kandev/docs/plans/plugins/host-data-api/plan.md

spec, adr, created, status

spec	adr	created	status
docs/specs/plugins/spec.md	docs/decisions/0043-plugin-host-data-api.md	2026-07-17	implemented

Implementation Plan: Plugin Host data API

Overview

Give plugins a typed, capability-gated path to read (phase 1) kandev domain data over the existing kandev.plugin.v1 Host gRPC channel, replacing direct SQLite access. Land the contract first (proto + regen) and the one net-new aggregation (per-session LOC), then the SDK's Go-native accessors, then kandev's server-side implementation, and finally the proof (rewrite the agent-stats plugin) plus tests and docs. Write RPCs (CreateTask/UpdateTask/CreateComment) are specified in the proto but deferred; handlers return Unimplemented this phase.

The draft contract at docs/plans/plugins/HOST-DATA-API.proto is largely settled — refine field-level details against the real DTOs (apps/backend/pkg/api/v1/), do not redesign.

---

Backend

Proto contract (apps/backend/proto/kandev/plugin/v1/plugin.proto)

Fold the read + (deferred) write RPCs and DTOs from docs/plans/plugins/HOST-DATA-API.proto into the frozen plugin proto. Decide placement: add the RPCs to the existing service Host (simplest — reuses the single Host broker connection and the existing capability interceptor) rather than a separate service HostData, unless a separate service is needed to keep the interceptor's per-RPC capability lookup clean. Regenerate stubs with make -C apps/backend proto. Read RPCs: ListTasks, GetTask, ListWorkspaces, ListWorkflows, ListWorkflowSteps, ListAgentProfiles, ListRepositories, ListSessions, ListSessionCodeStats. DTOs mirror apps/backend/pkg/api/v1/{task,workflow,agent,workspace}.go field-for-field with RFC3339 string timestamps and optional nullables.

Per-session LOC aggregation (internal/analytics/repository/sqlite/)

No existing method computes per-session committed + peak-pending LOC. internal/analytics/repository/sqlite/stats.go aggregates git stats at workspace and per-repository granularity only. Add a read-only aggregation returning, per session id: SUM(insertions) / SUM(deletions) from task_session_commits, and the peak pending-diff (MAX over each snapshot's SUM(json_each(files).additions / .deletions)) from task_session_git_snapshots. This mirrors the SQL the agent-stats plugin runs today (stats.go:sessionsQuery). Use the dialect helpers for SQLite/Postgres portability; filter by session-id set / workspace and support pagination bounds. Expose it through the analytics (or task) service so the Host handler calls a service method, not the repository.

Host data RPC implementation (internal/plugins/)

Extend pluginHost (or add a sibling under internal/plugins/) to implement the new Host data interface methods. Each read handler:

1. checks manifest.CanRead("<resource>") — reuse the existing per-plugin binding and permissionDenied("api_read:<resource>") helper shape;
2. calls the relevant service method (taskSvc for tasks/workspaces/repositories/ sessions, workflow service for workflows/steps, agent settings for profiles, the new aggregation for code stats) — never a repository directly;
3. maps internal models → the Go-native DTOs with explicit code. Wire the new service dependencies into plugins.Service / plugins.NewService / Provide, and thread them into hostForPlugin. Write handlers return codes.Unimplemented.

---

Frontend

None. This feature has no SPA surface; it is a plugin-facing gRPC contract.

---

SDK (apps/backend/pkg/pluginsdk/)

Extend the Go-native Host surface authors and kandev share:

• Add Go-native DTO structs (Task, Session, SessionCodeStats, Workspace, Workflow, WorkflowStep, AgentProfile, Repository, page types) mirroring the proto, with toProto/fromProto helpers alongside the existing ones in types.go.
• Add data methods to the Host interface (or a HostData sub-interface exposed via host.Tasks(), host.Sessions(), ...), with grpcHostClient (plugin-side) and grpcHostServer (kandev-side) conversions in host.go.
• Author-facing ergonomics: typed accessors returning Go-native structs, e.g. host.Tasks().List(ctx, filter, page), host.Sessions().List(...), host.Sessions().CodeStats(...).

---

Tests

• DTO mapping (internal/plugins/*_test.go): table-driven — internal model → DTO for each resource, including nullable/optional and timestamp formatting.
• Capability gating (internal/plugins/host_test.go): each read RPC returns PermissionDenied with capability 'api_read:<resource>' not declared when the manifest omits the resource, and succeeds when it declares it. Write RPC returns Unimplemented.
• Per-session LOC aggregation (internal/analytics/repository/sqlite/*_test.go): real SQLite fixture with commits + snapshots; assert committed sums and the peak-pending MAX semantics (peak, not latest; not double-counted).
• SDK conversions (pkg/pluginsdk/*_test.go): proto↔native round-trip for the new DTOs.
• Integration: a plugin (via the SDK Host client over the broker) calls ListSessions + ListSessionCodeStats and gets rows without DB access — extend the existing internal/plugins/host_test.go / integration harness.

---

E2E Tests

None. No user-visible UI surface. The integration test above covers the plugin↔host path.

---

Implementation Waves

Wave 1 (parallel):
- [x] [task-01-proto-contract](task-01-proto-contract.md)
- [x] [task-02-session-loc-aggregation](task-02-session-loc-aggregation.md)

Wave 2:
- [x] [task-03-sdk-data-accessors](task-03-sdk-data-accessors.md)

Wave 3:
- [x] [task-04-host-data-impl](task-04-host-data-impl.md)

Wave 4 (parallel):
- [x] [task-05-rewrite-agent-stats-plugin](task-05-rewrite-agent-stats-plugin.md)
- [x] [task-06-tests](task-06-tests.md)
- [x] [task-07-docs](task-07-docs.md)

Rationale: task-01 (contract) and task-02 (aggregation) touch disjoint packages and can run in parallel. The SDK (task-03) needs generated stubs from task-01. The kandev impl (task-04) needs the stubs (task-01), the aggregation (task-02), and the Go-native interface/DTOs (task-03). Wave 4 tasks touch disjoint trees (a separate plugin repo, backend tests, public docs) and parallelize.

---

Open Questions

• Proto placement: RPCs on the existing service Host vs a new service HostData served on the same broker connection. Recommendation: extend service Host to reuse the existing capability interceptor and single broker wiring; revisit only if the interceptor's per-RPC capability mapping gets unwieldy. Settle in task-01.
• Aggregation home: internal/analytics (where git stats already live) vs a new method on the task repository's git_snapshots.go. Recommendation: analytics repository, consistent with GetGitStats/GetRepositoryStats. Settle in task-02.