Orchestrata Ultimate › Plans › delegation-plan-s238
Backlog delegation plan — Claude ↔ kimi split (session 238)
Source: plans/delegation-plan-s238.md at 8a2cb42ad · last changed 2026-08-13 · first committed 2026-07-15 · 46 lines.
Prose pending.
Plan text
Backlog delegation plan — Claude ↔ kimi split (session 238)
Directive (Mohamed, IRC 2026-07-14): after #125, "create a plan and plan to delegate some tasks to kimi if that works to offload and speed up the process." Status: APPROVED by Mohamed on #agents 2026-07-15 ("claude approved"). K1 briefed. (Separate from
kimi-delegation-plan.md, the older SPEC-018 Fireflies brief.)
Split criteria
- Kimi gets items that are well-scoped, spec-anchored, and mechanically verifiable (a green test run or a zero-diff check proves completion) — the same profile as its successful generated-test regen and wiki work.
- Claude (me) keeps items needing design decisions, cross-subsystem integration, or live-KG/infra mutation, and reviews everything kimi lands (review-before-commit rule).
Ground rules for every kimi item: no mock/skip/xdist/monkeypatch, uv only, absolute imports, pathspec-scoped commits, no AI attribution, review by me before each lands.
Kimi track (parallel, independent items)
| # | Item | Scope anchor | Done-when |
|---|---|---|---|
| K1 | Backlog #217 — diagnose + fix the 5 pre-existing vizint full-suite failures (claude-chat RPC timeouts, project-explorer seeded mismatch, volk-status health flags) | vizint-ui test suite | vizint full suite green on Firefox |
| K2 | Backlog #47 — diagram_to_code round-trip test suite covering all 7 relationship arrows | orchestrata_tools.diagram | new test module green; test-review.py clean |
| K3 | Backlog #148 — annotate vizint-ui JS gadgets (specs 017 + 022, ~16 REQs) with the // @traces REQ-... comment convention (AGENTS.md § TS-side @traces) and confirm generate_rtm_evidence.py --ts ingests them |
AGENTS.md convention, commit cff3e01e | spec-017/022 rows move UNLINKED→DECLARED in reports/rtm-evidence.md |
| K4 | (Already owed) Memgraph wiki graph indexer retry — cluster is quiet now | knowledge-wiki | index run completes; chunk counts reported to #agents |
Claude track (sequential, design-heavy)
| # | Item | Why not delegated |
|---|---|---|
| C1 | Backlog #215 — kernel.events wiring: emit_event on real lifecycle transitions in cli_main.py boot + gateway _consume_kernel_events forwarder |
cross-subsystem (kernel boot ↔ gateway ↔ vizint NotificationCenter); design of event taxonomy |
| C2 | Backlog #223 — e2e_run relay: replace fragile canned --answer scripts |
needs a design choice (LLM-driven clarification answering vs. richer matcher) |
| C3 | Backlog #224/#231 — canonicalize ArtifactStore implementations (SPEC-107 FR-006 persistent ranked store) | architecture decision; blocked on the open SPEC-107 keep-or-rollback call |
| C4 | Review gate for K1–K4 | standing rule |
Sequencing
- Mohamed signs off (or edits the split) on #agents.
- K1–K3 briefs go to kimi one at a time (K4 is already agreed); each lands only after my review.
- C1 starts immediately after sign-off; C3 waits on the SPEC-107 decision.
Open user decisions this plan does NOT cover
SPEC-107 keep-or-rollback, gen-skeleton default-on, codewhale tabling plan adoption, diagram_to_code retirement, generated-file length-rule exemption, offsite neo4j-admin dump.