loki-mode 11.3.0 → 11.3.1

package/docs/SYNERGY-TASKS.md

@@ -1,143 +0,0 @@
-# Loki Mode Synergy Implementation Tasks
-
-This document tracks all implementation tasks to achieve tool synergy and competitive leadership.
-
----
-
-## Phase 1: Event Bus Foundation (v5.17.0) - COMPLETED
-
-| Task ID | Task | Status | Notes |
-|---------|------|--------|-------|
-| SYN-001 | Create Python event bus module | Done | events/bus.py |
-| SYN-002 | Create TypeScript event bus module | Done | events/bus.ts |
-| SYN-003 | Create Bash event emitter helper | Done | events/emit.sh |
-| SYN-004 | Add event bus tests | Done | 10/10 passing |
-| SYN-005 | Document event bus architecture | Done | SYNERGY-ROADMAP.md |
-
----
-
-## Phase 2: Memory Integration (v5.18.0) - COMPLETED
-
-| Task ID | Task | Status | Priority | Notes |
-|---------|------|--------|----------|-------|
-| SYN-006 | Create unified memory access layer | Done | High | memory/unified_access.py |
-| SYN-007 | Add memory context to VS Code sidebar | Done | High | memoryViewProvider.ts |
-| SYN-008 | Add memory retrieval to CLI start | Done | Medium | autonomy/loki - base64 encoding, atomic writes |
-| SYN-009 | Add automatic context loading to MCP | Done | High | MCP uses unified access |
-| SYN-010 | Add memory-informed suggestions to API | Done | Medium | /api/status has memory context |
-| SYN-011 | Integrate VS Code file edits with memory | Pending | Medium | Record edits as episodes |
-
----
-
-## Phase 3: Smart State Sync (v5.19.0) - IN PROGRESS
-
-| Task ID | Task | Status | Priority | Notes |
-|---------|------|--------|----------|-------|
-| SYN-012 | Create centralized state manager | Done | High | state/manager.py, state/manager.ts - peer reviewed |
-| SYN-013 | Replace direct file reads with state manager | Pending | High | All components use manager |
-| SYN-014 | Add optimistic updates with conflict resolution | Pending | Medium | Handle concurrent writes |
-| SYN-015 | Add state versioning for rollback | Pending | Low | Recovery capability |
-| SYN-016 | Add state change notifications | Pending | High | Real-time sync |
-
----
-
-## Phase 4: Cross-Tool Learning (v5.20.0) - IN PROGRESS
-
-| Task ID | Task | Status | Priority | Notes |
-|---------|------|--------|----------|-------|
-| SYN-017 | Define learning signal types | Done | High | learning/signals.py + learning/signals.ts - peer reviewed |
-| SYN-018 | Implement learning collectors in CLI | Pending | Medium | CLI emits learning signals |
-| SYN-019 | Implement learning collectors in API | Pending | Medium | API emits learning signals |
-| SYN-020 | Implement learning collectors in VS Code | Pending | Medium | VS Code emits learning signals |
-| SYN-021 | Implement learning collectors in MCP | Pending | Medium | MCP emits learning signals |
-| SYN-022 | Create learning aggregator | Pending | High | Consolidate signals |
-| SYN-023 | Add learning-based suggestions | Pending | High | Use learnings to improve |
-| SYN-024 | Add learning metrics dashboard | Pending | Low | Visualize learning |
-
----
-
-## Phase 5: Unified Dashboard (v5.21.0) - IN PROGRESS
-
-| Task ID | Task | Status | Priority | Notes |
-|---------|------|--------|----------|-------|
-| SYN-025 | Extract dashboard as reusable web components | Done | High | legacy-ui/ - ARIA, keyboard nav, focus mgmt - peer reviewed |
-| SYN-026 | Create VS Code webview integration | Pending | High | Embed dashboard in VS Code |
-| SYN-027 | Unify styling and behavior | Pending | Medium | Consistent UX |
-| SYN-028 | Add dashboard feature parity check | Pending | Low | Ensure all features available |
-
----
-
-## Competitive Gap Tasks (vs claude-flow, claude-mem, etc.)
