Memory Context Pipeline v1.2.1

Release Notes - v1.2.1

Added

• skill/scripts/orchestrate_memory_cycle.py
  • One-command chain: trigger -> route -> store
• docs/ORCHESTRATOR.md
• examples/orchestrator-example.ps1

Why

Turns individual helpers into a practical single-step operator workflow.

Memory Context Pipeline v1.2.0

Release Notes - v1.2.0

Added

• skill/scripts/validate_vision_tasks.py for multimodal route validation
• skill/scripts/enforcement_check.py for 10k compaction posture checks
• docs/VALIDATION.md with tuning guidance and operator runbook

Updated

• Config template now separates text vs vision confidence thresholds
• README includes validation commands and enforcement clarification

Intent

Close the loop on:

• screenshot/vision task validation
• confidence tuning for multimodal extraction
• clear understanding of soft vs practical hard enforcement above 10k

v1.1.3 - Deterministic Multimodal Router

Release Notes v1.1.3

Added

• Deterministic multimodal branch in context_router.py
  • --has-image flag for explicit image-bearing turns
  • vision task routing (image-caption, screenshot-summary, ocr-extract)
  • route output now includes model, fallback_chain, confidence, and decision_trace

Improved

• Default local fallback behavior now explicit: local (lfm2) -> premium (opus) for local-routed turns
• Router output now includes needs_premium_review signal for low-confidence paths

Tests

• Added unit tests for routing behavior in tests/test_context_router.py
  • premium default path
  • local task path
  • token threshold trigger
  • image/vision routing paths

Memory Context Pipeline v1.1.2

Housekeeping release: ignore runtime .trigger_state.json files from version control.

Memory Context Pipeline v1.1.1

Release Notes - v1.1.1

Added

• skill/scripts/auto_trigger.py

  • Auto-trigger decisions for channel-memory writes using:
    • context-length threshold
    • explicit remember intent
    • periodic digest interval
  • Per-channel trigger state file for throttling and consistency

• docs/AUTO_TRIGGERS.md

• examples/auto-trigger-example.ps1

Improved

• Clarified automation flow from trigger decision -> memory append
• Strengthened practical path toward fully automatic channel organization

Memory Context Pipeline v1.1.0

Release Notes - v1.1.0

Added

• skill/scripts/channel_memory_store.py for automatic channel-scoped append logs
• skill/references/channel_layout.md documenting the lean folder structure
• examples/channel-memory-example.ps1 for Windows usage

Channel memory layout

• memory/channels/<guild_id>/<channel_id>/YYYY-MM-DD.md
• memory/channels/<guild_id>/<channel_id>/summaries/YYYY-Www.md

Design intent

• Better long-run organization for active Discord environments
• Lower retrieval noise by scoping memory per channel
• Keep writes efficient via append-only + trigger-based events

Memory Context Pipeline v1.0.1

Release Notes - v1.0.1

Changed

• Clarified 10k behavior as a soft compaction threshold (not a hard cap)
• Added practical model guidance for choosing lfm2 vs qwen3.5:35b
• Added Windows gh CLI path note for environments where GitHub CLI is installed but not on PATH

Why this matters

• Prevents confusion about strict token limits
• Helps operators pick the best local model for speed vs quality
• Reduces friction when publishing releases from Windows hosts

Memory Context Pipeline v1.0.0

Release Notes - v1.0.0

Added

• Initial memory-context-pipeline skill
• Router script (context_router.py)
• Output validator (validate_output.py)
• Worker prompt reference
• JSON schema output contract
• Config template for local/premium routing
• Packaged skill artifact support (.skill)

Operational changes

• Discord thread-bound subagent spawn support configured
• Automatic memory flush compaction policy configured at 10k soft threshold

Recommended model strategy

• Fast path: lfm2
• High-quality local pass (if hardware allows): qwen3.5:35b

Known limitations

• 10k behavior is a soft compaction threshold, not a hard token firewall
• Full automation quality depends on local model health/latency