Issue resolution in a repository
Reproduce and fix a reported bug so hidden tests pass, in one or several languages.
20 project-authored proposals · taxonomy v1.1 · 0 new benchmark runs. Verifiers and partial-credit rules below are designs, not executed results.
- Typical applications
- git · pytest · Maven · Go
- Class GUI-need label
CL· CLI-sufficient
Digital-employee context
Project class assignments, not demonstrated employee capability or imported benchmark coverage. Acceptance criteria and value metrics are proposals; no ROI has been measured.
Related roles
- IT and software delivery · Separate deliverables cover service tickets, restored configuration, identity records, repository fixes, passing CI and release packages.
Task proposals · 20
Expand to inspect the complete original wording and proposed checks. Original proposal text remains English in both interfaces.
R02.01.01.01-T01 Fix an HTTP client that leaks the Authorization header on cross-host redirects
Original instruction
Repo `/work/httpkit` (Python, pinned commit). Issue #212: when a request with an Authorization header is redirected to a different host, the header is forwarded. Reproduce it with the local redirect test server in `tests/servers/`, fix the client so credentials are stripped on cross-origin redirects (same-host redirects keep them), and keep the public API unchanged.
- Platforms
lin· Linux- Applications
- git · pytest
- Interfaces
term· Terminal- GUI need · task setting
CL· CLI-sufficient- GUI need · ordinary access
CL· CLI-sufficient- Interaction primitives
code· Code & commands- Work product
WP06· Code/engineering- Conditions
- None declared
- Proposed difficulty
- 2 / 5
Proposed verifier
Hidden FAIL_TO_PASS tests (6) for cross-host, scheme downgrade and same-host cases pass; PASS_TO_PASS suite (412 tests) stays green; diff limited to src/ (tests dir changes ignored).
Proposed partial credit
60% fail-to-pass tests (scaled); 30% pass-to-pass green; 10% public API unchanged (signature check)
R02.01.01.01-T02 Resolve a Django ORM crash combining only() with select_related()
Original instruction
In a Django checkout at a pinned commit, the issue text in `ISSUE.md` reports a crash when `.only()` names a field on a related model reached through `select_related()` with a reverse one-to-one. Write the fix in `django/db/models/`, and add a regression test in `tests/queries/`.
- Platforms
lin· Linux- Applications
- git · Django test runner
- Interfaces
term· Terminal- GUI need · task setting
CL· CLI-sufficient- GUI need · ordinary access
CL· CLI-sufficient- Interaction primitives
code· Code & commands- Work product
WP06· Code/engineering- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
Hidden FAIL_TO_PASS tests (3) in tests/queries pass; full tests/queries and tests/select_related suites stay green; agent's own regression test fails on the base commit and passes on the patch.
Proposed partial credit
60% fail-to-pass; 25% pass-to-pass; 15% agent test discriminates base vs patch
R02.01.01.01-T03 Fix a DST off-by-one in a TypeScript date library
Original instruction
Repo `/work/tz-dates` (TypeScript, pnpm). Issue: `addDays(date, 1)` returns 23:00 of the same day when crossing the spring-forward DST transition in `Europe/Lisbon`. Fix the implementation for all IANA zones without changing behavior for UTC inputs. Run tests with `TZ` set to several zones.
- Platforms
lin· Linuxmac· macOS- Applications
- pnpm · vitest
- Interfaces
term· Terminal- GUI need · task setting
CL· CLI-sufficient- GUI need · ordinary access
CL· CLI-sufficient- Interaction primitives
code· Code & commands- Work product
WP06· Code/engineering- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
Hidden tests run under TZ=Europe/Lisbon, America/Sao_Paulo, Australia/Lord_Howe and UTC: 14 DST-boundary cases pass; existing 600-test suite green; tsc --noEmit clean.
Proposed partial credit
50% DST cases (scaled); 30% existing suite; 20% type check clean
R02.01.01.01-T04 Eliminate a deadlock in a Go worker pool under the race detector
Original instruction
Repo `/work/jobq` (Go 1.22). Issue: `Pool.Shutdown()` hangs about 1 in 50 runs when jobs are still being submitted, and `go test -race` reports a data race in `pool.go`. Fix both without removing the graceful-drain semantics described in `docs/shutdown.md`.