-
-| Task ID | Task | Status | Priority | Competitor Gap |
-|---------|------|--------|----------|----------------|
-| COMP-001 | Implement swarm intelligence patterns | Pending | Medium | claude-flow has 60+ agents |
-| COMP-002 | Add Byzantine fault tolerance | Pending | Low | claude-flow resilience |
-| COMP-003 | Improve embedding quality | Pending | Medium | claude-mem uses OpenAI |
-| COMP-004 | Add importance scoring for memories | Done | High | Now has decay + boost |
-| COMP-005 | Implement memory namespaces | Pending | Medium | claude-mem project isolation |
-| COMP-006 | Add real-time collaboration | Pending | Low | Multi-user support |
-| COMP-007 | Improve context window optimization | Done | High | Progressive disclosure + budget |
-
----
-
-## Security Tasks (from Peer Review)
-
-| Task ID | Task | Status | Priority | Notes |
-|---------|------|--------|----------|-------|
-| SEC-001 | Fix command injection in session-init.sh | Done | Critical | Fixed via env vars |
-| SEC-002 | Fix command injection in store-episode.sh | Done | Critical | Fixed via env vars |
-| SEC-003 | Fix path traversal in mcp/server.py | Done | High | Added validate_path() |
-| SEC-004 | Improve fork bomb pattern detection | Done | Medium | Regex updated in v5.16.0 |
-| SEC-005 | Add input sanitization to all hooks | Done | High | Via path validation |
-| SEC-006 | Fix path traversal in storage.py | Done | High | Path validation in v5.25.0 |
-| SEC-007 | Fix XSS in log-stream | Done | High | Output sanitization in v5.25.0 |
-| SEC-008 | Fix memory leak in session-control | Done | Medium | Resource cleanup in v5.25.0 |
-| SEC-009 | Fix Python injection in completion-council.sh | Done | Critical | Input sanitization in v5.25.0 |
-| SEC-010 | Make CORS configurable | Done | Medium | CORS_ALLOWED_ORIGINS env var in v5.25.0 |
-
----
-
-## Quick Wins (Can Do Now)
-
-| Task ID | Task | Status | Priority | Effort |
-|---------|------|--------|----------|--------|
-| QW-001 | CLI emits events on command execution | Done | High | Low |
-| QW-002 | API returns relevant patterns with status | Done | Medium | Low |
-| QW-003 | MCP emits tool call events | Done | Medium | Low |
-| QW-004 | Add suggestions endpoint to API | Done | Medium | Medium | api/routes/memory.ts - rate limiting, security fixes - peer reviewed |
-| QW-005 | VS Code shows memory stats in status bar | Done | Low | Low |
-
----
-
-## Task Dependencies
-
-```
-SYN-006 (unified memory) -> SYN-007, SYN-008, SYN-009, SYN-010, SYN-011
-SYN-012 (state manager) -> SYN-013, SYN-014, SYN-015, SYN-016
-SYN-017 (learning signals) -> SYN-018, SYN-019, SYN-020, SYN-021 -> SYN-022 -> SYN-023
-SYN-025 (web components) -> SYN-026, SYN-027
-
-SEC-001, SEC-002 -> Should be done ASAP (Critical)
-```
-
----
-
-## Progress Tracking
-
-| Phase | Total Tasks | Completed | In Progress | Pending |
-|-------|-------------|-----------|-------------|---------|
-| Phase 1 (Event Bus) | 5 | 5 | 0 | 0 |
-| Phase 2 (Memory) | 6 | 5 | 0 | 1 |
-| Phase 3 (State) | 5 | 1 | 0 | 4 |
-| Phase 4 (Learning) | 8 | 1 | 0 | 7 |
-| Phase 5 (Dashboard) | 4 | 1 | 0 | 3 |
-| Competitive | 7 | 2 | 0 | 5 |
-| Security | 10 | 10 | 0 | 0 |
-| Quick Wins | 5 | 5 | 0 | 0 |
-| **Total** | **50** | **30** | **0** | **20** |
-
----
-
-*Last Updated: 2026-02-06*
-*v5.25.0: 50 tasks tracked, 30 completed. Security hardening (5 new fixes) added in v5.25.0.*

package/docs/TASKLIST-ACCURACY-PLAN.md

@@ -1,355 +0,0 @@
-# Dashboard Task-List Accuracy Fix Plan
-
-Branch: fix/tasklist-accuracy
-Worktree: /Users/lokesh/git/loki-tasklist-wt
-Type: PATCH release (bug fix, no new features)
-Status: Plan only. No implementation in this document.
