v2.1.3

Bug Fix

• fix: Rename loop variable shadowing global info function in CLI

  The info loop variable in cmd_run() shadowed the global info() function, breaking barbossa run <agent> commands with UnboundLocalError.

v2.1.2

Full Changelog: v2.1.1...v2.1.2

v2.1.1

Full Changelog: v2.1.0...v2.1.1

v2.1.0 - CLI Improvements & Unified Prompts

What's New

CLI Improvements

• barbossa doctor - Comprehensive system diagnostics in one command
• barbossa watch - Tail all agent logs in real-time
• Command aliases: barbossa engineer, barbossa tl, barbossa disco, barbossa prod
• Agent descriptions in barbossa run help output
• Better post-run guidance

Prompts Overhauled

• Engineer: 326 → 199 lines (-39%)
• Tech Lead: 360 → 171 lines (-52%)
• All prompts now include "Team Context" section explaining agent roles
• Consistent GitHub CLI usage across all agents
• Clear handoff documentation (Discovery → Engineer → Tech Lead)

Full GitHub Integration

• Consistent labeling: backlog, discovery, product, feature, spec
• Product Manager now creates GitHub issues directly (was JSON-only)
• Auditor checks GitHub PRs/issues via gh CLI
• Discovery checks for duplicates before creating issues

Config

• Minimal Config Template - config/repositories.minimal.json for quick setup (7 lines)

Docs

• Quickstart, FAQ, troubleshooting updated with new commands

---

Full Changelog: https://github.com/ADWilkinson/barbossa-dev/blob/main/CHANGELOG.md

v2.0.2 - Simplified to GitHub Issues Only

Removed

• Linear Integration - Removed all Linear support in favor of GitHub Issues only
  • Deleted linear_client.py and associated tests (~900 lines)
  • Simplified issue_tracker.py to GitHub-only (~200 lines removed)
  • Removed Linear config from schema, env files, and docker-compose
  • Why: Simplifies codebase, GitHub Issues works out of the box with zero config

Changed

• README Rewrite - Clearer value proposition with "Why Barbossa?" section
• Docs Simplification - Removed Linear references from all documentation pages

Added

• OG Meta Tags - Social preview cards for docs site (Twitter, Open Graph)
• GitHub Stars Badge - Added to README

Full Changelog: v2.0.1...v2.0.2

v2.0.1

Full Changelog: v2.0.0...v2.0.1

v2.0.0 - Failure Backoff & Stale Cleanup

What's New in v2.0.0

Added

• Stale Issue Cleanup (Optional) - Auditor can close stale issues by label (GitHub only)
  • Configurable via settings.auditor.stale_issue_days and stale_issue_labels
• Failure Backoff Visibility - Engineer prompt now surfaces issues currently in backoff to prevent retry loops

Changed

• Pending CI Timeout - Tech Lead can request changes if checks stay pending past a configurable timeout
  • Default: 6 hours (settings.tech_lead.pending_checks_timeout_hours)
• Stale PR Detection - Tech Lead closes stale PRs based on last update time (not just creation time)
• Docs Updated - Configuration docs now include pending CI timeout, failure analyzer, and stale issue cleanup

Config Changes (optional)

{
  "settings": {
    "tech_lead": {
      "pending_checks_timeout_hours": 6
    },
    "auditor": {
      "stale_issue_days": 30,
      "stale_issue_labels": ["discovery", "product", "feature"]
    }
  }
}

Upgrade Steps

docker compose pull
docker compose up -d --build
docker exec barbossa barbossa health

Full Changelog: https://github.com/ADWilkinson/barbossa-dev/blob/main/CHANGELOG.md

v1.8.3

Changed

Auditor Refocus: System Health Only 🎯

The auditor now focuses purely on system health, not code quality.

Rationale: Code quality is Discovery/Product's domain. The Auditor should monitor the pipeline itself, not the code being managed.

Removed

Quality Issue Creation

Auditor no longer creates backlog issues for code quality:

• Discovery agent handles code quality issue creation (missing tests, TODOs, etc.)
• Prevents duplicate/overlapping responsibilities between agents
• Quality patterns still logged and saved for Tech Lead insights

Fixed

Health Score Calculation

Now based only on system metrics:

• Session success/failure rate (are agents running?)
• Error rate (API failures, auth issues)
• Merge rate (is the pipeline producing usable PRs?)
• Timeouts (are tasks completing?)

NOT penalized for: missing integration tests, UI issues, code bloat (these are quality issues, not system issues)

Result: Health score now 75/100 (was 0/100) - reflects actual system health

Discord Notifications

• "Critical Issues" renamed to "System Issues"
• Only counts actual system problems, not code quality patterns

v1.8.2

Fixed

Auditor Session Detection Bug 🔧

Fixed false 0/100 health scores caused by flawed session success/failure detection.

Root Cause:

• Success detection looked for narrow patterns ("Successfully", "PR created successfully") that didn't match actual log output ("SUCCESS", "PR created:")
• Failure detection was too broad ('error' in content.lower()) catching false positives

Fix:

• Now detects actual success patterns: ": SUCCESS", "PR created:", ": ADDRESSED"
• Now detects actual failure patterns: "- ERROR -" (log level), ": FAILED"
• Sessions with neither clear success nor failure are not counted (e.g., no work to do)

This caused 100% session failure rate even when all sessions were successful, driving health scores to 0/100.

v1.8.1

Full Changelog: v1.8.0...v1.8.1