Skip to content

Transmission-FileBot Media Processing Pipeline

Imported from smartwatermelon/transmission-filebot (archived 2026-04-03). For full git history, see the archived repository.

This file provides operational reference for the Transmission-FileBot media processing pipeline.

Note: Common protocols and standards are in ~/.claude/CLAUDE.md (global configuration). This file contains only project-specific additions and modifications.

For reviewers: The global configuration file may not be accessible during review. This is expected - the global file provides common standards across all projects.

Project Overview

Transmission-Plex Media Manager: A bash-based automation system that integrates Transmission BitTorrent client with FileBot and Plex Media Server. When downloads complete, it automatically cleans, organizes, renames media files, and triggers Plex library updates.

Stack: Pure Bash 5.x, strict POSIX compliance, zero external frameworks, comprehensive BATS test coverage

Architecture

Core Components

  1. transmission-done.sh - Main processing script (1,390 lines)
  2. Entry point triggered by Transmission on download completion
  3. Orchestrates cleanup → FileBot processing → Plex notification
  4. Contains inline test suite (run with TEST_MODE=true)
  5. Triple-mode design: production execution, inline test harness, BATS test support

  6. install.sh - Interactive installation wizard (478 lines)

  7. Dependency checking and installation
  8. Plex server validation and token authentication
  9. Auto-detects Plex library paths and presents as menu
  10. Generates ~/.config/transmission-done/config.yml
  11. Creates symlink at ~/.local/bin/transmission-done

  12. run_tests.sh - BATS test runner (122 lines)

  13. Executes comprehensive test suite
  14. Auto-installs bats-core if missing
  15. Runs unit tests (test/unit/*.bats)
  16. Runs integration tests (test/integration/*.bats)
  17. 114 total tests covering all functionality

  18. config.yml - YAML configuration

  19. Located at ~/.config/transmission-done/config.yml (production)
  20. Or ./config.yml (local development override)
  21. Contains paths, Plex credentials, logging settings

Test Infrastructure

Structure:

test/
├── test_helper.bash           - Shared test utilities and assertions
├── unit/                      - Unit tests (84 tests)
│   ├── test_error_logging.bats
│   ├── test_file_safety.bats
│   ├── test_filebot.bats
│   ├── test_mode_detection.bats
│   ├── test_plex_api.bats
│   └── test_type_detection.bats
└── integration/               - Integration tests (30 tests)
    ├── test_manual_mode.bats
    ├── test_movie_workflow.bats
    └── test_tv_workflow.bats

Test modes:

  • TEST_MODE=true - Inline tests in transmission-done.sh
  • TEST_RUNNER=true - BATS test execution mode
  • ./run_tests.sh - Runs full BATS suite (114 tests)

Processing Flow

Transmission download completes
  ↓
transmission-done.sh triggered with TR_TORRENT_* env vars
  ↓
Config validation (read_config)
  ↓
Cleanup phase (cleanup_torrent)
  - Removes .nfo, .exe, .txt files
  ↓
Media detection (process_media)
  - Pattern matching: S01E01 → TV, (19|20)XX → Movie
  - Tries most likely type first, falls back if needed
  ↓
FileBot processing (process_media_with_fallback)
  - TV (episode-style names): TheTVDB → TheMovieDB::TV → AniDB,
    then auto-detection, then movie databases
  - Everything else: auto-detection first, then database chains
  - A library conflict stops the chain (see Duplicate handling)
  - Actions: move, artwork, metadata, subtitles
  ↓
Plex notification (trigger_plex_scan)
  - Section 1 (movies) or Section 2 (TV shows)
  - Retry logic: 3 attempts, 5s delays
  ↓
Cleanup empty directories

Key Design Patterns

  • Strict mode everywhere: set -euo pipefail with IFS=$'\n\t'
  • Test mode injection: Functions check TEST_MODE flag and mock external calls
  • Config precedence: Local ./config.yml → User ~/.config/transmission-done/config.yml
  • Retry with exponential backoff: plex_make_request supports configurable retries
  • Pattern-based media type detection: Filename analysis before attempting FileBot processing
  • stdout/stderr separation: All user messages to stderr, only data to stdout

Database choice and strict matching

Episode-style files go to TheTVDB first, not to FileBot's auto-detection. Auto-detection uses TheMovieDB, and TheMovieDB can be missing whole seasons: its Great British Bake Off entry (id 34549) stops at series 7, the last BBC series. With -non-strict, FileBot does not give up on a season it cannot find. It remaps the file by episode title instead, so S17E01 "Cake Week" was filed as S04E01 "Cake", and every Channel 4 season landed in series 1–7 folders.

