Skip to content

Changelog

Changelog

All notable changes to atb-cli are documented here. Format follows Keep a Changelog.

[Unreleased]

Added

  • atb agc reads AGC (Assembled Genomes Compressor) archives. atb agc install fetches the upstream agc binary (Linux/macOS x64+arm64, Windows x64); atb agc ls lists samples or a sample's contigs; atb agc info prints archive metadata; and atb agc get extracts FASTA by contig (contig[@sample][:from-to]), by --sample, or the whole collection with --all. Output streams to stdout or a file (-o), with optional gzip (--gzip) and line wrapping (-l).
  • atb agc download fetches assembled genomes from AllTheBacteria's OSF storage. --species <name> resolves and extracts every .agc batch for a species; given sample accessions instead (positional, a --from file, or piped stdin) it resolves each accession to the batch that holds it and downloads only those batches, so you extract just the samples of interest. Batches resolve from a cache-first, md5-verified index that by default crawls the full v202505 collection across its OSF nodes; --osf-node <id> targets a single node's published index and --agc-index <file> uses a local TSV. atb agc index builds or refreshes the by-species index. This AGC path is a preview under active development, so archive locations and the index may change.
  • atb agc locate [accession...] is the search half of atb agc download: it resolves sample accessions to the AGC batch (and OSF collection part) that holds them without downloading anything, reading the same map and index. Accessions come from arguments, a --from file, or piped stdin; output is TSV or --format json.

v0.17.1 - 2026-06-04

Fixed

  • atb query --sample-file now uses the fast SQLite index path instead of scanning the whole table. The index lookup read only the --samples slice and ignored accessions loaded from --sample-file, so a file-based query fell back to a full scan: the reporter's --sample-file run had not finished after a minute, versus ~0.1s for the same 10 accessions passed via --samples. Both sources are now merged and de-duplicated through Filters.SampleAccessions() before building the sample_accession IN (...) clause, so --sample-file and --samples perform identically (#19).
  • atb directory help no longer conflates the metadata index with genome downloads. The --data-dir help read "data directory for downloaded files", which led users to expect --data-dir/$ATB_DATA_DIR to redirect where atb download saves assemblies; those are written to the separate download.output_dir (-o/--output-dir). --data-dir now reads "directory for the local metadata index", and atb download --output-dir shows its effective default path and the download.output_dir config key (#20).

v0.17.0 - 2026-05-29

Added

  • atb osf download can now extract downloaded tar archives with new --extract, --compress (none/gz/xz, default gz), and --delete-archive flags — including AllTheBacteria .tar.xz assembly tarballs (#18).

v0.16.1 - 2026-05-24

Fixed

  • --data-dir help text now reflects the effective default and surfaces the ATB_DATA_DIR override. Previously hardcoded to default $HOME/.atb/data, the usage string ignored the OS-specific data directory (e.g. ~/Library/Application Support/atb/data on macOS) and gave no indication that ATB_DATA_DIR was honored, leaving shared-install users unable to confirm via --help that the env var was wired up (#15). The string is now built at runtime from config.DefaultDataDir() and explicitly mentions $ATB_DATA_DIR.

v0.16.0 - 2026-05-20

Added

  • Environment-variable overrides for shared installations. ATB_DATA_DIR (alias: ATB_DATADIR) overrides the data directory and ATB_CONFIG overrides the config file path. Both precede the loaded config but yield to an explicit --data-dir flag, so admins on multi-user servers can point every user at a shared data volume via /etc/profile.d without writing per-user config files (#12). All CLI subcommands route through a single resolveDataDir helper so the precedence is consistent.
  • atb info now prints run_accession as the first field of the ENA Metadata section, surfacing the SRA run accession directly instead of asking users to scrape it out of the fastq_ftp URL (#13).