- Platforms
lin· Linux- Applications
- go · git
- Interfaces
term· Terminal- GUI need · task setting
CL· CLI-sufficient- GUI need · ordinary access
CL· CLI-sufficient- Interaction primitives
code· Code & commands- Work product
WP06· Code/engineering- Conditions
- None declared
- Proposed difficulty
- 4 / 5
Proposed verifier
Hidden stress test runs 500 iterations with -race and -timeout 120s: zero hangs, zero race reports; drain-semantics tests (jobs accepted before Shutdown complete) pass; go vet clean.
Proposed partial credit
40% no deadlock in stress run; 30% race-free; 30% drain semantics preserved
R02.01.01.01-T05 Patch an integer overflow in a Rust chunked-format parser
Original instruction
Crate `/work/chunkread` (Rust). Issue #77 includes a 12-byte file that makes `Reader::next_chunk` panic with 'attempt to add with overflow' in debug and read out of bounds logic in release. Return a proper `Error::Malformed` instead, for this and similar length fields.
- Platforms
lin· Linuxmac· macOS- Applications
- cargo · git
- Interfaces
term· Terminal- GUI need · task setting
CL· CLI-sufficient- GUI need · ordinary access
CL· CLI-sufficient- Interaction primitives
code· Code & commands- Work product
WP06· Code/engineering- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
Hidden tests: the reproducer and 8 other crafted headers return Error::Malformed; cargo test green in debug and release; a 5-minute cargo-fuzz run on the provided target finds no panic.
Proposed partial credit
40% reproducer fixed; 30% similar fields hardened; 30% fuzz run clean
R02.01.01.01-T06 Fix Optional deserialization in a Java JSON mapper module
Original instruction
Maven project `/work/jsonmap` (Java 17). Issue: fields declared as `Optional<String>` deserialize to `null` instead of `Optional.empty()` when the JSON key is absent, but to `Optional.empty()` when the value is `null`; both should be `Optional.empty()`. Fix it in the `jdk8-types` module.
- Platforms
lin· Linuxwin· Windows- Applications
- Maven · JUnit
- Interfaces
term· Terminal- GUI need · task setting
CL· CLI-sufficient- GUI need · ordinary access
CL· CLI-sufficient- Interaction primitives
code· Code & commands- Work product
WP06· Code/engineering- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
`mvn -pl jdk8-types test` with hidden tests: absent, null and present cases for Optional, OptionalInt and nested Optional fields pass; full reactor build `mvn verify` green.
Proposed partial credit
60% hidden cases; 40% reactor verify green
R02.01.01.01-T07 Fix a heap overflow in a C CSV parser found by AddressSanitizer
Original instruction
Repo `/work/libcsvx` (C, CMake). A user attached `crash.csv` that triggers an ASan heap-buffer-overflow in `parse_field()` when a quoted field ends at EOF without a closing quote. Fix the bug, keep the documented lenient-quote behavior, and make the ASan build pass all tests.
- Platforms
lin· Linux- Applications
- CMake · gcc · ASan
- Interfaces
term· Terminal- GUI need · task setting
CL· CLI-sufficient- GUI need · ordinary access
CL· CLI-sufficient- Interaction primitives
code· Code & commands- Work product
WP06· Code/engineering- Conditions
- None declared
- Proposed difficulty
- 4 / 5
Proposed verifier
ASan+UBSan build: hidden tests (crash.csv plus 10 edge cases) pass with no sanitizer reports; lenient-quote output for 5 documented examples matches exactly; 10-minute libFuzzer run clean.
Proposed partial credit
35% crash fixed; 35% semantics preserved; 30% fuzz clean
R02.01.01.01-T08 Resolve trailing-slash route mismatches in a PHP micro-framework
Original instruction
Repo `/work/slimroute` (PHP 8.3, Composer). Issue: routes defined as `/users/{id}` do not match `/users/42/`, but the docs promise optional trailing slashes unless `strict` is set. Fix the matcher, respecting the `strict` flag and not breaking routes whose last segment is optional.