-
-## 1. Summary
-
-On a live run (~/git/anonima, loki start reusing a generated PRD) the dashboard
-board at :57375/#section=overview shows:
-
-- DONE column: 30 bare "Iteration N" cards with empty body (click -> empty modal).
-- PENDING column: 12 rich PRD stories (prd-001..prd-012) that never move.
-- IN-PROGRESS: iteration-13, which correctly borrows the active PRD story
-  title/description (intentional).
-
-This plan verifies the mechanism against source, recommends an honesty-first
-board model, lists the exact per-file change set with line references, gives a
-test plan, and notes the patch-release steps.
-
-## 2. Verified root cause (against source + real data)
-
-Reproduced by replaying the /api/tasks merge logic over the real
-~/git/anonima/.loki files. Rendered output shape:
-
-- TOTAL 43 tasks: done=30, pending=12, in_progress=1.
-- EMPTY-description done cards: 30 of 30.
-- Duplicate ids in the output: iteration-1 x5, iteration-2 x3,
-  iteration-3..12 x2 each, iteration-13 x2 (appears in BOTH in_progress and
-  done). This is a distinct, second defect (see 2C).
-
-There are four concrete defects.
-
-### 2A. Thin completed markers (primary defect -> empty done cards)
-
-`track_iteration_start` (autonomy/run.sh:5370-5569) writes a RICH in-progress
-record: it reads pending[0] for context (5395-5416) and emits a task with
-id `iteration-N`, type `iteration`, a real title/description, acceptance_criteria,
-notes, and a `logs` array (5462-5488, with fallbacks at 5499-5532 and
-5545-5568). `append_iteration_log` (5013-5072) appends further per-phase log
-entries to that same in-progress record.
-
-`track_iteration_complete` (autonomy/run.sh:6276-6455) writes the COMPLETED
-marker with a hardcoded THIN body (6400-6411):
-
-    {
-      "id": "iteration-N",
-      "type": "iteration",
-      "title": "Iteration N",
-      "status": "completed"|"failed",
-      "completedAt": "...",
-      "exitCode": N,
-      "provider": "..."
-    }
-
-No title-from-context, no description, no logs. It appends this to
-completed.json/failed.json (6414-6429), then DELETES the rich in-progress
-record (6431-6443). So the rich in-progress card becomes an empty done card,
-and the logs that lived only in the in-progress record are discarded.
-
-Confirmed data: ~/git/anonima/.loki/queue/completed.json items are all
-`{"id":"iteration-N","title":"Iteration N","description":""...}`.
-
-Key consequence for the fix: the body (logs/phase) exists ONLY in the
-in-progress record at completion time. The API cannot recover it after run.sh
-deletes it. run.sh MUST lift it before removal. This is why run.sh is the
-required fix site for the card body, not the API.
-
-### 2B. PRD stories never reconcile pending -> done
-
-PRD tasks are populated once into pending.json by the PRD parser
-(autonomy/run.sh:15167-15213, ids `prd-001`..). Nothing ever pops or moves a
-prd task out of pending: grep of every pending.json write site
-(2625-2720 dedup, 2918-2938 GitHub sync read-only, 14663 openspec purge,
-14679 populate) shows no path that transitions a prd story to completed. The
-RARV loop only READS pending[0] for iteration context (5403-5413); it never
-mutates prd status. So the 12 stories sit in pending forever regardless of how
-many iterations run.
-
-Honesty check (decisive): ~/git/anonima/.loki/checklist/checklist.json has 19
-items, ALL status `pending`; verification-results.json summary is
-verified=0/failing=0/pending=19. Nothing in this run is verified-complete.
