Skip to content

TARS v2.0.0

Choose a tag to compare

View all tags
@github-actions github-actions released this 10 Feb 01:41
· 131 commits to main since this release
v2.0.0
968b4fe

Architecture overhaul: skill consolidation, integration abstraction, automated housekeeping, and enterprise distribution

Breaking changes

  • Consolidated 28 skills into 12 (7 background skills merged into single core skill, workflow skills merged by domain, 2 composite skills absorbed)
  • Reduced 23 commands to 11 thin wrappers matching consolidated skills
  • Renamed /setup to /welcome with progressive onboarding
  • Replaced pipe-delimited index tables with YAML format
  • Replaced hardcoded Eventlink/remindctl references with provider-agnostic integration registry
  • Skill help metadata changed from flat string to structured object (purpose, use_cases, invoke_examples, common_questions, related_skills)

Added

  • Skill consolidation: core (7→1), meeting (2→1), tasks (2→1), learn (2→1), think (5→1), initiative (2→1), maintain (3→1)
  • 9 automation scripts: health-check.py, rebuild-indexes.py, scaffold.sh, sync.py, archive.py, verify-integrations.py, scan-secrets.py, bump-version.py (all Python stdlib only)
  • Provider-agnostic integration registry: category-based design supporting http-api, cli, and mcp provider types
  • Progressive welcome flow: 3-phase onboarding (instant setup, first win, background learning) replacing 4-round interrogation
  • YAML-based indexes: improved editability and parseability over pipe tables
  • 4-tier content archival: durable/seasonal/transient/ephemeral with automatic expiry
  • Batch processing inbox: inbox/pending → processing → completed/failed with isolated sub-agent processing
  • Sub-agent parallelization: meeting (task extraction || memory extraction), think deep mode (validation || executive council), briefing (calendar || tasks || memory)
  • Automated daily housekeeping: session-start check triggers maintenance scripts if not run today
  • Inline help metadata: structured help objects in all skill YAML frontmatter
  • Sensitive data guardrails: script-based scanning (scan-secrets.py) with configurable patterns in reference/guardrails.yaml
  • CI/CD pipeline: GitHub Actions for PR validation (validate.yml) and auto-release (release.yml)
  • Test suite: 6 validation scripts + test runner with dynamic test selection
  • CATALOG.md: leadership adoption document for enterprise distribution
  • Maturity model: reference/maturity.yaml tracking onboarding progress
  • Getting started guide: reference/getting-started.md for new users
  • Workflow patterns: reference/workflows.md documenting multi-skill patterns
  • Communication rules go global: BLUF, anti-sycophancy, banned phrases now apply to ALL TARS outputs
  • AskUserQuestion integration: structured UI clarification in Cowork mode
  • TodoWrite progress tracking: real-time visibility during long-running operations
  • Context management directive: intermediate results offloaded to files
  • Proactive learning triggers: TARS suggests memory extraction when user corrects facts or shares context
  • Help routing: core skill handles "how do I", "what can you do" queries via inline help metadata
  • Cowork shortcut support: scheduled daily housekeeping via create-shortcut skill

Changed

  • Session baseline: ~115 tokens → ~48 tokens (12 skills × ~4 tokens vs 28 × ~4)
  • Version: 1.5.0 → 2.0.0
  • Plugin name: lowercase "tars" in manifest
  • Skill descriptions: optimized with trigger keywords for better auto-routing accuracy
  • Index format: pipe-delimited markdown tables → YAML entries
  • Welcome flow: 4-round interrogation (10+ min) → 2 questions + auto-discovery (5 min)
  • Integration references: hardcoded tool names → category-based ("query calendar integration")

Removed

  • 7 individual background skill directories (identity, communication, memory-management, task-management, decision-frameworks, clarification, routing) — merged into core
  • 2 composite skill directories (meeting-processor, deep-analysis) — absorbed into meeting and think
  • 8 individual workflow skill directories (process-meeting, extract-tasks, manage-tasks, extract-memory, extract-wisdom, housekeeping, rebuild-index, update, setup, quick-answer, strategic-analysis, executive-council, validation-council, discovery-mode, performance-report, create-artifact) — consolidated into domain skills
  • 12 command files — consolidated to match 11 merged skills
Assets 3
Loading