What can we actually do — and where are the gaps?
From Dependency Map: 1 hard dependency (auth/identity), 8 soft. The question is no longer "what blocks us" but "what can we execute with what we have?"
CAPABILITY MAP: AGENT PLATFORM
════════════════════════════════════════════════════════════
CAPABILITY MATURITY GAP? CATEGORY
───────────────────────────────── ──────── ──── ────────
CONCERN 1: IDENTITY & MEMORY
Agent profile schema ████░ (4) Core
Memory store schema ████░ (4) Core
Agent ETL (load profiles) ███░░ (3) Core
Flexible path flag ░░░░░ (0) P1 Core
Session bootstrap █░░░░ (1) P1 Core
Session extract / consolidation ░░░░░ (0) P1 Core
Semantic/procedural recall ██░░░ (2) Core
Drift detection ░░░░░ (0) P3 Core
CONCERN 2: COMMUNICATION
Convex channels (5 seeded) ███░░ (3) Core
Message types (8 typed) ███░░ (3) Core
Agent comms CLI ████░ (4) Core
Block signalling ██░░░ (2) Core
Cross-team handoff ████░ (4) Core
Session recovery via messages ░░░░░ (0) P2 Core
Live activity feed ░░░░░ (0) P3 Supporting
CONCERN 3: DISPATCH & EXECUTION
Plan CLI (DB-native) █████ (5) Core
23 plan templates ████░ (4) Supporting
Priority dispatch █░░░░ (1) P1 Core
PRD→template mapping ░░░░░ (0) P2 Supporting
Session bootstrap from dispatch ░░░░░ (0) P2 Core
Commissioning dispatch ░░░░░ (0) P2 Core
CONCERN 4: VVFL ENFORCEMENT
Context graph schema █░░░░ (1) P2 Core
8-dimension auditors ░░░░░ (0) P1 Core
Pattern extractor ░░░░░ (0) P2 Core
Memory writer (cross-session) ░░░░░ (0) P2 Core
Action generator ░░░░░ (0) P3 Supporting
CONCERN 5: UNIFIED CLI (drmg)
Shared DB context ░░░░░ (0) P1 Core
Thin router ░░░░░ (0) P1 Core
Plan wrapper █░░░░ (1) Supporting
Agent module █░░░░ (1) Supporting
VVFL module ░░░░░ (0) P2 Supporting
Data wrapper █░░░░ (1) Supporting
GRADUATION PATH
API routes ░░░░░ (0) P3 Core
Agent Card (A2A) ░░░░░ (0) P3 Core
════════════════════════════════════════════════════════════
Summary
| Maturity | Count | % |
|---|---|---|
| 0 (Not Built) | 16 | 48% |
| 1 (Ad-hoc) | 5 | 15% |
| 2 (Repeatable) | 2 | 6% |
| 3 (Defined) | 4 | 12% |
| 4 (Managed) | 5 | 15% |
| 5 (Optimised) | 1 | 3% |
| Total | 33 |
Gap Prioritisation
| Priority | Count | Capabilities | Phase |
|---|---|---|---|
| P1 (Critical) | 7 | Flexible path, Session bootstrap, Session extract, Priority dispatch, 8-dimension auditors, Shared DB context, Thin router | Phase 0-1 |
| P2 (Important) | 8 | Session recovery, PRD→template, Bootstrap from dispatch, Commissioning dispatch, Context graph, Pattern extractor, Memory writer, VVFL module | Phase 2-4 |
| P3 (Future) | 4 | Drift detection, Live activity feed, Action generator, Graduation (API+A2A) | Phase 5-7 |
Category Analysis
| Category | Count | Investment Strategy |
|---|---|---|
| Core (differentiates — this IS the product) | 26 | Build internally, maximum investment |
| Supporting (enables core) | 7 | Make efficient, leverage existing CLIs |
| Generic (commodity) | 0 | N/A — no generic capabilities needed |
The platform IS the product. 79% of capabilities are Core — this is correct for a platform PRD. Supporting capabilities are wrappers around existing CLIs that don't differentiate.
Composition Analysis
| Source | Count | % |
|---|---|---|
| Existing infrastructure (maturity ≥ 2) | 12 | 36% |
| Partial (maturity 1) | 5 | 15% |
| New build (maturity 0) | 16 | 48% |
Unlike the Sales Dev Agent (80% composition, 20% new), the Agent Platform has a heavier build ratio (48% new). This is expected — you build the platform once, then compose agents on top of it.
Dependency Cross-Check
From Dependency Map: P1 gaps have only soft dependencies. Critical finding:
| P1 Gap | Blocking Dependency | Type | Implication |
|---|---|---|---|
| Flexible path flag | None | — | Can start immediately (0.5 day) |
| Session bootstrap | D1 (shared DB) | SOFT | Mock-able, start with local JSON |
| Session extract | D1 (shared DB) | SOFT | Mock-able |
| Priority dispatch | D1, D2, D5 (Convex) | SOFT | D5 already done, D1+D2 are first tasks |
| 8-dimension auditors | D1, D2 (thin router) | SOFT | Chain: D1→D2→D8, all engineering |
| Shared DB context | None | — | Can start immediately (0.5 day) |
| Thin router | D1 | SOFT | Immediate after D1 |
No P1 gap is blocked by a hard dependency. Engineering can start the entire P1 backlog immediately. The only hard dependency (D4: Identity PRD) blocks production auth, which is a P3 concern.
Gate
Before moving to A&ID:
- All capabilities assessed with evidence-based maturity — YES (33 capabilities, 0-5 scale with infrastructure evidence)
- Capabilities categorised — YES (26 Core, 7 Supporting, 0 Generic)
- Gaps identified and prioritised — YES (7 P1, 8 P2, 4 P3)
- Investment strategy matches category — YES (Core = build, Supporting = leverage existing)
- Dependency cross-check present — YES (table confirms no P1 gap blocked by hard dep)
Context
- Dependency Map — Previous: what blocks us
- A&ID — Next: how agents orchestrate
- Capability Map Template — The empty pattern
- Agent Platform PRD — Full depth