Fixed

  • Release binaries are built with CGO_ENABLED=0, producing statically linked executables that run on RHEL 7/8/9/10, Debian 11, and other older-glibc systems. The previous dynamic builds failed at startup with GLIBC_2.32/GLIBC_2.34 not found (#10). All dependencies are pure Go (including modernc.org/sqlite), so nothing is lost from dropping CGO; the Makefile sets the same flag for local builds.
  • atb config show and config.Save no longer emit the BurntSushi/toml default two-space indent. Long URL values like base_url were wrapping in narrow terminals and looked like they had unmatched quotes (e.g. a visible line of AllTheBacteria"); standard unindented TOML keeps values column-aligned (#11).
  • atb info no longer pauses ~24s after the MLST section. The ENA join was using ReadFiltered, which materialised the entire ~2M-row ENA parquet into memory before applying the sample predicate. All five single-sample lookups in atb info (ENA, Assembly, Stats, CheckM2, MLST) now use ReadStreamFiltered with limit=1, filtering during deserialization and exiting on the first match (#14).

Docs

  • README's curl ... | bash install example now places ATB_INSTALL_DIR=... after the pipe so it scopes to the install script rather than to curl. Adds a short note explaining why the placement matters (#9).
  • New Shared installations (multi-user servers) section documenting ATB_DATA_DIR and ATB_CONFIG.

v0.15.2 - 2026-05-16

Fixed

  • Summary statistics now exclude unprocessed/rejected samples (asm_fasta_on_osf == 0) so totals match the published OSF release (#7). The fix applies to both code paths: atb summarise (default, no --from) and the MCP-facing atb_stats / atb_species_list tools backed by internal/index.QueryStats and SpeciesList. On v0.15.1 production data this drops Total genomes by ~460k and removes the spurious NA entry from the top-species list. atb summarise --from <file> is unchanged — user-supplied rows are summarised as-is.
  • QueryStats.TopSpecies now retains the unknown category (assembled samples sylph could not classify) since it's a real published bucket; SpeciesList still drops it because its purpose is to enumerate named species.

v0.15.1 - 2026-05-05

Fixed

  • atb summarise with no arguments now prints the default summary instead of the help screen. The earlier guard treated zero flags + zero positional args as "show help", contradicting the documented Default summary of the full database example (#6). Stray positional args (e.g. atb summarise foo) now error via cobra.NoArgs rather than being silently ignored.

v0.15.0 - 2026-05-01

Added

  • atb amr now accepts --samples (comma-separated accessions) and --sample-file (one accession per line, # comments allowed) for parity with atb query. They intersect with --hq-only and ENA filters so combined restrictions tighten rather than override the sample set.
  • Output writers across atb amr, atb query, and atb mlst auto-gzip when -o ends in .gz. Format is inferred from the destination filename — results.csv.gz produces gzipped CSV regardless of the configured default — and explicit --format still wins.

Fixed

  • Per-genus AMR partition and SQLite-index lookup is now case-insensitive. GTDB letter clades (e.g. Legionella_C) were previously normalised to Legionella_c, missing the on-disk Legionella_C.parquet partition; atb amr --genus Legionella_C silently fell back to a full monolithic scan. Results were correct but the optimisation was bypassed for any genus name where casing changed after the first character.
  • atb amr --type help text no longer claims amr is the default. The default is unset (matches all element types); the help string now reads default: all types.

v0.14.2 - 2026-04-30

Added

  • atb fetch now prompts before downloading amrfinderplus.parquet, surfacing that the ~1.2 GB download expands to ~35 GB of per-genus partitions and SQLite indexes during the post-fetch index build. Answering n drops only AMR from the target list and continues fetching the other core tables, so users on small home volumes can opt out without aborting the whole download. Non-interactive runs (no TTY) and --yes skip the prompt.

Docs

  • README install section now spells out that ATB_INSTALL_DIR controls the binary location only, not the data directory, and documents --data-dir / atb config set general.data_dir for steering the database to a larger volume. The default-disk figure ("~3.5 GB typical install") was stale and is corrected to ~35 GB for the default install (core parquet + AMR indexes).

v0.14.1 - 2026-04-30

Performance

  • AMR index build now streams rows from parquet into SQLite instead of loading the full per-genus partition into memory first. Combined with journal_mode=OFF + in-memory temp store during the build (the .tmp→final rename keeps crash safety), peak heap during partition→SQLite conversion drops from O(partition size) to O(buffer ≈ 512 rows). Large genera that previously needed multiple GB now run in tens of MB.
  • Per-genus partition writers cap row groups at 100k rows (parquet-go MaxRowsPerRowGroup). The previous unlimited default kept all compressed pages of every open writer in memory until close, which was the dominant cost during atb fetch --tables amrfinderplus.parquet. Each writer now flushes a row group to disk every 100k rows.

