vibrantlabsai/pool-desk
pool-desk: Harbor dataset on Hugging Face with 50 tasks. An OpenEnv environment for the OpenEnv Arena: 50 terminal tasks built on real public data, covering all eight Arena domains. Each episode is one task in a fresh container with its own image. The model works in a shell (exec, write file)…
Tasks
- List the CVEs in a findings export whose EPSS score rose by 0.05 or more over a quarter, from two pinned EPSS daily files.
- From a Cowrie honeypot login log, find sources whose first successful login followed 30+ failed attempts across sessions, with the winning…
- Diff a hand-edited local KEV tracker CSV against the official KEV catalog JSON, ignoring formatting noise, and report the drift as JSON.
- Audit a sample of NVD 2024 CVE records for large disagreements between NVD and CNA CVSS v3 base scores, telling the CNA entry apart from…
- Audit a Python package inventory against PyPA OSV advisories (ECOSYSTEM ranges, withdrawn entries, numeric version compare).
- Write a Python regex detection rule for failed SSH password attempts with named user and ip groups, scored on hidden real sshd lines.
- Reconstruct the ancestry of six alerted processes from merged Sysmon process-creation events by ProcessGuid (PIDs repeat), writing JSON…
- Compute noncurrent-loan and allowance-coverage ratios for 28 banks from Call Report items found by caption, and assign ordered…
- Compute the Texas ratio for 30 AZ/NV banks from Call Report items identified by caption, with RCFD/RCON selection and a 10% watch flag.
- Reconcile FDIC BankFind financials against Call Report items for 23 RI/NH banks and list every field that does not tie.
- Audit a FERC Form 1 plant in service extract (PUDL, 10 disguised utilities, 2023-2024): list accounts whose signed roll-forward or…
- Pick face-of-balance-sheet XBRL facts from SEC FSDS num.txt and report or derive total liabilities for 70 small-filer 10-Qs.
- Pick face-of-statement XBRL facts from SEC FSDS num.txt for 30 small-filer 10-Qs, compute current ratio and nine-month cash runway, and…
- Flag unavailable-speed, duplicate and position-jump reports in a six-hour AIS feed for 25 vessels and count them per vessel.
- Per-station arrival delay minutes by cause, with midnight-wrapped delays and unattributed delays, from two weeks of BTS on-time data.
- Reconcile EPA CAMPD unit heat input with EIA-923 plant fuel consumption for one state's 2024 fleet: total vs electric fuel, nuclear…
- Plant-group heat rates from EIA-923 fuel and generation, with combined-cycle parts pooled, flagged against plausibility bounds.
- Group FRA Form 54 rail accident reports into accidents through shared incident numbers and joint-report links, total the deduplicated…
- Flag structures removed, added, recoded, reconstructed or suspiciously improved between two National Bridge Inventory snapshots.
- Crude and directly age-adjusted weighted smoking prevalence for three disguised states from BRFSS 2023 respondent records.
- Check a disguised Census county adjacency excerpt for one-way pairs, missing self lines and duplicates, and list the fixes.
- Labor force, unemployment rate and BLS U-6 underutilization for three disguised states from December 2024 CPS person records.
- Audit damaged integral-point lists for 50 re-modelled Cremona elliptic curves: report bogus and missing x-coordinates.
- Survey-weighted prevalence of high blood pressure in NHANES 2017-2018 adults, with Taylor-linearized standard errors over strata and PSUs.
- CKD-EPI 2021 eGFR, KDIGO stage counts and weighted chronic kidney disease prevalence from NHANES 2017-2018 examination data.
- Audit a table of elliptic curves (Cremona ecdata, re-modelled) for wrong a p values and points not on the curve.
- Build display credits and primary/featured artist splits from MusicBrainz track artist credits (JSONL) for a streaming feed CSV.
- Write FFmpeg FFMETADATA1 chapter files for LibriVox audiobooks: contiguous ms chapters, HTML-cleaned and escaped section titles.
- QA a Polish Fluent translation against the current en-US source: missing and obsolete ids, unknown variables, missing attributes.
- Audit ISBN-10/ISBN-13 values of Open Library editions (check digits, 10- 13 consistency) and emit a feed ISBN-13 and status per edition.
- Parse free-text museum dimension lines (inches with fractions, cm or mm), convert to cm and flag mistyped metric conversions.
- Sum Discogs vinyl tracklists per side (A/B/AA ...) and flag sides over the cutting limit for the record speed.
- Rebuild a composite radius/true-mass table for 49 planets from per-reference NASA Exoplanet Archive parameter sets under stated selection…
- Annual precipitation accounting for five GHCN-Daily stations: fixed-width parsing, QFLAG and sentinel rules, MDPR/DAPR multi-day totals…
- Compute background CO2 annual means for three flask sites from NOAA GML event data: QC flag positions, pair averaging, mean of monthly…
- Compute per-storm accumulated cyclone energy for the 2024 global season from an IBTrACS-format extract (units line, synoptic-only fixes…
- Harmonise four NDBC stdmet buoy files across format eras (header drift, sentinels, sampling change, year-boundary record) into per-year…
- Resolve hand-typed legacy PDB ids through the wwPDB obsolete-entry list (multi-hop, multi-successor, dead-end and self-loop chains) to…
- Determine Type A programs (2 CFR 200.518 threshold, clusters as one program) for 30 single audits from Federal Audit Clearinghouse SEFA…
- Compute 2025 lobbying income per firm and client from LDA filings where the latest amendment or termination of each quarter supersedes…
- Build a per-title pay benchmark (active headcount, median annualized base by pay basis, OT share) from NYC payroll records of four agencies.
- Build a May-to-June 2026 headcount bridge per sub-agency from OPM employment snapshots, accessions and separations, with grouped flow…
- Add a live SUMIFS summary sheet (object class by legislative year, with shares) to a House disbursements workbook.
- Harmonise three UK departments' spending-over-£25k files (encodings, £ strings, schemas) and total spend for 14 suppliers matched by a…
- Audit how 68 real GitHub Actions workflows from seven projects pin their actions (SHA, tag, branch, local, reusable) and flag unpinned…
- Resolve GitHub CODEOWNERS for 160 paths of a real repository: gitignore pattern semantics, last match wins, owner-less rules.
- Reconcile GHSA- and PYSEC-style OSV advisory records with per-branch and multi-span ranges into the exact set of affected released…
- Summarize which RHEL 9 minor releases received fixes for 40 CVEs per support offering (EUS spellings, E4S, mainline) from Red Hat VEX…
- Assign ordered supply-chain risk flags to 57 PyPI packages from registry metadata and OpenSSF Scorecard checks, where missing checks are…
- Build a per-session ledger (client, reattaches, expiry vs client close, lifetime) from a real three-server ZooKeeper log where socket…