ROCm/aorta

Stars: 11
Forks: 7
Open beginner issues: 0
Indexed issues: 68
Avg merge: 1d 17h
Merged PRs (30d): 31

Dominant language: Python
License: MIT
Last GitHub push: Sep 18, 2026
Latest indexed: Sep 20, 2026
Contributing guide: No contributing guide
Code of conduct: No code of conduct
Beginner labels: No beginner labels indexed

Beginner friendly Everything

Open All

68 open issues indexed

probe: the diagnostic axis has never changed a verdict in any archived matrix, and the cell summary cannot carry the signal that would Open

Difficulty 4/5 3-5 days Newbie friendliness 48/100

ROCm/aorta#516 · Opened Sep 18, 2026
agent: a detector should declare its autopsy category, so the offline heuristic stops inferring one from a free-form name Open

Difficulty 4/5 3-5 days Newbie friendliness 35/100

ROCm/aorta#515 · Opened Sep 18, 2026
registry: nccl_launch_order_implicit segfaults every rank instead of resolving the launch-order deadlock it exists for Open

Difficulty 4/5 3-5 days Newbie friendliness 45/100

ROCm/aorta#512 · Opened Sep 18, 2026
registry: seven of twenty-one built-in mitigations are non-functional on the stack probes run on, in four distinct modes Open

Difficulty 5/5 Over a week Newbie friendliness 35/100

ROCm/aorta#511 · 3 comments · Opened Sep 18, 2026
agent: a grammar-constrained serving path proposes near-full-menu mitigation lists on ~11% of replies, and the iteration budget charges one unit for the whole cycle Open

Difficulty 4/5 3-5 days Newbie friendliness 48/100

ROCm/aorta#510 · Opened Sep 18, 2026
examples/rl: five guards encode "could not check" as "checked and passed", so an unreachable verdict scores as a favourable one Open

Difficulty 4/5 3-5 days Newbie friendliness 32/100

ROCm/aorta#506 · Opened Sep 18, 2026
agent: `winning_mitigation` credits a pass only on `{m}-none`, so the same registered name is a resolver on one probe axis and uncreditable on the other Open

Difficulty 5/5 Over a week Newbie friendliness 35/100

ROCm/aorta#504 · 1 comment · Opened Sep 17, 2026
agent: validate_step silently drops `none` and duplicate names from next_mitigations, so an all-`none` proposal empties the list and the loop records a genuine agent_stop Open

Difficulty 4/5 3-5 days Newbie friendliness 48/100

ROCm/aorta#501 · Opened Sep 17, 2026
`tf32_off` sets an environment variable that nothing reads, and it is the canonical worked example throughout the probe documentation Open

Difficulty 4/5 3-5 days Newbie friendliness 38/100

ROCm/aorta#500 · 2 comments · Opened Sep 11, 2026
ci: the Chat tests aggregator was written to be a required check and was never added to main protection Open

Difficulty 2/5 1-3 hours Newbie friendliness 68/100

ROCm/aorta#499 · Opened Sep 11, 2026
chat: the duplicate-call guard signature only canonicalises top-level kwargs, so nested dicts can re-run a side-effecting tool Open

Difficulty 3/5 1-2 days Newbie friendliness 72/100

ROCm/aorta#495 · Opened Sep 10, 2026
chat: a router reply naming both routes moved from question to action, which the narrowing did not intend Open

Difficulty 5/5 Over a week Newbie friendliness 30/100

ROCm/aorta#494 · Opened Sep 10, 2026
chat: interrupted index builds leak a full copy of the staged index into the cache dir, and nothing reaps it Open

Difficulty 3/5 1-2 days Newbie friendliness 72/100

ROCm/aorta#493 · Opened Sep 10, 2026
chat: two concurrent index builds can cross one build's store with the other's manifest, undetectably Open

Difficulty 5/5 Over a week Newbie friendliness 45/100

ROCm/aorta#492 · Opened Sep 10, 2026
chat: the LLM model settings accept a blank value, unlike the embedding ones Open

Difficulty 4/5 3-5 days Newbie friendliness 45/100

ROCm/aorta#491 · Opened Sep 10, 2026
chat: duplicate keyword arguments in an ACTION: call are silently last-wins Open