Auto-detection and TheMovieDB::TV therefore run strict, and fail rather than guess. TheTVDB, AniDB, and the movie databases keep -non-strict, because strict mode refuses any name shared with a spin-off ("The Great British Bake Off: An Extra Slice"). Tested against FileBot 5.3.0 on 2026-09-22:

Database -non-strict strict
auto-detect (TheMovieDB) S04E01 (wrong) fails
TheMovieDB::TV S10E01 (wrong) fails
TheTVDB S17E01 (correct) fails

Year guard on movie matches (#216)

Non-strict matching has the same flaw on the movie side: it takes the closest title even when the year is wrong. On 2026-10-15 the 1989 NBC miniseries Around.The.World.In.80.Days.1989.Part{1,2,3}.DVDRip.XviD.avi was refused by strict auto-detection, then matched by non-strict TheMovieDB to the 1988 animated film and filed as Movies/Around the World in 80 Days (1988)/... - CD{1,2,3}.avi, with that film's poster, backdrop and tmdb.url. Plex prefers those local files, so fixing the match in Plex did not stick.

Before TheMovieDB or OMDb moves anything, process_with_database dry-runs the match (--action test). filebot_year_mismatch compares each destination's (YYYY) with the years in the source filename (or its folder name, when the file has none). The preview step runs the same check. If they differ:

  • nothing is moved, and the database chain stops (filebot_chain_should_stop); the next candidates, OMDb and the non-strict TV chain, are no safer
  • a YEAR MISMATCH: line is added to the output, so classify_failure returns no-match, even when the output also holds a conflict line. A conflict with a wrong-title file must never reach media-compare.

The years must be equal. A one-year tolerance would not have caught 1989 → 1988. The cost of that is that an occasional correct match, where TMDB's release year is a year off the release name, goes to triage/no-match/ with an email instead of into the library. A filename with no year is not checked. TV databases are not guarded: an episode's destination carries the show's first-aired year, which date-named episodes legitimately contradict.

Email when a download cannot be processed

When a download is triaged as no-match or failed, notify_unprocessed emails alerts.email from config.yml through alert-lib.sh (see monitoring.md). The email names the torrent, the triage folder, and either the year-guard reason or the last 15 lines of FileBot output. already-in-plex sends nothing: FileBot identified the download, and media-compare has handled it.

  • The email goes out only after the triage move succeeds. A failed move makes the trigger watcher retry, and each retry would send another email.
  • alert-lib.sh is loaded from ~/.local/lib/ (where msmtp-setup.sh deploys it), or else from beside the script. Without it, or without alerts.email, the download is still triaged and a log line says no email was sent.
  • transmission-filebot-setup.sh writes alerts.email from MONITORING_EMAIL in config/config.conf. An existing config.yml gets the key only when setup is re-run, or by hand:
alerts:
  email: you@example.com

Duplicate handling and quality upgrades

FileBot runs with --conflict skip, which never overwrites a library file. (It used to run with --conflict auto, which is not a skip: it deletes the library file whenever it judges the download better, and together with a wrong match that destroyed correct episodes.) A skipped download is classified already-in-plex by classify_failure() and moved to triage/already-in-plex/.

Most duplicates are caught at the preview step, because FileBot prints the conflict line even under --action test. preview_filebot_changes() keeps its output on failure so that line reaches classify_failure(). When a conflict shows up later in the fallback chain, the chain stops there: another database could only re-match the same file somewhere wrong.

media-compare.sh decides which of the two copies is better, so that the download is not left sitting in triage waiting for someone to compare them by hand (issue #178). It is a sourced library, like pia-port-guard.sh, and is optional: when the file is absent, transmission-done.sh keeps its previous behaviour of triaging every duplicate.

Pairing. The download and the library file it collides with are read from FileBot's own conflict message, which names both paths:

[SKIP] Skipped [SOURCE] because [DEST] already exists
[IMPORT] Destination file already exists: DEST (SOURCE)

FileBot has already matched the download to a title and worked out the exact destination, so taking the pair from its output is authoritative. Guessing it back from filenames is not. Anything unparsable yields no pair, and the download goes to triage as before.

