Overview - Sverka

sverkaWorkspace

CI Pipeline Executions
- Filtered
Runs
Demo

Insights

Compare tasks
Analytics

Overview

Distributed computation caching and task execution are currently disabled for this organization.

sverka

Loading workspace stats

Loading workspace insights...

Statistics interval

7 days30 days

Latest CI Pipeline Executions

Filter

Fuzzy

Sort by

© 2026 - Nx Cloud

Terms of Service Privacy Policy Changelog Status Docs Contact Nx Cloud Pricing Company @NxDevTools

sverkaWorkspace

Average execution time

1m

105 CI pipeline executions

Computation time saved

1h 32m

Cache hit rate

49%

Protect cache integrity with sandboxing

Time to Green

1m

Was 1m 2s, improved by 16%

Succeeded
feat/arena-real-task
20a3d7fb arena: tighten credential redaction to exact key names; stop echoing the agent-authored verify script into committed check output
by Petr Plenkov
Succeeded
feat/arena-real-task
64f6f62b arena/reporter: close remaining review gaps - verify check asserts the script composes all five checks or delegates to a pipeline tool — `verify: echo ok` no longer passes - integrity check rejects any vite/vitest config filename (readdir glob, covers .mjs/.cjs and vite.config.test variants) - redactValue strips credential-bearing object keys (apiKey &co) in tool-call arguments - log search folds full text once (contextual ς/σ) with per-char fold-length offset map
by Petr Plenkov
Succeeded
feat/arena-real-task
96f8060a chore: prettier
by Petr Plenkov
Succeeded
feat/arena-real-task
265af3b8 arena: require letter TLD in email redaction `pkg@0.1.29` specifiers were being mangled into [email] — the TLD must be at least two letters.
by Petr Plenkov
Succeeded
feat/arena-real-task
ce1dd4aa arena/reporter: close sanitize + integrity gaps from review - trace reports render from the sanitized trace clone, not the raw one - clipboard-disabled state now lives in updateActionState so opening a step can't re-enable it on insecure contexts - held Space/Enter still prevents default (page scroll) but skips the synthetic click - integrity check rejects new vitest configs (include:[] bypass) and the prompt limits additions to pipeline files, no new tool configs
by Petr Plenkov
Succeeded
feat/arena-real-task
ec1e0c65 arena: complete trace sanitization; dedupe tasks unconditionally - redactText also strips emails and bearer/token/api-key patterns; sanitization now covers run output, checkResults output, and verdicts (check output was still publishing /tmp sandbox paths) - filterTasks dedupes configured ids even without --task and the unknown-id diagnostic lists unique ids - cases.json updatedAt only advances when a case actually changed
by Petr Plenkov
Succeeded
feat/arena-real-task
700b7f71 arena: harness-side integrity digests; reporter: unicode-fold search - protected-files check embeds sha256 digests + pins the five existing package.json script values via node -e — a workspace manifest the agent can rewrite no longer gates itself; also drops the GNU-only sha256sum dependency - fixture test script drops --passWithNoTests so the tests-pass check can't succeed vacuously - trace JSON now emits the configs:<comboLabel> shape trace.html actually reads (llmCallCount was copied to a level nothing consumed) - log search folds per-char with an offset map: keeps unicode-fold matches (i → İ) and correct highlight ranges — fixes both directions of the lowercase-index bug
by Petr Plenkov
Succeeded
feat/arena-real-task
9fa7c4c1 arena: enforce pipeline re-run contract; sanitize committed traces - ship-package prompt now requires `bun run verify` as the single re-runnable entry point (was only a successCriteria the judge never reads); adds a `pipeline` check executing it, so both cells must wire a real pipeline — baseline can't pass on source fixes alone - prompt explicitly permits new files (sverka.config.ts) and package.json scripts; fixture .gitignore/.prettierignore cover .sverka/.devin so run artifacts can't fail the format check - update-arena-results: redact sessionId, /home/*, /tmp/*, kernel version; dedupe replayed steps by stepId — applied to per-run trace files and the committed arena-results.json - bin: extract filterTasks() (dedup + unknown-id rejection) with unit tests for the happy path - reporter: attach download anchor before click + defer revoke (Firefox); clear DAG drag-guard before synthetic keyboard click; guard e.repeat; disable copy button when clipboard API is unavailable
by Petr Plenkov
Succeeded
feat/arena-real-task
cfe77a93 test(reporter): assert regex search emit; arena: refresh run on fixed fixture The fixture's intentional bugs were already fixed by a previous agent run, so re-ran the benchmark to keep committed results honest — plugin run now shows -4.9k tokens / -13 tool calls / -31s vs baseline.
by Petr Plenkov
Succeeded
feat/arena-real-task
aef97f1c review: --task scoping, trace llmCallCount, unicode-safe log search - --task rejected on report/doctor instead of silently ignored; duplicate config task ids dedupe to one run each - trace JSON now carries top-level llmCallCount (from RunResult.metrics) — the SPA reads it there - drawer search matches via case-insensitive regex on the original text — indexOf on a lowercased copy misaligns marks when lowercasing expands a char (İ → i̇) The serialize.ts finding is a real bug in packages/storage (the fixture is verbatim) — deferred as sv-0g6t against the source package. Generated with [Devin](https://devin.ai) Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>
by Petr Plenkov
Succeeded
feat/arena-real-task
718d5aeb arena: real-package task + --task filter + live dogfood results - fixtures/ship-package: verbatim copy of the real @sverka/storage package (49 real vitest tests, tsdown build, @sverka/runtime from npm) with three accumulated-without-CI failures — a dropped process.cwd() fallback (typecheck), unused imports (lint) and prettier drift (format) - arena.config.ts: 'ship-package' task — prompt never mentions sverka; the skill cell must discover the plugin, author sverka.config.ts and drive sverka run; baseline does the same work by hand - bin: repeatable --task <id> filter, unknown ids exit 2 with the known-task list - website/public/benchmark: refreshed snapshot — real run on swe-2-high, both cells pass all 5 checks Honest result worth keeping: the skill cell costs more on this small task (~37k vs ~30k tokens, 187s vs 110s) because it installs the CLI, authors the config and also compiles a GitHub workflow — while producing a richer artifact (sverka.config.ts + ci.yml + run report). Generated with [Devin](https://devin.ai) Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>
by Petr Plenkov