docs(status): decisions + observatory taosctl groups merged

docs/STATUS.md

@@ -1,4 +1,5 @@
 SINGLE SOURCE OF TRUTH for cross-agent handoff.
+Last updated: 2026-06-23 ~18:10Z, @taOS-dev (DECISIONS + OBSERVATORY taosctl groups landed -- both approved-spec backends now have an agent-native CLI surface. MERGED since last entry (both on dev): #1413 taosctl decisions (post/list/get/history/answer over the shipped Decisions inbox API #1331/#1339/#1336; status+project filters, positive-int limit, json_array argtype validating --options-json at parse, arrays-only --value coercion that keeps bracketed free-text literal), #1414 taosctl observatory (fleet/pause-status/pause/resume/throttle-status/throttle over the queue-control API #1341/#1343/#1346/#1347/#1365; global or per-lane scope, --max/--clear concurrency dial). Folded all bot findings before merge: gitar Edge-Case (--options-json traceback), kilo regression (bracketed free-text hard-error), kilo footgun (phantom 'expired' status dropped), gitar Quality (absolute argtypes import aligned across both + the sibling modules), kilo per-lane throttle-clear test gap. Both gated by tests/test_taosctl_route_coverage.py. taosctl now covers the two human-in-the-loop ops surfaces from the shell. REMAINING on Decisions/Observatory = frontend app surfaces (live session). DISCIPLINE NOTE: held at 2 PRs in flight + GitHub-native auto-merge rather than a third group, to avoid out-producing review (the exact failure the Observatory spec was written to fix). Frontier otherwise unchanged: canvas slices 3/5/6 live-gated, grant-on-install consent flow Jay-gated (pending-decisions 28/29). five_hour 79% / seven_day 67%. PRIOR ENTRY BELOW.)
 Last updated: 2026-06-23 ~15:38Z, @taOS-dev (CANVAS EXCALIDRAW MIGRATION -- headless slices all merged. MERGED since last entry (all on dev): #1410 (removed the superseded Konva render layer KonvaBoard.tsx + element-to-konva.ts + konva/react-konva deps -- dead code, never wired into CanvasView, replaced by the Excalidraw view), #1411 (slice 4: mermaid/flowchart CanvasElements now render REAL diagrams via @excalidraw/mermaid-to-excalidraw -> mermaid-to-elements.ts converter + an async effect in ExcalidrawBoard, placeholder rectangle fallback while converting/on-error; folded all bot findings incl a CodeRabbit Major z-order fix -- scene now walks elements in z_index order so a diagram is not forced above a higher-z regular element). CANVAS STATE on dev: slice 1 mapping (element-to-excalidraw) + slice 2 read-only ExcalidrawBoard + slice 4 mermaid render + Konva removal ALL DONE; the read-only board + mermaid were screenshot-verified via a local vite preview + playwright (not the live app). REMAINING canvas slices ALL need the running app + backend (live session, not a static screenshot): slice 3 write-back interactions (Excalidraw onChange -> diff -> CanvasElement PATCH/POST), slice 5 swap CanvasView's tldraw CanvasBoard for ExcalidrawBoard behind a flag (store/SSE-wired, verify both themes + mobile), slice 6 remove tldraw + @tldraw/* deps. tldraw stays the LIVE renderer until slice 5. PENDING-ON-JAY: grant-on-install consent flow design (pending-decisions items 28 + 29: decision-shape/when/answer-dispatch, AND reconcile the TWO consent surfaces -- existing set_permissions UI vs new Decisions flow, recommend Decisions-inbox canonical, no schema change). The 3 live-MC decisions (25 /v1/chat opencode-host turn, 26 canvas-canonical confirmed, 27 Design Studio shell) remain confirmed-direction + live-session. five_hour 50% / seven_day 64%. PRIOR ENTRY BELOW.)
