dev__tilt-6246 | tilt-dev/tilt | [
{
"role": "system",
"content": "You are an autonomous software-engineering assistant operating a shell to resolve a repository issue.\n\nRespond with a short THOUGHT explaining your reasoning, then EXACTLY ONE bash command in a fenced\ncode block. Keep all prose in the THOUGHT; never add `#` comments inside... | mini-swe-agent-1 | incomplete | 0 |
tilt-dev__tilt-6246 | tilt-dev/tilt | [
{
"role": "system",
"content": "You are an autonomous software-engineering assistant operating a shell to resolve a repository issue.\n\nRespond with a short THOUGHT explaining your reasoning, then EXACTLY ONE bash command in a fenced\ncode block. Keep all prose in the THOUGHT; never add `#` comments inside... | mini-swe-agent-1 | Submitted | 0 |
tilt-dev__tilt-6246 | tilt-dev/tilt | [
{
"role": "system",
"content": "You are an autonomous software-engineering assistant operating a shell to resolve a repository issue.\n\nRespond with a short THOUGHT explaining your reasoning, then EXACTLY ONE bash command in a fenced\ncode block. Keep all prose in the THOUGHT; never add `#` comments inside... | mini-swe-agent-1 | incomplete | 0 |
tilt-dev__tilt-6246 | tilt-dev/tilt | [
{
"role": "system",
"content": "You are an autonomous software-engineering assistant operating a shell to resolve a repository issue.\n\nRespond with a short THOUGHT explaining your reasoning, then EXACTLY ONE bash command in a fenced\ncode block. Keep all prose in the THOUGHT; never add `#` comments inside... | mini-swe-agent-1 | incomplete | 0 |
tilt-dev__tilt-6246 | tilt-dev/tilt | [
{
"role": "system",
"content": "You are an autonomous software-engineering assistant operating a shell to resolve a repository issue.\n\nRespond with a short THOUGHT explaining your reasoning, then EXACTLY ONE bash command in a fenced\ncode block. Keep all prose in the THOUGHT; never add `#` comments inside... | mini-swe-agent-1 | incomplete | 0 |
End of preview. Expand in Data Studio
swe-cleanroom-trajs
Procedurally generated SWE-agent trajectories — no LLM, no GPU — as a clean-room counterpart to
AlienKevin/SWE-ZERO-12M-trajectories (which rolls out a 1.7B model). ~40k trajectories across 124 repos
(5 languages), mini-swe-agent-1 format.
How it's generated (no LLM):
- TOOL rhythm: a per-step 2nd-order Markov chain fit on
ricdomolm/mini-coder-trajs-400k(verb sequence). - ARGUMENTS: a tree-sitter code graph + the gold PR patch decide which file/symbol/lines (privileged generation).
- OBSERVATIONS: real — commands executed against a real checkout (execution-free POSIX, no Docker).
- THOUGHTS: grounded templates (the one synthetic part).
- Priors: gold-file convergence, ~20% struggle/fail, grounded backtracking, best-first triage.
Clean-room: structure, search targets, edits, and observations use only the public task source
(nebius/SWE-rebench-V2-PRs) and the real repos — zero data from SWE-ZERO-12M.
Purpose: a free baseline to train against and compare downstream (SWE-bench Verified) vs LLM-generated data.
- Downloads last month
- 58