Every domain of the platform, specified before it was built. Each document is written from source and links to the repository that implements it.
Building a Claude Code plugin ecosystem — hooks, skills, agents, commands, and MCP integration.
PRD 2 of 25Multi-agent orchestration — 37 agents, 2,046-line state schema, tiered quality gates, Gemini validation, NHI lifecycle, BRD-driven development.
PRD 3 of 25Context window management — 4-tier escalation, velocity prediction, auto-memory, CLAUDE.md hierarchy, and compaction recovery.
PRD 4 of 25The post-Barman tiered persistent vector memory system — Qdrant (hot/warm) + pgvector cold tier + Memgraph links, 74 MCP tools, 34 n8n workflows…
PRD 5 of 25Agent governance — identity manifests, trust levels, 3-tier tool classification, append-only audit, constitutional contracts, memory governor…
PRD 6 of 25HTML-to-markdown proxy — converts web pages to clean markdown for AI agent consumption with Playwright fallback.
PRD 7 of 25A unified code assurance platform — 27 integrated tools powering 37 analyzers, scan profiles, 6-stage enrichment, 1000-point scoring, cryptographic…
PRD 8 of 25Data lineage, quality validation, pipeline observability, and continuous classification for agent-generated insurance systems.
PRD 9 of 25Per-agent cost tracking, budget caps, token optimization, model routing, and cost dashboards for an autonomous AI development factory targeting…
PRD 10 of 25Runtime behavioral monitoring, agentic identity lifecycle, memory integrity verification, inter-agent coordination scoring, and guardian agent…
PRD 11 of 25Self-Healing Workflows — automated recovery for multi-agent orchestration failures with pattern-based classification, YAML strategy playbooks, and…
PRD 12 of 25Event-Driven Automation — centralized event taxonomy with 11 categories, YAML-defined routing rules, n8n workflow integration, SQLite dead letter…
PRD 13 of 25Process Knowledge Base — structured knowledge store for decision rules, SOPs, decision trees, and edge-case catalogs organized into domain packages…
PRD 14 of 25Outcome Measurement — Quantify the value delivered by multi-agent orchestration through passive observation of task completion, quality trends, and…
PRD 15 of 25Predictive Scaling — Anticipate workload patterns, pre-warm caches, and optimize model routing using statistical forecasting instead of reactive cost…
PRD 16 of 25A2A External Agent Interoperability — Enable external agents to participate in conductor workflows and expose internal agent capabilities via A2A…
PRD 17 of 25Regulatory compliance framework — human attribution, immutable audit trail, evidence packages, data subject rights, model cards, incident response.
PRD 18 of 25Compliance portal — auditor access, human gate decisions, DSR management, evidence packages, model card governance, regulatory reporting for GDPR…
PRD 19 of 25A Claude Code `Stop` hook that intercepts an assistant's unverified “done” claims and routes them to a local model whose only job is to disprove them…
PRD 20 of 25A single-file FastAPI application that discovers every project under a documentation root, renders their Markdown and HTML docs in the browser, and…
PRD 21 of 25A FastAPI service running four always-on schedulers that publish hourly Merkle-chained roots over audit databases, enforce per-class retention with…
PRD 22 of 25A governed data pipeline that extracts, transforms, masks, and loads data under JSON-schema contracts — with consistent tokenization, PROV-AGENT…
PRD 23 of 25A self-hosted brand and design-asset library: a Next.js admin UI and a standalone Express REST API serving brand identities, design tokens, component…
PRD 24 of 25A self-hostable, always-on personal assistant reachable over Telegram and SMS, driven by Claude Code CLI sessions with vector memory, a memory-hooks…
PRD 25 of 25The Conductor plugin pair — conductor-kernel (19 agents, 14 skills) and conductor-dev (18 agents), with tiered quality gates, 5-signal classification…