Comparison ladder. ffprobe (from the ffmpeg formula, already in config/formulae.txt) reads the attributes. The first rule that separates the two files decides:

  1. Identical bytes → keep (a size check first, then cmp)
  2. Video resolution, as width × height → more pixels wins
  3. Video bitrate → higher wins, past a 10% noise margin
  4. Audio channels → more wins
  5. File size → larger wins, as a last resort only

Everything uncertain resolves to keep. A wrong "replace" destroys the better copy; a wrong "keep" costs disk space and a triage entry. Those are not symmetric, so an unreadable file, a probe timeout, or a tie on every rule all leave the library alone.

Replacement is ordered so the library is never short an episode: the incumbent moves to triage/quarantine/ first, then the candidate moves into its place, and a failure of the second move restores the first. Quarantined files are pruned after MEDIA_COMPARE_RETENTION_DAYS (default 14).

A Plex rescan fires only when something was actually replaced — this code runs on every duplicate, and a scan with nothing to find is wasted work.

Configuration (environment variables, all with working defaults):

Variable Default Purpose
MEDIA_COMPARE_DRY_RUN false Log decisions, change nothing
MEDIA_COMPARE_RETENTION_DAYS 14 Quarantine lifetime
MEDIA_COMPARE_BITRATE_MARGIN 10 Percent difference before bitrate decides
MEDIA_COMPARE_TIMEOUT 30 Seconds before an ffprobe read is abandoned
MEDIA_COMPARE_FFPROBE ${HOMEBREW_PREFIX}/bin/ffprobe Probe binary

Tests live in tests/media-compare.bats and probe real media synthesised by ffmpeg rather than mocking ffprobe, so a wrong -show_entries spec or a misread field name fails the suite instead of passing it.

Critical Environment Variables

When Transmission invokes the script, it provides:

  • TR_TORRENT_DIR - Download location (required)
  • TR_TORRENT_NAME - Torrent name (required)
  • TR_TORRENT_HASH, TR_TORRENT_ID - Identifiers
  • TR_TIME_LOCALTIME - Completion timestamp
  • Limited PATH - Script explicitly adds /usr/local/bin:/opt/homebrew/bin
  • No HOME variable - Script uses config.yml:paths.default_home fallback

The script's validate_environment function enforces these requirements.

Development Commands

Testing

# Run full BATS test suite (110 tests)
./run_tests.sh

# Run only unit tests
bats test/unit/*.bats

# Run only integration tests
bats test/integration/*.bats

# Run inline tests (legacy, use run_tests.sh instead)
TEST_MODE=true ./transmission-done.sh

# Test specific scenario with mocked environment
TR_TORRENT_DIR=/path/to/test \
TR_TORRENT_NAME="Test.Show.S01E01" \
TEST_MODE=true \
./transmission-done.sh

Installation & Configuration

# Run installation wizard (checks deps, validates Plex, generates config)
./install.sh

# Validate config file syntax
yq eval '.' ~/.config/transmission-done/config.yml

# Test Plex connectivity
curl -H "X-Plex-Token: YOUR_TOKEN" http://localhost:32400/identity

Linting

# Check shell script quality (must be clean, no disable directives)
shellcheck transmission-done.sh install.sh run_tests.sh

# Check test scripts
shellcheck test/test_helper.bash
shellcheck test/unit/*.bats
shellcheck test/integration/*.bats

# Verify executable permissions
chmod +x transmission-done.sh install.sh run_tests.sh

Debugging

# Watch processing logs in real-time
tail -f ~/.local/state/transmission-processing.log

# Inspect Transmission environment (add to script temporarily)
env > /tmp/transmission-env.log

# Verify Plex library sections
curl -H "X-Plex-Token: YOUR_TOKEN" http://localhost:32400/library/sections

Configuration File Structure

version: 1.0
paths:
  default_home: /Users/username  # HOME fallback for Transmission context
plex:
  server: http://localhost:32400  # Plex server URL
  token: abc123xyz                # Obtained during install.sh
  media_path: /path/to/plex/media # Target directory for organized media
logging:
  file: .local/state/transmission-processing.log  # Relative to home
  max_size: 10485760  # Bytes (10MB) - triggers log rotation

Important Constraints

Bash Script Rules (from Andrew's guidelines)

  1. GNU Bash 5.x compatible only - No Bash 3.x fallbacks
  2. All shellcheck issues must be resolved - Errors, warnings, AND info
  3. NEVER use # shellcheck disable directives
  4. Critical: Never use ((var++)) with set -e - when var=0, this exits silently. Use ((var += 1)) instead
  5. Remove unused variables completely - Don't suppress with _var or shellcheck disables
  6. stdout/stderr discipline: User messages to stderr (>&2), only data to stdout for capture with $(...)