- Platforms
lin· Linux- Applications
- Composer · PHPUnit
- Interfaces
term· Terminal- GUI need · task setting
CL· CLI-sufficient- GUI need · ordinary access
CL· CLI-sufficient- Interaction primitives
code· Code & commands- Work product
WP06· Code/engineering- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
Hidden PHPUnit cases (12) for strict/non-strict, optional segments and root route pass; existing suite green; PHPStan level 6 no new errors.
Proposed partial credit
60% hidden cases; 30% suite; 10% static analysis
R02.01.01.01-T09 Fix a time-zone parsing bug in a Ruby date helper gem
Original instruction
Gem `/work/timeparse` (Ruby 3.3). Issue: `TimeParse.parse('2026-03-29 02:30', zone: 'Europe/Berlin')` raises instead of resolving the nonexistent local time per the `:nonexistent` option. Implement the documented options `:forward`, `:backward`, `:raise` and fix the default.
- Platforms
lin· Linuxmac· macOS- Applications
- Bundler · RSpec
- Interfaces
term· Terminal- GUI need · task setting
CL· CLI-sufficient- GUI need · ordinary access
CL· CLI-sufficient- Interaction primitives
code· Code & commands- Work product
WP06· Code/engineering- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
Hidden RSpec examples for the three options across 4 zones and both DST transitions pass; existing specs green; rubocop no new offenses.
Proposed partial credit
60% hidden examples; 30% existing specs; 10% lint
R02.01.01.01-T10 Fix culture-dependent decimal output in a .NET serializer on Windows
Original instruction
Solution `C:\work\LedgerJson` (.NET 8). Issue: on Windows with the `de-DE` culture, decimals serialize as `12,5` instead of `12.5`. Reproduce on the VM by setting the thread culture in a test, fix the serializer to always use invariant formatting, and keep custom format attributes working.
- Platforms
win· Windows- Applications
- dotnet · xUnit
- Interfaces
term· Terminal- GUI need · task setting
CL· CLI-sufficient- GUI need · ordinary access
CL· CLI-sufficient- Interaction primitives
code· Code & commands- Work product
WP06· Code/engineering- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
`dotnet test` with hidden tests under de-DE, fr-FR, ar-SA and invariant cultures: round-trip and exact string output pass; [JsonFormat] attribute tests pass; build has no new warnings.
Proposed partial credit
60% culture cases; 25% attribute cases; 15% warnings
R02.01.01.01-T11 Fix a pagination bug that spans a Python API and a TypeScript client
Original instruction
Monorepo `/work/catalog` has `api/` (FastAPI) and `web/` (TypeScript client). Issue: the last page of `/items` repeats the first item of the previous page when `sort=price`. The root cause is a non-unique sort key on the server plus a client cursor bug. Fix both sides so iterating all pages yields each item exactly once.
- Platforms
lin· Linux- Applications
- pytest · vitest · uvicorn
- Interfaces
term· Terminal- GUI need · task setting
CL· CLI-sufficient- GUI need · ordinary access
CL· CLI-sufficient- Interaction primitives
code· Code & commands- Work product
WP06· Code/engineering- Conditions
- None declared
- Proposed difficulty
- 4 / 5
Proposed verifier
Hidden end-to-end test seeds 1,037 items with many price ties, pages through the client with page sizes 7, 50, 100: every id exactly once and in order; api and web unit suites green.
Proposed partial credit
40% server stable ordering; 30% client cursor; 30% e2e exactly-once
R02.01.01.01-T12 Fix a Kotlin coroutine cancellation bug in an Android library
Original instruction
Gradle project `/work/netcache` (Kotlin, Android library + JVM tests). Issue: cancelling a `fetch()` coroutine leaves the disk-cache entry locked, so the next fetch for the same key times out. Fix it so cancellation releases the lock and leaves no partial cache file.
- Platforms
lin· Linuxmac· macOS- Applications
- Gradle · Kotlin · JUnit
- Interfaces
term· Terminal- GUI need · task setting
CL· CLI-sufficient- GUI need · ordinary access
CL· CLI-sufficient- Interaction primitives
code· Code & commands- Work product
WP06· Code/engineering- Conditions
- None declared
- Proposed difficulty
- 4 / 5
Proposed verifier
`./gradlew test` with hidden kotlinx-coroutines-test cases: cancel at 5 points then refetch succeeds within 100 ms virtual time; no *.tmp files left; existing tests green; detekt clean.