Difficulty 2/5 1-3 hours Newbie friendliness 85/100

ROCm/aorta#490 · Opened Sep 10, 2026
chat: a critic retry can mislabel its fallback "no tool ran" and discard the earlier pass's tool_trace Open

Difficulty 4/5 3-5 days Newbie friendliness 48/100

ROCm/aorta#488 · Opened Sep 10, 2026
chat: the _send chokepoint guard misses wildcard imports, and its completeness oracle cannot see a partially-missed binding Open

Difficulty 4/5 3-5 days Newbie friendliness 48/100

ROCm/aorta#487 · Opened Sep 10, 2026
ci: nightly-eval inherits nightly.yml's completion time, so a 46-minute chat-index run silently moves the perf measurement slot by three quarters of an hour Open

Difficulty 4/5 3-5 days Newbie friendliness 48/100

ROCm/aorta#486 · Opened Sep 10, 2026
chat: refresh advice should read corpus provenance before recommending 'index fetch' Open

Difficulty 4/5 3-5 days Newbie friendliness 45/100

ROCm/aorta#483 · Opened Sep 10, 2026
chat: the local model-cache probe reports a truncated cache as warm, and the pre-seed procedure it prints is unquoted Open

bug

Difficulty 3/5 1-2 days Newbie friendliness 74/100

ROCm/aorta#478 · 1 comment · Opened Sep 9, 2026
chat: retriever.py's two index remedies hardcode 'aorta chat index fetch', which a remote embedder can never run Open

bug

Difficulty 3/5 1-2 days Newbie friendliness 68/100

ROCm/aorta#477 · Opened Sep 9, 2026
chat: run_collection_name() appends _runs to an at-cap name, exceeding MAX_COLLECTION_NAME by up to 5 characters Open

bug

Difficulty 3/5 1-2 days Newbie friendliness 75/100

ROCm/aorta#476 · 1 comment · Opened Sep 9, 2026
chat: a give-up that already gathered tool results discards them instead of answering from them Open

Difficulty 4/5 3-5 days Newbie friendliness 67/100

ROCm/aorta#475 · 1 comment · Opened Sep 8, 2026
pytest-timeout is installed but never armed, so a blocking chat test hangs CI instead of failing Open

bug

Difficulty 3/5 1-2 days Newbie friendliness 75/100

ROCm/aorta#474 · Opened Sep 8, 2026
chat: get_retriever() still blocks the event loop, and its offload needs a cache lock first Open

bug

Difficulty 4/5 3-5 days Newbie friendliness 64/100

ROCm/aorta#472 · 2 comments · Opened Sep 8, 2026
chat: `aorta chat doctor --help` enumerates its checks by hand, and the list is now incomplete Open

bug

Difficulty 3/5 1-2 days Newbie friendliness 68/100

ROCm/aorta#471 · 1 comment · Opened Sep 8, 2026
chat: two cli/chat.py handlers still print a rejected credential via pydantic's input_value Open

bug

Difficulty 3/5 1-2 days Newbie friendliness 45/100

ROCm/aorta#470 · Opened Sep 8, 2026
redaction: tool-call arguments bypass the outbound gate, and the native protocol is now reachable by default Open

Difficulty 3/5 1-2 days Newbie friendliness 74/100

ROCm/aorta#469 · Opened Sep 8, 2026
chat: the Chainlit welcome banner names the provider but not the tool protocol Open

Difficulty 2/5 1-3 hours Newbie friendliness 78/100

ROCm/aorta#468 · Opened Sep 8, 2026
chat: manifest.validate() silently skips the version-drift check instead of reporting it skipped Open

bug

Difficulty 4/5 3-5 days Newbie friendliness 58/100

ROCm/aorta#467 · Opened Sep 8, 2026
ci: a dispatched nightly-eval overwrites that date's scheduled ci-results record, destroying a night's observation during a record-only baseline window Open

bug

ROCm/aorta#457 · 1 assignee · Opened Sep 7, 2026
sanitizers: reject consan_policy: strict paired with a non-dispatching driver at recipe-validation time, honouring declared expectations Open

enhancement

Difficulty 5/5 Over a week Newbie friendliness 42/100

ROCm/aorta#455 · Opened Sep 3, 2026
sanitizers: nightly red since 2026-09-02 — rj_waitcheck CFG dataflow does not converge on ROCm 10's gemm_NT_M256_N4096_K1024, so the case yields error against an expected warn Open