Testing Philosophy

  • Tests are NOT optional - every behavior change requires test updates
  • Test mode mocks external dependencies (Plex API, FileBot, df, stat)
  • Tests run in isolated temp directories (TEST_TEMP_DIR)
  • BATS tests use custom assertions in test_helper.bash
  • Expected: 114 tests passing (84 unit + 30 integration)

Plex API Hardcoded Sections

The script assumes:

  • Section ID 1 = Movies
  • Section ID 2 = TV Shows

If your Plex has different IDs, modify trigger_plex_scan function.

FileBot Requirements

  • Licensed version required (free trial won't work for automation)
  • Must be in PATH or at /usr/local/bin/filebot or /opt/homebrew/bin/filebot
  • Format string {plex} is Plex-optimized naming convention
  • Database selection: TheTVDB for TV, TheMovieDB for movies

NAS Storage Configuration

When media files are stored on a NAS (e.g., Synology) mounted via NFS, the NFS export settings are critical for FileBot to successfully move files.

NFS Export Requirements

The NFS rule on the NAS shared folder must be configured as follows:

Setting Required Value Why
Privilege Read/Write FileBot needs to move files and create directories
Squash Map all users to admin See UID mapping below
Security sys Standard AUTH_SYS authentication
Enable asynchronous Yes Performance
Allow non-privileged ports Yes Required for macOS NFS clients
Allow subfolder access Yes Media lives in subdirectories

UID Mapping Explained

NFS permission checks happen server-side, not client-side. The macOS noowners mount option only affects local display — the NAS still enforces permissions based on the UID it receives.

The challenge: macOS user UIDs (typically 501, 502) don't match Synology user UIDs (admin = 1024, other users = 1025+). With "No mapping" squash, the NAS receives the macOS UID and checks it against file ownership — which will fail because no Synology user has UID 501/502.

"Map all users to admin" solves this by mapping all NFS requests to the Synology admin user (UID 1024), which has full access to the shared folder.

Directory Permissions

With "Map all users to admin" squash, directories must be rwxrwxrwx (777). This is because:

  • All NFS requests arrive as UID 1024 (Synology admin)
  • Files/directories may be owned by different UIDs depending on what created them
  • The admin user needs write access via the "other" permission bits
  • This is acceptable for a media server on a firewalled LAN

If directories become non-writable (e.g., after an accidental chmod 755), fix from the NAS via SSH:

# On the Synology (requires sudo even for admin user):
sudo find /volume1/MediaShare/Media/ -type d -exec chmod 777 {} +

Do NOT attempt chmod from the NFS client — it will fail silently or with "Operation not permitted". Always SSH into the NAS.

NFS Attribute Cache

After changing permissions or ownership on the NAS, the NFS client will serve stale metadata from its attribute cache. To pick up changes:

# Unmount and remount the NFS share:
sudo diskutil unmount force /path/to/mount
sudo mount -t nfs -o noowners nas.local:/volume1/share /path/to/mount

Synology ACLs

Synology DSM defaults to "Windows ACL" mode on shared folders, which adds an ACL layer on top of UNIX permissions. The + suffix in ls -la output on the NAS indicates ACLs are present. These ACLs can deny writes even when UNIX permissions allow them. If experiencing unexplained permission issues, check the shared folder's permission model in DSM Control Panel → Shared Folder → Edit → Advanced.

Containerized Transmission Setup

When Transmission runs in a container (e.g., podman/haugene), it cannot directly invoke macOS scripts. The transmission-trigger-watcher.sh daemon bridges this gap:

  1. Container writes trigger files to a host-mounted directory on download completion
  2. The watcher daemon polls for trigger files every 60 seconds
  3. On finding a trigger, it maps the container's /data path to the macOS NFS mount path
  4. It invokes transmission-done.sh with the correct TR_TORRENT_* environment variables
  5. Dead-letter handling: after 5 failures, triggers are moved to .dead for manual inspection

VirtioFS NFS Access and Full Disk Access

The Podman VM shares host directories into the container via VirtioFS. For NFS-backed paths (like /Users/operator/.local/mnt/<share>), macOS enforces TCC/sandbox restrictions on the com.apple.Virtualization.VirtualMachine XPC service that handles VirtioFS I/O.