-Auto-moving any prd story to done would be fake-green.
-
-### 2C. No intra-source dedup in /api/tasks (duplicate cards, broken keys)
-
-dashboard/server.py /api/tasks (1772-1914) merges two sources:
-1. dashboard-state.json `.tasks` groups (1783-1825) mapped
-   pending->pending, inProgress->in_progress, completed->done, failed->done.
-2. queue/*.json files (1827-1881).
-
-The queue merge skips ids already present (1854 `any(t["id"] == tid ...)`), but
-there is NO dedup WITHIN the dashboard-state groups themselves, and no
-cross-column dedup. The real dashboard-state completed group already contains
-iteration-1 x5 and iteration-13 (which is ALSO in the inProgress group). Result:
-the same id renders as multiple cards, and iteration-13 appears in both the
-in-progress and done columns simultaneously. In the React board these become
-duplicate `key={task.id}` / dnd-kit sortable ids (TaskCard.tsx:24), which is a
-correctness bug (duplicate keys, unstable drag).
-
-Root of the id collision: iteration ids are `iteration-N` and N resets/repeats
-across sub-runs, and completed.json is capped with `data[-50:]` (run.sh:6426)
-which retains stale duplicate-id entries rather than replacing by id.
-
-### 2D. iteration-13 leaks into completed while still in-progress
-
-The completed group contains iteration-13 even though it is the active
-in-progress card. Combined with 2C this is what puts one id in two columns.
-The dedup fix (3C) resolves the display; the run.sh completed-writer dedup (3B)
-resolves the source accumulation.
-
-## 3. Recommended board model (honesty-first)
-
-Recommendation: Option A now; defer Option B as a scoped follow-up. Do NOT ship
-Option B (or C) in this patch.
-
-Rationale (honesty + accuracy):
-
-- Iterations that ran are real events. Making their done cards honest and
-  informative is truthful and low-risk.
-- LITERAL Option A is a trap and must be avoided: `track_iteration_start`
-  borrows pending[0] for the title, and because prd stories never leave pending
-  (2B), pending[0] is ALWAYS prd-001. Carrying that borrowed PRD title forward
-  onto every done card would render 13+ done cards all titled
-  "server.js (single backend module)", implying server.js was built and
-  completed 13 times, when the checklist says it was never verified. That is the
-  same fake-green we reject for Option B, spread across the done column. So the
-  corrected Option A carries forward the REAL, iteration-scoped parts (logs,
-  phase, exit code, duration) and keeps an iteration-scoped title
-  ("Iteration N complete - <phase>, exit 0"), NOT the borrowed PRD story title.
-- Option B (reconcile prd pending -> done) requires a trustworthy per-story
-  completion signal. The only per-item truth signal is the checklist, keyed by
-  `be-001`/`api-001` item ids with NO stored `prd-NNN` link
-  (verified in checklist.json/verification-results.json). Mapping checklist ->
-  prd would require fuzzy title matching, which introduces inaccuracy
-  (mismatch -> fake-green or fake-red). Shipping a fuzzy reconciler in a patch
-  contradicts the high-accuracy mandate. On this run the correct verified signal
-  would move ZERO stories anyway (verified=0), so B delivers no value now and
-  carries real regression risk. Defer to a follow-up that either (a) stamps a
-  `prd-NNN` source id onto checklist items at generation time so the join is
-  exact, or (b) surfaces the checklist as its own honest progress panel instead
-  of mutating prd story status.
-
-Net board after this patch: DONE shows honest iteration cards (informative
-title + real logs, no empty modals); PENDING keeps the 12 prd stories
-(truthful - none are verified-done); IN-PROGRESS shows the single active
-iteration; no duplicate cards; no card in two columns.
-
-## 4. Exact change set (per file, with line refs)
-
-All engine (run.sh) edits are additive/corrective to the task-record shape only.
-No trust-gate, verdict, or completion-council logic is touched.