Fixed

  • Stale-data errors from atb amr now suggest atb fetch --tables amrfinderplus.parquet --force --yes instead of a bare atb fetch --force. The previous suggestion re-downloaded every core table (~hundreds of MB to GBs of unrelated assembly/checkm2/mlst data) when only the AMR artifact was stale.

Added

  • BenchmarkBuildOneIndex and BenchmarkBuildOneIndexPeakRSS in internal/amr track allocs/op and peak heap during a 1M-row build, so future memory regressions in the streamed indexer surface in CI bench output.

v0.14.0 - 2026-04-29

Changed

  • AMRFinderPlus dataset upgraded from v3.12.8 to v4.2.5 (~25.6M → ~58.5M rows). atb amr now reads the new OSF artifact (amrfinderplus.parquet, AMRFP v4.2.5) by default.
  • AMR output now exposes all 26 AMRFinderPlus columns in TSV, CSV, JSON, and table format (previously a curated subset). Column headers in TSV/CSV/table and JSON keys for atb amr --format json and the atb_amr MCP tool match the AMRFinderPlus v4.2.5 TSV verbatim — including spaces and the leading % in % Coverage of reference / % Identity to reference.

Added

  • Schema guard on every atb amr invocation: detects an older v3.12.8 parquet on disk and returns an actionable error pointing to atb fetch --force instead of silently zeroing out renamed columns.
  • SQLite index error wrapping: stale per-genus indexes (built from a v3 dataset) now surface a clear "stale index, run atb fetch --force to rebuild" error rather than the raw no such column SQLite message.

Breaking changes

  • The AMR column set, header names, and JSON key names have changed. Pipelines that grep specific columns by name (e.g. Gene symbol) must update to the v4.2.5 names (e.g. Element symbol). Consumers that select columns by index also need to update — the column count went from a smaller curated set to the full 26.
  • On-disk v3.12.8 parquet files and indexes are not forward-compatible with this release.

Migration notes

  • Run atb fetch --force to download the v4.2.5 AMR parquet and rebuild genus partitions and SQLite indexes. No metadata re-download is required; only the AMR artifact changed.
  • If you skip the refetch, atb amr will fail fast with a friendly error rather than returning silently corrupted rows.

v0.13.2 - 2026-04-29

Fixed

  • atb amr invoked with no flags now reaches the unfiltered-scan confirmation prompt instead of short-circuiting to help text. Regression introduced in v0.13.1.

v0.13.1 - 2026-04-29

Added

  • --genus flag on atb amr for case-insensitive bacterial genus filtering, independent of --species.
  • Confirmation prompt on atb amr when invoked without any filter, to guard against accidental full-table scans of the AMR dataset.

Fixed

  • atb amr --species "Genus species" now applies a true species-level filter on the row's species column; previously the species token was being conflated with the genus partition lookup.
  • install.sh now points at the allthebacteria/atb-cli repository URL (post-migration permanent target). Closes the temporary v0.12.4 redirect that targeted the old repo while releases were being cut on the new one.

v0.13.0 - 2026-04-22

Changed

  • Repository moved to allthebacteria/atb-cli. This is the first release under the new home.
  • Go module path renamed from github.com/AMR-genomics-hackathon-2026/atb-cli-claude to github.com/allthebacteria/atb-cli. Users running go install github.com/AMR-genomics-hackathon-2026/atb-cli-claude/cmd/atb@latest must switch to go install github.com/allthebacteria/atb-cli/cmd/atb@latest.
  • Documentation URLs (README.md, docs/deployment.md) updated to reference the new repository.
  • Container image path updated to ghcr.io/allthebacteria/atb-cli.