Proposed partial credit
50% lock released; 30% no partial files; 20% suites/lint
R02.01.01.01-T13 Fix a locale-dependent number formatting bug in a Swift package
Original instruction
Swift package `/work/UnitsKit` on macOS. Issue: `Measurement.formattedShort()` drops the unit for `km/h` when the locale is `pt_PT`. Reproduce with `swift test`, fix the formatting path, and add tests for at least 3 locales.
- Platforms
mac· macOS- Applications
- swift · Xcode toolchain
- Interfaces
term· Terminal- GUI need · task setting
CL· CLI-sufficient- GUI need · ordinary access
CL· CLI-sufficient- Interaction primitives
code· Code & commands- Work product
WP06· Code/engineering- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
Hidden XCTest cases across pt_PT, de_DE, ja_JP, en_US for 6 unit types pass; existing tests green; agent-added tests present and passing.
Proposed partial credit
60% hidden cases; 25% existing suite; 15% own tests
R02.01.01.01-T14 Reproduce a production crash from a stack trace and env-specific config
Original instruction
Service `/work/invoicer` (Python). The issue contains only a production stack trace (`trace.txt`) and says it happens 'only in EU deployments'. Find the environment setting that triggers it (see `deploy/`), write a failing test that reproduces it, then fix the code.
- Platforms
lin· Linux- Applications
- pytest · git
- Interfaces
term· Terminal- GUI need · task setting
CL· CLI-sufficient- GUI need · ordinary access
CL· CLI-sufficient- Interaction primitives
code· Code & commands- Work product
WP06· Code/engineering- Conditions
hidden· Hidden/stale state- Proposed difficulty
- 4 / 5
Proposed verifier
Agent's reproduction test fails on base and passes on patch; hidden tests for the EU config (VAT rounding mode + locale) pass; full suite green.
Proposed partial credit
30% trigger identified (test fails on base); 50% hidden tests pass; 20% suite green
R02.01.01.01-T15 Find and fix the root cause of a flaky test without touching the test
Original instruction
Repo `/work/scheduler` (Python). `test_next_run_order` fails about 1 in 20 runs. The issue asks for a fix in the library, not the test: the flake comes from nondeterminism in the code under test. Find it and make the behavior deterministic.
- Platforms
lin· Linux- Applications
- pytest · git
- Interfaces
term· Terminal- GUI need · task setting
CL· CLI-sufficient- GUI need · ordinary access
CL· CLI-sufficient- Interaction primitives
code· Code & commands- Work product
WP06· Code/engineering- Conditions
- None declared
- Proposed difficulty
- 4 / 5
Proposed verifier
Test files unchanged (hash); the flaky test passes 500/500 under pytest-randomly with varied PYTHONHASHSEED; hidden determinism tests pass; suite green.
Proposed partial credit
60% 500/500 passes; 25% hidden determinism tests; 15% tests untouched
R02.01.01.01-T16 Bisect a regression between two releases and fix it
Original instruction
Repo `/work/mdlite` (JavaScript). Issue: nested ordered lists render flat since v2.4.0; v2.3.0 was fine. Use `git bisect` to find the offending commit, write its hash to `BISECT.txt`, then fix the regression on `main` without reverting unrelated changes from that commit.
- Platforms
lin· Linuxmac· macOS- Applications
- git · npm · jest
- Interfaces
term· Terminal- GUI need · task setting
CL· CLI-sufficient- GUI need · ordinary access
CL· CLI-sufficient- Interaction primitives
code· Code & commands- Work product
WP06· Code/engineering- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
BISECT.txt equals the seeded first-bad commit; hidden nested-list snapshot tests pass; the other behavior introduced by that commit (table alignment) still passes its tests.
