docs/v6-spec-and-roadmap/v6-1/v6-1-priorities
v6.1 Priorities — Actionable Work Items reference
Work items derived from the gap inventory, ordered by impact and feasibility.
Continue reading
Nearby pages in the same section.
v6.1 Priorities — Actionable Work Items
Work items derived from the gap inventory, ordered by impact and feasibility.
P0: Fix Before Next Release
These are active regressions or blockers that prevent stable CI/CD.
| Item | Gap ID | Layer | Description | Status |
|---|---|---|---|---|
| Codex+google websocket proxy bypass | G-L3-02 | L3 | Codex websocket connections bypass OPENAI_BASE_URL (#200) | Open |
| hooks-adapter-cli version hash inconsistency | G-X-04 | Cross | 12-char vs variable-length SHA in staging versions causes ETARGET | Fixed (publish.yml) |
| Windows npm spawn | G-X-03 | Cross | shell: true needed on Windows for npm.cmd | Fixed |
| ESM process module loading | G-L8-01 | L8 | NODE_PATH hack for ESM import resolution (#196) | Fixed (workaround) |
| Process code drift | G-L13-02 | L13 | Stable invocation keys for replay (#169, PR #178) | In review |
P1: Architectural Improvements
These address S2 (Major) gaps that cause ongoing friction.
1. Unified Provider Abstraction (G-L2-01, G-L2-02)
**Problem:** Provider routing is split across transport-adapter (proxy), adapters-cli (translation), and agent-catalog (metadata).
**Proposed:** Extract a ProviderResolver interface that:
- Resolves provider + model → transport config + credentials
- Lives in agent-catalog (metadata-backed) or a new thin package
- Used by transport-adapter proxy and adapters-cli launch
**Effort:** Medium. Seam already exists in agent-catalog's getProviderTranslation().
2. Explicit Workspace Contract (G-L7-01)
**Problem:** The workspace concept (what the agent sees/modifies) is implicit in .a5c/runs/ layout.
**Proposed:** Define a WorkspaceContract interface in babysitter-sdk:
resolveWorkspace(runDir)→{ root, processDir, artifactsDir, runsDir }- Used by ProcessContext, task execution, and plugin install
- Maps to graph's Workspace and Worktree node kinds
**Effort:** Small. Most paths already exist in scattered utilities.
3. Transport Codec Completion (G-L3-01)
**Problem:** Codec architecture (tool schema translation, cost normalization) was planned but partially implemented.
**Proposed:** Complete the codec refactor per the existing plan:
- TransportCodec interface per protocol (anthropic, openai-chat, google, bedrock)
- Tool schema translation between formats
- Cost/usage normalization across providers
- Already has a plan document in the repo
**Effort:** Large. Significant API surface across multiple protocols.
4. Unified Governance API (G-L14-01)
**Problem:** Governance concerns scattered across 3 packages.
**Proposed:** Define a GovernancePolicy interface in babysitter-sdk:
- Approval routing (delegates to tasks-adapter)
- Evidence claims (delegates to atlas)
- Trust level evaluation (backed by graph TrustLevel records)
**Effort:** Medium. Integration layer, not new functionality.
5. Graph Cluster Alignment (G-X-01)
**Problem:** Some node kinds are in wrong clusters (ProviderTranslation in "extensions", TransportRuntime in "extensions").
**Proposed:** Move graph YAML files to align node kinds with their architectural layer clusters:
- ProviderTranslation → compute cluster
- TransportRuntime → compute cluster
- AdapterModel → capabilities-and-models cluster
**Effort:** Small. YAML file moves + index rebuild.
P2: Package Metadata
Low-effort improvements that help navigation.
| Item | Gap ID | Effort | Description |
|---|---|---|---|
| Add layer tags to package.json | G-X-02 | Small | Add "atlas": { "layers": ["L13"] } to each package.json |
| Package README layer references | G-X-02 | Small | Each README should state which layer(s) the package implements |
| Graph node kind references in exports | G-X-02 | Small | TypeScript exports should reference their graph node kinds in JSDoc |
P3: Future Architecture (Deferred from v6.0)
These are explicitly deferred. Tracked here for completeness.
| Item | Gap IDs | Layer | Notes |
|---|---|---|---|
| Durable agent memory | G-L12-01, G-L12-02 | L12 | Requires vector store, embedding models, retrieval pipeline |
| Filesystem/network sandbox | G-L9-01, G-L9-02, G-L9-03 | L9 | Requires runtime policy enforcement engine |
| Model selection engine | G-L1-01 | L1 | Requires capability negotiation protocol |
| Team agent coordination | G-L6-01 | L6 | Beyond subagent dispatch — shared context, handoff |
| Marketplace identity/billing | G-L6-02 | L6 | Enterprise feature |
| NIST/OWASP compliance mapping | G-L14-02 | L14 | Enterprise governance |
Decision Criteria
Work items should be prioritized based on:
1. **User impact:** Does this gap cause visible failures or confusion? 2. **Agent impact:** Does this gap prevent autonomous agent workflows? 3. **Graph alignment:** Does fixing this bring packages closer to the graph source of truth? 4. **Incremental value:** Can this be shipped independently without blocking other work?
P0 items ship immediately. P1 items are v6.1 scope. P2 items are opportunistic. P3 items are tracked but not planned.