ADR 0008 hierarchical memory scoping

ADR 0008: Hierarchical Memory Scoping

Status: Implemented Date: 2026-02-04 Memory ID: 20260204191624-14925ea2

Context

The hive monorepo contains multiple subprojects. We need:

• Parents see children's memories (downward visibility)
• Children isolated from parents (upward isolation)

Decision

Configuration Cascade (local precedence)

~/.config/hive.edn              # Global defaults
~/project/.hive-project.edn     # Project overrides
~/project/sub/.hive-project.edn # Submodule overrides (highest)

Memory Visibility (asymmetric scoping)

hive-mcp/                    # Sees all below
├── hive-connectors/         # Sees its children only
│   ├── slack/               # Isolated
│   └── github/              # Isolated
└── hive-addons/             # Sees its children only
    ├── diagrams/            # Isolated
    └── presentations/       # Isolated

Hybrid Approach (Option E - Recommended)

Resolution Priority:

1. Check .hive-project.edn for explicit :parent-id
2. Check persistent index (DataScript) for cached hierarchy
3. If stale or missing, scan filesystem and update index
4. Fall back to string-based inference (colon-delimited)

Implementation Phases

Phase	Scope	Effort
1	Add .hive-project.edn to subprojects	Low
2	Persistent DataScript index	Medium
3	Downward visibility (descendant-scopes)	Medium
4	Multi-repo workspaces (~/.config/hive.edn)	Future

NIH Research

Tool	Hierarchy Model	Key Insight
Nx	workspace.json	"Affected" detection via dep graph
Polylith	workspace.edn	Single source of truth
deps.edn monorepos	:local/root	Sean Corfield pattern
Git submodules	.gitmodules	Recursive traversal

Consequences

Positive

• Subprojects get isolated memories
• Parent aggregates children's memories
• Incremental adoption

Negative

• Requires adding .hive-project.edn to each subproject
• Cache invalidation complexity