 Last updated: 2026-06-23 ~14:38Z, @taOS-dev (APP-PERMS RECONCILIATION COMPLETE + CANVAS EXCALIDRAW MIGRATION STARTED, in a live Jay session. Jay answered 4 gated decisions live (pending-decisions batch 7): 24 app-perms = broker stays the runtime enforcer + app_grants ledger FEEDS it; 25 /v1/chat turn = reuse the opencode host-server path (live-session build); 26 canvas = CanvasElement canonical + Excalidraw as the VIEW (bidirectional map); 27 Design Studio shell = proceed + restyle, DROP Plugins/Automations/Integrations (Automation Studio owns automations, plugins/integrations are OS-managed). MERGED since last entry: #1404 (broker route unions the per-user app_grants ledger into the gated `granted` set for community apps, additive/best-effort, first-party unchanged), #1405 (grant API rejects unknown/typo'd capabilities via is_known_capability now that the vocabulary is settled). IN FLIGHT (auto-merge armed): #1407 feat/canvas-excalidraw-slice2 -- slices 1-2 of #75 tldraw->Excalidraw: element-to-excalidraw.ts pure CanvasElement->Excalidraw-skeleton mapping (mirrors element-to-konva) + read-only ExcalidrawBoard.tsx (@excalidraw/excalidraw MIT, viewMode+zenMode, scrollToContent fit-on-mount), LIVE-VERIFIED via a throwaway vite preview + playwright screenshot (notes/text/link/mermaid render); #1408 folds 3 gitar Edge-Cases (broker ledger lookup now try/except best-effort so an uninitialised/erroring store no longer 500s; grant API rejects malformed network: origins via a centralised NET_ORIGIN_RE + is_valid_network_grant in capabilities.py reused by the package parser; mindmap_edge omits empty arrow bindings). #56 app-perms now has: vocabulary+manifest validation (#1389), grant store+API (#1369/#1374), broker<->ledger (#1404), grant validation (#1405). PENDING-ON-JAY: grant-on-install consent flow design (pending-decisions item 28: decision-shape/when/answer-side-effect). CANVAS remaining slices: 3 write-back interactions, 4 real mermaid/flowchart + mindmap arrows, 5 swap behind a flag (live verify both themes/mobile), 6 remove tldraw + superseded Konva layer + deps. The 3 other live-session decisions (25/26/27) are confirmed-direction, ready when Jay is around. five_hour 24% / seven_day 61%. PRIOR ENTRY BELOW.)
 Last updated: 2026-06-23 ~13:10Z, @taOS-dev (TAOSCTL AGENT-NATIVE SWEEP COMPLETE + PAGINATION-ARG HARDENING. MERGED since last entry (all on dev): #1393 knowledge, #1394 catalog, #1395 office, #1396 recycle, #1397 benchmarks, #1399 dashboard, #1400 agent-registry -- seven new taosctl command groups wrapping their respective routes/*.py read/write surfaces so agents drive them from the shell (pattern: NOUN + register() + handlers returning data; each gated by tests/test_taosctl_route_coverage.py which ast-checks every client.<method>(path) maps to a real registered route+method). #1398 fix-forwarded a gitar Bug on the knowledge group (ingest sent None for str-typed IngestRequest fields -> 422; delete-rule rule_id now int). Folded must-fix findings before merge: #1399 CodeRabbit non-positive --limit (now rejected at parse), #1400 gitar Edge-Case empty-update PATCH (now exits 2 with a hint). IN FLIGHT: #1401 feat/taosctl-arg-validation (auto-merge armed, waiting on slow GitHub-hosted test runners) -- generalized the --limit validator into a shared tinyagentos/cli/taosctl/argtypes.py (positive_int/nonneg_int) applied across 10 merged groups so non-positive --limit / negative --offset fail at parse time. PROCESS: switched the in-flight PRs from session-bound bash watchers to GitHub-native auto-merge (--auto --squash), which lands them when the 4 required checks (test 3.12/3.13, lint, spa-build) go green regardless of session. FOLLOW-UP (blocked on #1401 landing): dedupe dashboard.py's inline _positive_int to import the shared argtypes helper. UNBLOCKED HEADLESS QUEUE IS DRAINED: decision-17 memory-framework-manager backend turns out already-built (memory_management.py: settings-schema, recipes list/get/apply/recommend, backend capabilities, per-agent memory-config) so its remainder is frontend; AgentModelKeyStore is well-covered; remaining feature work is Jay-gated (Design Studio shell, app-perms broker-vs-app_grants reconciliation, canvas item 22, /v1/chat turn seam) or live-session-gated. five_hour 51% / seven_day 59%. PRIOR ENTRY BELOW.)