c25. AI Ecosystem Inventory Map Plan

Goal

Build a qmap inventory/index view that shows the company-wide AI ecosystem for a selected timeframe:

• active agents as the center of gravity
• repositories around those agents
• files attached to repositories
• MCP servers, tools, and sub-agents as surrounding capability surfaces

The result should behave like a product-facing inventory view, not just a session debug harness.

Phase A: Lock The Semantic Contract

Before deepening the wrapper, confirm the intended model:

• one scenario represents one timeframe-bounded ecosystem snapshot
• agents are the top-level active entities
• repositories are first-class actors, not just tags on files
• files carry repo tags so grouping can nest file actors under repository names
• tools and sub-agents remain distinct via entityType tags even if both use the skill mesh family

Recommended wrapper name:

• QMapInventoryIndex

Recommended first-pass actor mapping:

• agent -> AI agents
• app -> repositories
• file -> files
• mcp-server -> MCP servers
• skill -> tools and sub-agents

Phase B: Stand Up The Synthetic Data Harness

Create a deterministic data module first so layout tuning is repeatable.

Required scenario tiers:

• demo — tiny hand-checkable story for screenshots and wrapper tuning
• teamspace — one realistic cross-team working set
• enterprise — large but still interpretable enterprise topology
• global-enterprise — intentionally oversized stress case

Each scenario should export:

• actors
• dataflows
• stats
• meta

Recommended graph behavior:

• agent -> repository flows encode active work
• repository -> file flows encode file membership and file-touch concentration
• agent -> MCP server flows encode capability usage
• agent -> tool flows encode direct tool use
• agent -> sub-agent flows encode delegation
• optional sub-agent -> repository flows encode delegated work when useful

Phase C: Build The qmap Experiment First

In dev/experiments/c25.ai-ecosystem-inventory-map/:

1. create the experiment shell
2. load the synthetic scenarios
3. wire grouping mode controls
4. show compact stats for actors, flows, and major entity families

The experiment should remain the place where density is judged, not the wrapper alone.

Phase D: Build And Tune The Wrapper

Create src/components/QMapInventoryIndex.vue.

The wrapper should:

• accept one inventory snapshot object
• reduce that snapshot to normal qmap actors and dataflows
• apply tuned defaults distinct from single-session wrappers
• expose a small, stable grouping API

The wrapper should not depend on experiment-local files.

Phase E: Pressure-Test Inventory Density

Focus the tuning pass on the problems this view is uniquely likely to hit:

• many agents in the same focal region
• repository fans that are wider than single-session MCP fans
• file rugs that may become much denser than c21
• mixed skill semantics where tools and sub-agents coexist
• long repository and file labels

Judgment criteria:

• active agents remain easy to spot first
• repositories feel like the middle layer
• files visually read as attached to repositories rather than as a separate random cloud
• MCP/tool/sub-agent surfaces stay interpretable as ecosystem capabilities

Phase F: Decide What Needs Promotion

Only after the experiment feels convincing should we decide whether qmap needs:

• a dedicated repository mesh
• a dedicated tool mesh
• inventory-specific default presets beyond the initial wrapper
• a summarization helper for extremely large repo/file sets

Default recommendation:

• defer new mesh work until the wrapper proves the semantic shape is right

Scenario Matrix

The deterministic generator is designed to land roughly at:

• demo -> ~3 agents, ~3 repos, ~12 to 14 files, 3 MCP servers, 4 tools, 2 sub-agents
• teamspace -> ~9 agents, ~9 repos, ~50 files, 6 MCP servers, 8 tools, 5 sub-agents
• enterprise -> ~38 agents, ~25 repos, ~240 files, 10 MCP servers, 16 tools, 14 sub-agents
• global-enterprise -> ~110 agents, ~48 repos, ~700 files, 14 MCP servers, 24 tools, 32 sub-agents

The exact totals should remain deterministic per seed, but slightly irregular counts are fine if they produce a more believable topology.

Near-Term Recommendation

Do not start by inventing new visuals.

Start by proving:

• existing qmap meshes can express the inventory semantics
• repository/file relations are spatially readable
• the deterministic synthetic scenarios are rich enough to tune against

If those three hold, the cycle is worth deepening immediately.