Migration notes

  • Users on v0.12.3 or v0.12.4 (the migration bridge) auto-discover this release via atb update; no manual reinstall required.
  • Users on v0.12.2 or earlier should first run atb update to reach the bridge (v0.12.4), then run atb update again to reach this release. Alternatively: reinstall via curl -fsSL https://raw.githubusercontent.com/allthebacteria/atb-cli/main/install.sh | bash.
  • Historical release assets for v0.1.0–v0.12.4 remain downloadable from the archived old repo at github.com/AMR-genomics-hackathon-2026/atb-cli-claude.

v0.12.4 - 2026-04-22

Fixed

  • install.sh now fetches the latest release from this repository during the migration gap. In v0.12.3 the installer was pointed at the new repository before it had any releases, causing curl | bash fresh installs to fail with a 404. Installed binaries continue to self-update to the new repository on atb update; only the curl | bash path was affected.

v0.12.3 - 2026-04-22

Changed

  • Project has moved to allthebacteria/atb-cli. This release redirects the built-in self-update mechanism (atb update) and the curl | bash installer to the new repository. Running atb update from this version onward will discover future releases published on the new repo — no action required, users will upgrade automatically.
  • New installations should use the new URL: 
    curl -fsSL https://raw.githubusercontent.com/allthebacteria/atb-cli/main/install.sh | bash
  • Historical releases (v0.1.0–v0.12.2) remain downloadable from this archived repository.

v0.12.2 - 2026-04-20

Fixed

  • atb mlst now excludes samples that have no MLST scheme assigned (mlst_scheme empty or -) by default. Previously, default queries returned rows with blank ST and alleles columns. Use --status NONE to inspect untyped samples explicitly.

v0.12.1 - 2026-04-17

Fixed

  • atb query, atb mlst, and atb amr date filters now accept ISO 8601 interval collection_date values (e.g. 2020-01-01/2020-06-30). Previously ~50k ENA rows using interval notation were silently excluded.

Changed

  • Help text for --collection-date-from / --collection-date-to clarifies that rows with missing or unparseable dates are excluded from date-filtered results.

v0.12.0 - 2026-04-17

Added

  • --with-ena flag on atb mlst and atb amr appends country, collection_date, and instrument_platform columns from the ENA table
  • Any ENA filter (--country, --platform, --collection-date-from, --collection-date-to) now implies --with-ena, so the columns driving the filter are visible in the output

Changed

  • atb mlst and atb amr no longer scan ena_20250506.parquet unless --with-ena or an ENA filter is set — default queries stay in the millisecond tier even when the ENA table is installed

v0.9.0 - 2026-03-28

Added

  • atb sketch query -- find closest ATB genomes via MinHash sketch distances (ANI)
  • atb sketch install -- auto-download sketchlib binary from GitHub releases (Linux/macOS)
  • atb sketch fetch -- download ATB sketch database from OSF (~4.2 GB, ~3.2M genomes)
  • atb sketch query --download -- fetch matched genome assemblies alongside results
  • atb sketch info -- show sketch database stats (sample count, k-mer sizes, size)
  • --dry-run flag on sketch query download for previewing
  • --knn and --threads flags for controlling match count and parallelism

Changed

  • Consolidate all external URLs into internal/sources/sources.go (single source of truth)
  • Format large numbers with thousand separators across all commands (e.g. 1,868,526)
  • Default sketch threads to NumCPU - 1 for faster queries

v0.8.2 - 2026-03-28

Changed

  • atb osf ls defaults to table format in terminal, TSV when piped
  • Show human-readable sizes (KB/MB/GB) instead of raw MB in OSF listings

v0.8.1 - 2026-03-28

Fixed

  • Background update check now saves state before process exit (was racing against main())
  • Update notice shows on every run when newer version exists (not just once per 24h)

v0.8.0 - 2026-03-28

Added

  • atb osf ls -- browse ~3,000 ATB files across 75+ OSF project categories
  • atb osf download -- download files by regex pattern with MD5 verification
  • --grep, --project, --sort, --refresh flags for flexible file discovery
  • OSF file index cached locally with 7-day TTL

Changed

  • Download module: transport-level timeouts (no more killed large transfers)
  • DownloadAllFiles() accepts FileTask structs with filenames and MD5 checksums
  • Progress reporting via callback during downloads
  • Handle HTTP 416 by clearing stale .part files

v0.7.3 - 2026-03-27

Added

  • Comprehensive benchmarks in docs/BENCHMARKS.md with 3-tier AMR query results

Fixed

  • SQL LIKE pattern escaping for underscores in gene names

v0.7.2 - 2026-03-27

Changed

  • Default output format is now TSV (override with --format table)
  • Update check shows release notes and changelog

v0.7.1 - 2026-03-27

Fixed

  • SQLite variable limit error when using --hq-only with AMR queries

Changed

  • Show help when subcommands are called without flags

v0.7.0 - 2026-03-27

Added

  • Per-genus SQLite indexes for instant AMR queries (<10ms)
  • Interactive prompt before building indexes after fetch
  • --yes/-y flag for non-interactive index builds
  • Database summary with file sizes after fetch