Symptom: Container shows "Operation not permitted" on /data/ while the host NFS mount works fine. stat() succeeds but opendir() fails (EPERM). Other VirtioFS mounts to local paths (/scripts, /config, /watch) work.

Root cause: The Virtualization.framework XPC service lacks Full Disk Access (FDA). On startup, VirtioFS opens directory handles that work for days/weeks. When the NFS server's idle timeout expires those handles, the XPC service must re-open them — and gets blocked by the sandbox.

Fix: Grant FDA to the Virtualization.framework binary via System Settings > Privacy & Security > Full Disk Access:

/System/Library/Frameworks/Virtualization.framework/Versions/A/XPCServices/
  com.apple.Virtualization.VirtualMachine.xpc/Contents/MacOS/
  com.apple.Virtualization.VirtualMachine

Defense in depth: The supervision loop in podman-machine-start.sh includes a /data/ read health check. If /data/ is inaccessible for 3 consecutive checks (15 minutes at 300s intervals), the loop automatically cycles the Podman VM and recreates the container to obtain fresh VirtioFS handles.

Note: This is the same class of issue as the LaunchAgent NFS opendir() EPERM (fixed 2026-04-04 by granting FDA to /bin/bash). macOS Tahoe enforces NFS access restrictions at the kernel level for any process without FDA, including system XPC services.

"podman-remote would like to access files on a network volume"

A separate grant (kTCCServiceSystemPolicyNetworkVolumes, per-user) is keyed on the podman-remote that ran podman machine start, because macOS holds that process responsible for the VM's VirtioFS access. Homebrew's podman-remote changes path and code hash on every brew upgrade, so the grant died daily and the supervisor blocked on the prompt after every VM start. The supervisor now runs podman from a stably-signed mirror at /usr/local/stable/podman/bin, maintained by the daily upgrade job. Full write-up, evidence and operations: docs/components/stable-signing.md.

"bash would like to access files on a network volume"

The same failure, on the FileBot side. The trigger watcher runs transmission-done (#!/usr/bin/env bash), and that bash is the binary macOS holds responsible for FileBot's reads on the NFS mount. With Homebrew bash, every bash upgrade voided the grant; on 2026-09-17 FileBot blocked on the prompt for 19 hours. The watcher now puts /usr/local/stable/bash/bin first in PATH. See docs/components/stable-signing.md.

Every podman call is bounded by a timeout

The supervision loop is single-threaded: it calls ensure_machine, ensure_container, and check_data_access synchronously. A podman command that never returns therefore freezes the whole loop — no health checks, no recovery, no log output.

That is not hypothetical. On 2026-08-16 the /data/ health check correctly detected a VirtioFS failure and triggered its documented recovery. The recovery's podman run -d was issued through a Homebrew-upgraded 6.1.0 CLI while a leftover 6.0.2 gvproxy, orphaned by the upgrade seven days earlier, still held the machine's socket paths. That podman run hung and was still running 27 hours later. Nothing was listening on port 9091 the entire time.

Two changes close that hole:

  • podman_t wraps every podman invocation in timeout(1). Command timeouts default to 60s, machine operations to 180s; override with PODMAN_CMD_TIMEOUT / PODMAN_MACHINE_TIMEOUT. A timeout is logged as TIMEOUT: 'podman run -d' exceeded 180s and was killed — deliberately distinct from an ordinary failure, because not being able to tell a hang from a failure is what made the outage take 27 hours to spot. The caller treats a timeout like any other failure and retries next cycle.

timeout(1) is required, not optional. It comes from coreutils, which config/formulae.txt installs. If it is missing the loop exits at startup with instructions rather than running unsupervised. A perl -e 'alarm' fallback was tried and rejected: SIGALRM kills perl but leaves the command's children running, so they hold the command-substitution pipe open and state=$(podman_t ... machine inspect) blocks forever regardless.

  • reap_machine_helpers verifies the helper processes actually exited. podman machine stop returning 0 does not mean gvproxy and vfkit are gone — the 6.0.2 gvproxy in the August incident survived for seven days. Before any machine start, and after the recovery path's machine stop, any surviving helper whose argv names transmission-vm gets SIGTERM, then SIGKILL if it ignores that. The match is scoped to the machine name so other podman machines on the host are untouched.