bug

ROCm/aorta#453 · 2 comments · 1 assignee · Opened Sep 3, 2026
sanitizers: a ConSan finding names neither its kernel nor its offset, and its code is the raw numeric kind, so a race cannot be attributed Open

bug

Difficulty 4/5 3-5 days Newbie friendliness 45/100

ROCm/aorta#451 · Opened Sep 3, 2026
sanitizers: daily-consan-tiny and daily-consan-gemm pair a load-only driver with consan_policy: strict, so they can never return a verdict Open

bug

Difficulty 2/5 1-3 hours Newbie friendliness 76/100

ROCm/aorta#450 · Opened Sep 3, 2026
agent: an unregistered mitigation name is dropped before validation, so a name-resolution failure is recorded as a genuine agent_stop Open

bug

Difficulty 3/5 1-2 days Newbie friendliness 72/100

ROCm/aorta#449 · Opened Sep 3, 2026
aorta.race.modes: PEP 562 `__getattr__` with no `__dir__`, so `__all__` names are missing from `dir()` Open

Difficulty 2/5 1-3 hours Newbie friendliness 88/100

ROCm/aorta#448 · Opened Sep 3, 2026
chat: embedding collection names are not collision-safe, and the run collection has no model guard Open

Difficulty 4/5 3-5 days Newbie friendliness 52/100

ROCm/aorta#447 · 3 comments · Opened Sep 3, 2026
chat: run_terminal_command has no OS-level isolation -- allowlist plus cwd is not containment Open

enhancement

Difficulty 5/5 Over a week Newbie friendliness 30/100

ROCm/aorta#440 · Opened Sep 2, 2026
proton: remove the roctracer mode: cli guard once the Triton floor reaches 3.8 Open

Difficulty 4/5 3-5 days Newbie friendliness 35/100

ROCm/aorta#439 · Opened Sep 2, 2026
discover_workloads() silently overwrites a duplicate workload name; all four sibling registries refuse the collision Open

bug

Difficulty 4/5 3-5 days Newbie friendliness 45/100

ROCm/aorta#435 · Opened Sep 2, 2026
Three subsystems, three exception bases: a caller cannot catch "aorta failed" as one category Open

question

Difficulty 5/5 Over a week Newbie friendliness 35/100

ROCm/aorta#431 · Opened Sep 2, 2026
Reconcile the two path-safety layers: should the chat tools sit on aorta.run._fsafe? Open

question

Difficulty 5/5 Over a week Newbie friendliness 35/100

ROCm/aorta#430 · Opened Sep 2, 2026
chat: add an `aorta chat` command group -- a LangGraph assistant over the codebase and your run artifacts, behind opt-in extras Open

enhancement

Difficulty 5/5 Over a week Newbie friendliness 20/100

ROCm/aorta#421 · Opened Sep 1, 2026
redaction: outbound scrubbing covers only paths and IP addresses -- no credential or customer-identifier scrubber Open

enhancement

Difficulty 5/5 Over a week Newbie friendliness 35/100

ROCm/aorta#420 · Opened Sep 1, 2026
packaging: py.typed is declared in package-data but the file has never existed, so amd-aorta ships untyped Open

bug

Difficulty 4/5 3-5 days Newbie friendliness 45/100

ROCm/aorta#419 · Opened Sep 1, 2026
test(triage): guard the published `matrix.json` key set so a dropped field fails CI instead of silently misreporting Open

Difficulty 3/5 1-2 days Newbie friendliness 78/100

ROCm/aorta#418 · 1 comment · Opened Sep 1, 2026
sanitizers: the nightly publish job renders the dashboard with an unpinned Python Open

bug

Difficulty 4/5 3-5 days Newbie friendliness 45/100

ROCm/aorta#401 · Opened Aug 25, 2026
Run-area renderers index env.json fields directly, so one malformed manifest fails the whole dashboard render Open

bug

Difficulty 4/5 3-5 days Newbie friendliness 72/100

ROCm/aorta#398 · 2 comments · Opened Aug 25, 2026

A short digest of beginner-friendly GitHub issues.