Performance

  • Index build uses up to 8 workers concurrently

v0.6.0 - 2026-03-27

Added

  • Comma-separated species in --species flag for multi-species queries
  • --gene/--class queries across all genera without --species

Performance

  • Stream-filter parquet rows during deserialization (276-556x faster with --limit)
  • Genus-partitioned parquet files eliminate 65-96% of I/O per query

v0.5.2 - 2026-03-27

Fixed

  • Auto-detect CSV/TSV delimiter from file content instead of file extension

v0.5.1 - 2026-03-27

Performance

  • Parallelize parquet reads during index build

Fixed

  • AMR test fixtures missing from repo
  • Self-update "text file busy" on running binary

v0.5.0 - 2026-03-27

Added

  • atb mlst -- query MLST data with scheme/ST/status filters (2.44M records, 156 schemes)
  • MLST section in atb info output
  • atb_mlst MCP tool for LLM integration

Changed

  • AMR data unified into single amrfinderplus.parquet (81 MB, 25.6M rows)
  • atb fetch now downloads 7 core tables including AMR and MLST
  • All data sourced from OSF

v0.4.1 - 2026-03-27

Fixed

  • Self-update: rm before cp to avoid "Text file busy" error in /usr/local/bin

v0.4.0 - 2026-03-27

Added

  • atb mlst -- query Multi-Locus Sequence Typing data
  • Filter by species, sequence type (--st), scheme, status
  • MLST data in sample info (atb info)
  • atb_mlst MCP tool
  • mlst.parquet added to core fetch tables

v0.3.1 - 2026-03-26

Added

  • HTTP/SSE transport for MCP server (atb mcp --http :8080)
  • CORS middleware for cross-origin access
  • Codex CLI setup instructions
  • MIT license

v0.3.0 - 2026-03-26

Added

  • SQLite query index (700-4000x faster queries, <10ms vs 7-40s)
  • atb mcp -- Model Context Protocol server for LLM integration
  • 5 MCP tools: atb_query, atb_amr, atb_info, atb_stats, atb_species_list
  • atb index command to manually rebuild index
  • Transparent fallback to parquet scan when index is absent

Performance

  • Single sample info: <10ms, 14 MB RAM (was 39.5s, 2.2 GB)
  • Species query: <10ms, 15 MB RAM (was 7.3s, 2.1 GB)

v0.2.2 - 2026-03-26

Changed

  • Practical examples in every command's --help output

v0.2.1 - 2026-03-26

Fixed

  • Self-update uses sudo when binary is in a privileged directory

v0.2.0 - 2026-03-26

Added

  • atb amr -- query AMRFinderPlus gene hits by species, drug class, gene symbol
  • AMR data fetch from GitHub by genus (hive-partitioned parquet)
  • Support for AMR, stress, and virulence element types
  • Filters: --class, --gene (wildcards), --min-coverage, --min-identity

Fixed

  • Install script: version parsing on macOS/BSD
  • Install script: unbound variable in cleanup trap

v0.1.0 - 2026-03-26

Added

  • Query 3.2M bacterial genomes by species, genus, quality, N50, country, platform
  • Filter via CLI flags or reproducible TOML filter files
  • Multi-table joins: assembly + checkm2 + assembly_stats + ENA metadata
  • Fuzzy species name suggestion (Levenshtein distance)
  • Parallel HTTP download with resume and retry
  • Disk space checking before downloads
  • Summary statistics with group-by dimensions
  • Detailed sample info across all tables
  • Auto-detect output format (table for terminal, TSV for pipes)
  • TSV, CSV, JSON, and table output formats
  • TOML configuration with OS-standard paths
  • Self-update from GitHub releases with background version check
  • One-line install script (curl | bash)