Repository Issue Activity (beta)

#2359 Support for other runtimes like apptainer or enroot

#2570 Served runs rebuild every task as the taskset's declared `Task` type, silently dropping per-case subclasses

#2571 Env-server pool never notices a dead worker; a worker that fails at startup hangs the eval

#2177 Lean taskset: injected `axiom` bypasses the reward (#1271 `#print axioms` follow-up)

#1558 Question about DSPy Flight Example

#2197 v1: apply agentic-judge task weights when an isolated solver fails

#2196 v1: harden capture_patch handling of pre-existing untracked filenames

#2202 v1: mcp utils

#1982 Hub quality scan tests the v0 API, so a v1 taskset can't pass it

#2345 eval CLI persists resolved secrets (default_headers) into the run's config.toml and log

#2198 v1: reject or normalize duplicate artifact sources before collection

#2447 XMLParser.parse_answer is inconsistent between string and message completions

#2321 v1: how should Harbor tasks that provision services via ENTRYPOINT bring up the agent-side environment?

#2354 Support Harbor [environment.env] and [environment.healthcheck]

#2250 v1: Bash harness leaks its uv environment into agent commands

#2319 DockerRuntime uses --network host for unrestricted tasks, silently corrupting grading when host ports collide

#1272 Add braintrust integration for tracing

#2180 Trace failed to push (Request too large)

#2195 v1: make isolated agentic-judge collect generic task artifacts and honor judge image

#2194 v1: enforce artifact bounds during transfer and validate restore archives

#2189 v1: model Harbor artifacts as durable outputs, not only grading transport

#2030 Env server fails to start on Windows: ZMQ ipc:// transport is unsupported

#2150 doc: fix links on the prime intellect docs site

#1871 User simulator must be an explicit harness feature (transparent injection forks the graph with tools)

#1477 Enable network=false for v1 sandboxes