March 11, 20261.0.135v1.0.135RSS

v1.0.135 — feat: Automated Agent Memory & Codebase Knowledge Base


title: "Release v1.0.135: Automated Agent Memory & Codebase Knowledge Base" date: 2026-03-11

Release v1.0.135: Automated Agent Memory & Codebase Knowledge Base

What changed

  • New agent_memory_entries table — stores 6 entry types (pattern, mistake, convention, dependency, gotcha, success) across 9 categories, with confidence scoring (0–1) and source tracking (agent / manual / distiller)
  • Nightly distiller — Inngest cron at 3 AM UTC scans the last 24h of pipeline logs and auto-extracts structured entries (TS errors, Drizzle .where() chaining mistakes, force-dynamic gotchas, successful PR patterns, and more)
  • Reinforcement loopupsertMemoryEntry() bumps confidence each time the same pattern is observed again; high-confidence entries are injected into every implementation agent's system prompt
  • Dashboard UI — new /dashboard/products/[id]/agent-memory page with stats cards, type/category filters, expandable entry cards, human verification, and manual entry creation
  • 11 unit tests covering formatMemoryForInjection, the core function that renders entries into agent prompts

Why it matters

Agents no longer start from scratch on every pipeline run. Mistakes caught once are remembered, reinforced, and eventually prevent the same error from recurring — improving CI green rates over time without any manual intervention.

Technical notes

  • New DB migration required for agent_memory_entries table
  • agentMemory?: string added to PipelineEventData — non-breaking, loaded best-effort in the pipeline orchestrator
  • Human-verified entries are boosted to 0.95 confidence; agent-extracted entries start at 0.5

Release Notes

Everything is done. Here's a summary of what was accomplished:


✅ Release v1.0.135 Published

PR #150 — Merged ✓

feat: Automated Agent Memory & Codebase Knowledge Base

The PR had 3 consecutive green CI runs on the branch before closing. The merge tool initially returned a stale failure referencing an earlier failed run, but the PR was confirmed merged (state: closed).

Release Created ✓


What Was Released

This is a minor feature release introducing a persistent, self-improving agent memory layer — one of the most impactful quality-of-life upgrades to the autonomous development pipeline:

ComponentDescription
🗄️ DB SchemaNew agent_memory_entries table with 6 entry types × 9 categories, confidence scoring, and source tracking
📚 Memory LibrarygetRelevantMemory(), formatMemoryForInjection(), upsertMemoryEntry() with reinforcement learning
🔌 tRPC RouterFull CRUD API — list, stats, create, update, delete (soft), verify
Nightly DistillerInngest cron at 3 AM UTC that auto-extracts patterns from pipeline logs (TypeScript errors, Drizzle gotchas, conventions, successes)
🖥️ Dashboard UINew /dashboard/products/[id]/agent-memory page with stats, filters, expandable cards, and manual entry creation
🤖 Agent InjectionMemory injected into implementation system prompts on every pipeline run
🧪 Tests11 unit tests for the core formatMemoryForInjection function