-
-### 4A. autonomy/run.sh - track_iteration_complete (6390-6444)
-
-Change the completed-marker construction so it lifts the honest body from the
-in-progress record BEFORE that record is removed.
-
-- Before building task_json (currently 6400-6411), read the matching
-  `iteration-N` entry from in-progress.json (the file already read at 6432-6443
-  for removal). Capture its `logs`, `title`, `description`, `acceptance_criteria`,
-  `notes`, `startedAt`, `user_story`, `source`, `project` if present.
-- Build the completed entry with:
-  - `title`: iteration-scoped and honest, e.g.
-    "Iteration N complete - <phase>" (exit 0) or
-    "Iteration N failed (exit <code>)" - derived from `$phase` (already computed
-    at 6355-6356) and `$exit_code`. Do NOT reuse the borrowed PRD story title
-    (see 3 rationale). If a truthful iteration description is wanted, synthesize
-    "Iteration N: <phase>, exit <code>, <duration>ms" from values already in
-    scope (`$phase`, `$exit_code`, `$duration_ms` at 6285).
-  - `logs`: the lifted logs array from the in-progress record (preserves the
-    real per-phase entries so the modal is populated).
-  - keep `type`, `status`, `completedAt`, `exitCode`, `provider` as today; add
-    `startedAt` (from in-progress) so the card can show duration.
-- Do this with the same python-embedded approach already used at 6417-6429;
-  read in-progress inside that python block so it is atomic with the append.
-- Dedup by id on write (see 4B) instead of blind append.
-
-Keep the existing lock/atomic-write and the last-resort fallback shape so a
-python-unavailable environment still writes a valid (if minimal) card.
-
-### 4B. autonomy/run.sh - completed.json/failed.json dedup on write (6417-6429)
-
-Replace the blind `data.append(...)` + `data[-50:]` (6424-6426) with a
-replace-by-id upsert: drop any existing entry with the same `id`, append the
-new one, then cap at 50. This stops the cross-sub-run iteration-N accumulation
-(iteration-1 x5) at the source. Failed.json (same writer, target chosen at
-6414-6415) gets the same upsert.
-
-Note: also prevents an id being simultaneously in completed and failed (real
-data has iteration-1 in both). Upsert-by-id in the chosen target plus removal
-from the other target (add a small removal of `id` from the non-target file)
-makes completed/failed mutually exclusive per id.
-
-### 4C. dashboard/server.py - /api/tasks global dedup (1780-1914)
-
-Add a single global dedup pass keyed by id with column priority so one id yields
-exactly one card, and it lands in the most-active column:
-
-- Priority order: in_progress > review > pending > done. (in_progress beats done
-  so iteration-13-in-both resolves to the in-progress card; pending beats done
-  is irrelevant here but keeps a story that is both queued and archived in the
-  live column.)
-- Implementation: keep a dict keyed by id while appending; when a duplicate id
-  arrives, keep the entry whose status has higher priority, and if equal
-  priority prefer the one with a non-empty description/logs (richer wins). Apply
-  this across BOTH the dashboard-state pass (1796-1823) and the queue pass
-  (1850-1879), replacing the current one-directional skip at 1854.
-- Return `list(dict.values())` (or rebuild all_tasks from the dict) at 1906
-  before the project_id/status filters, preserving current ordering by
-  first-seen where possible.
-
-Backward compatibility: the response is still a JSON array of the same
-task-entry shape (same keys as today at 1808-1822 and 1856-1878). No field is
-removed or renamed. Consumers see fewer/deduped items and populated
-description/logs on done cards. This is strictly a correctness improvement to an
-existing shape - compatible.
-
-### 4D. dashboard/frontend - defense-in-depth (optional, low risk)
-
-The React TaskCard already guards empty description (TaskCard.tsx:129-133) and
-renders title from `task.title` (124-126); the empty cards are a DATA problem
-fixed by 4A/4C, so no functional SPA change is required. Optional hardening:
-
-- TaskModal.tsx: when a task has neither description nor logs nor
-  acceptance_criteria, render an explicit honest placeholder
-  ("No details recorded for this iteration.") instead of an empty modal body.
-  This is a safety net for any legacy/thin records still on disk from before the
-  patch.