Proposed partial credit
30% correct commit; 50% regression fixed; 20% unrelated changes kept
R02.01.01.01-T17 Fix a quadratic slowdown that makes a hidden test time out
Original instruction
Repo `/work/difftext` (Python). Issue: diffing two 50k-line files takes minutes; users say v1.2 was fast. Find the accidental O(n^2) behavior and restore near-linear performance without changing diff output.
- Platforms
lin· Linux- Applications
- pytest · cProfile
- Interfaces
term· Terminal- GUI need · task setting
CL· CLI-sufficient- GUI need · ordinary access
CL· CLI-sufficient- Interaction primitives
code· Code & commands- Work product
WP06· Code/engineering- Conditions
- None declared
- Proposed difficulty
- 4 / 5
Proposed verifier
Hidden perf test: 50k-line diff under 3 s on the grader box (baseline ~180 s); output byte-identical to reference on 40 corpus pairs; suite green.
Proposed partial credit
50% performance threshold; 40% identical outputs; 10% suite
R02.01.01.01-T18 Fix a shared-utility bug that breaks three packages in a monorepo
Original instruction
pnpm/Turborepo monorepo `/work/shop` with packages `pricing`, `cart`, `checkout` and shared `utils`. Issue: totals are off by one cent for some currencies. The bug is in `utils/money.ts` rounding, but `cart` contains a workaround that double-corrects. Fix the root cause and remove the workaround so all packages agree.
- Platforms
lin· Linuxmac· macOS- Applications
- pnpm · turbo · vitest
- Interfaces
term· Terminal- GUI need · task setting
CL· CLI-sufficient- GUI need · ordinary access
CL· CLI-sufficient- Interaction primitives
code· Code & commands- Work product
WP06· Code/engineering- Conditions
- None declared
- Proposed difficulty
- 5 / 5
Proposed verifier
Hidden cross-package tests over 30 currencies (incl. JPY, BHD 3-decimals) pass; workaround removed (AST check for the compensating call); all package suites and `turbo build` green.
Proposed partial credit
40% root-cause fix; 30% workaround removed; 30% all suites green
R02.01.01.01-T19 Resolve an issue reported in Chinese with a screenshot of output
Original instruction
Repo `/work/csvkit-lite` (Python). Issue #301 is written in Chinese with a pasted terminal screenshot (`issue_301.png`): converting a GBK-encoded CSV with `--encoding auto` mangles column headers. Reproduce with the attached sample, fix encoding detection, and reply in `REPLY.md` (Chinese or English) summarising the fix.
- Platforms
lin· Linux- Applications
- pytest · git
- Interfaces
term· Terminal- GUI need · task setting
CL· CLI-sufficient- GUI need · ordinary access
CL· CLI-sufficient- Interaction primitives
code· Code & commandsvis· Visual judgment- Work product
WP06· Code/engineering- Conditions
- None declared
- Proposed difficulty
- 2 / 5
Proposed verifier
Hidden tests with GBK, Big5, Shift-JIS and UTF-8-BOM samples pass; suite green; REPLY.md non-empty and mentions the detection change (keyword check).
Proposed partial credit
60% encoding tests; 25% suite; 15% reply
R02.01.01.01-T20 Fix a cancellation race in an asyncio connection pool
Original instruction
Library `/work/aiopool` (Python 3.12). Issue: under load, cancelling tasks waiting in `Pool.acquire()` sometimes leaks a connection (pool size drifts down until all requests hang). Fix the race in acquire/release so cancellation at any await point never leaks or double-releases a connection.
- Platforms
lin· Linux- Applications
- pytest · asyncio
- Interfaces
term· Terminal- GUI need · task setting
CL· CLI-sufficient- GUI need · ordinary access
CL· CLI-sufficient- Interaction primitives
code· Code & commands- Work product
WP06· Code/engineering- Conditions
- None declared
- Proposed difficulty
- 5 / 5
Proposed verifier
Hidden stress harness: 20k acquire/cancel cycles with random cancellation points over 50 seeds; pool size invariant holds and no double-release assertions; existing suite green; mypy --strict clean.
Proposed partial credit
50% no leaks across seeds; 30% no double release; 20% suite + typing
Mapped research references
Existing benchmark-to-class links from the research catalog. They are not record-level annotations, executions or product endorsements.