Podman version drift is the underlying trigger. A brew upgrade replaces the CLI without restarting the running VM or its helpers. The timeout and reap changes make the resulting mismatch recoverable rather than fatal; they do not prevent the drift itself. After upgrading podman, cycle the machine:

launchctl kickstart -k "gui/$(id -u)/com.<hostname>.podman-transmission-vm"

Common Gotchas

  1. PATH issues: Transmission runs scripts with minimal PATH (/usr/bin:/bin:/usr/sbin:/sbin). Script adds /usr/local/bin:/opt/homebrew/bin explicitly for Homebrew tools.

  2. Config location: Script checks ./config.yml first (development), then ~/.config/transmission-done/config.yml (production). Keep secrets out of repo - use config.yml.template.

  3. Log rotation: Happens on every script run if log exceeds max_size. Old log moved to .log.old (overwrites previous).

  4. Media type detection: Pattern matching is case-insensitive but looks for specific formats. If filename has neither TV episode markers nor year, script tries movie first, then TV as fallback.

  5. Plex token exposure: config.yml contains plaintext token. NEVER commit real config.yml (use config.yml.template).

  6. Test mode environment: TEST_RUNNER=true bypasses Transmission variable validation during BATS test execution. Don't use in production.

  7. Command substitution: Use var=$(function) to capture stdout. Functions must send user messages to stderr (>&2), not stdout, to avoid contamination.

  8. FileBot [MOVE] output parsing: FileBot uses the same [MOVE] prefix for both successful and failed moves. A successful move logs [MOVE] from [X] to [Y], while a failed move logs [MOVE] from [X] to [Y] failed due to I/O error [...]. When counting moved files, always exclude lines containing "failed" — use grep "\[MOVE\]" | grep -vc "failed", not grep -c "\[MOVE\]".

  9. NFS permission errors appear as FileBot I/O errors: If FileBot reports Access Denied with details like (rwxrwxrwx 502:20 file.mkv -> r-xr-xr-x 502:20 Season 02), the issue is NAS-side directory permissions, not FileBot configuration. See the NAS Storage Configuration section.

Dependencies

All must be available in PATH:

  • yq (brew install yq) - YAML parsing, required for config
  • xmlstarlet (brew install xmlstarlet) - XML parsing for Plex API responses
  • jq (brew install jq) - JSON parsing for Plex API authentication
  • curl (pre-installed on macOS) - HTTP requests
  • bats-core (brew install bats-core) - Test framework
  • filebot (licensed) - Media organization engine
  • find, grep, stat, df - Standard Unix utilities

The install.sh wizard checks and can auto-install missing dependencies.

Transmission Integration

Configure in Transmission:

  1. Preferences → Downloading tab
  2. Check "Run script when download completes"
  3. Enter symlink path: ~/.local/bin/transmission-done (or direct path: /path/to/transmission-done.sh)

Ensure script is executable: chmod +x transmission-done.sh

Security Notes

  • config.yml contains sensitive data: Plex token grants full server access
  • Git ignore prevents accidental commits: config.yml in .gitignore
  • Template provided: config.yml.template shows structure without secrets
  • Token validation: install.sh validates token before saving
  • Symlink safety: install.sh refuses to overwrite regular files, only symlinks

CI/CD Pipeline

GitHub Actions workflow (.github/workflows/test.yml):

  • BATS Unit Tests - Runs test/unit/*.bats
  • BATS Integration Tests - Runs test/integration/*.bats
  • Inline Tests - Runs TEST_MODE=true ./transmission-done.sh
  • ShellCheck Validation - Lints all shell scripts
  • Test Summary - Aggregates results

All 114 tests must pass before merging PRs.

Recent Major Changes

  • PR #23: Fixed false success detection — FileBot [MOVE] ... failed lines were counted as successes, masking I/O errors (e.g., NFS permission denied)
  • PR #21: Corrected torrent path and Plex scan section detection
  • PR #11: Installer now queries Plex for library paths, improved symlink handling
  • PR #10: Fixed stdout/stderr separation in install.sh (critical bug fix)
  • PR #8: Simplified test runner to only require bats-core
  • PR #2: Implemented comprehensive BATS test suite