-- No change to api.ts transformTask (api.ts:41-56) or types.ts is needed; the
-  status enum already covers done/in_progress/pending.
-
-Prefer to include 4D TaskModal placeholder (small, defensive) but keep it out of
-scope if the patch must be minimal - 4A/4C alone fix the reported bug.
-
-## 5. Dedup correctness argument
-
-- Source (run.sh): 4B upsert-by-id makes completed.json / failed.json contain at
-  most one entry per iteration id, and mutual exclusion across the two files.
-- API (server.py): 4C global id-dedup with column priority guarantees each id
-  appears in exactly one column. iteration-13 (currently in inProgress state
-  group AND completed queue) collapses to the in_progress card because
-  in_progress outranks done. The 5x iteration-1 collapses to one done card.
-- Result on the anonima fixture: 43 -> ~26 unique cards (12 pending + 1
-  in_progress + ~13 unique iteration done cards), zero duplicate ids, zero
-  cross-column duplication, zero empty-body done cards.
-
-## 6. Test plan
-
-### 6A. pytest for /api/tasks (primary)
-
-Add a test module under the dashboard test suite (co-locate with existing
-server/API tests; confirm the exact dir during implementation, e.g.
-dashboard/tests/ or tests/). Build a fixture .loki dir mirroring anonima:
-
-- dashboard-state.json .tasks with: pending = 12 prd-NNN rich stories;
-  inProgress = [iteration-13 rich, borrowed prd title]; completed = duplicate
-  iteration-1..13 thin markers INCLUDING iteration-13 (to reproduce cross-column
-  leak); failed = [iteration-14 thin, iteration-1 thin].
-- queue/completed.json + queue/pending.json mirroring the same.
-
-Assertions:
-1. No done-column card has an empty title AND empty description AND empty logs
-   (no empty modal payloads). After the fix, done iteration cards carry logs.
-2. Every returned id is unique (len(ids) == len(set(ids))).
-3. iteration-13 appears exactly once, with status in_progress (column-priority
-   rule).
-4. All 12 prd-NNN remain in pending (no fake-green: none promoted to done).
-5. Counts: done has one card per unique completed/failed iteration id; no
-   duplicates.
-6. Backward-compat: response is a list; each entry still has id/title/
-   description/status/priority/type keys.
-
-Add a focused unit-style test for the dedup helper (column-priority + richer-
-wins) if it is factored into a function.
-
-### 6B. run.sh completed-writer test (shell/behavioral)
-
-If the shell test harness supports it (see scripts/local-ci.sh 14/14 dual-route
-suite), add a case that: seeds an in-progress.json with a rich iteration-N record
-(title/desc/logs), calls track_iteration_complete N 0, then asserts the resulting
-completed.json entry (a) has a non-empty logs array carried from in-progress,
-(b) has an honest iteration-scoped title (not "Iteration N" empty-body, not the
-borrowed PRD story title), and (c) contains no duplicate id after two calls with
-the same N (upsert). Keep it provider-agnostic and offline.
-
-### 6C. Playwright screenshot of the board
-
-Capture :57375/#section=overview against the fixture-backed dashboard (or the
-anonima state) after the fix: assert the DONE column cards show titles + preview
-text (not bare "Iteration N"), clicking a done card opens a populated modal, and
-no card ID is visually duplicated across columns. Store under
-artifacts/<release>-screens/ per CLAUDE.md:227.
-
-## 7. Disjoint build lanes
-
-Two independent lanes (can be done in parallel, no shared edit region):
-
-- Lane 1 (engine): autonomy/run.sh 6390-6444 (completed writer + upsert). Owner
-  must respect: additive/corrective to record shape only; no trust/verdict/
-  council logic; no version bump/commit; no emoji/em-dash.
-- Lane 2 (dashboard): dashboard/server.py /api/tasks 1780-1914 (global dedup) +
-  optional dashboard/frontend TaskModal.tsx placeholder.
-
-Lane 3 (tests, after 1+2): pytest fixture (6A), optional shell test (6B),
-Playwright shot (6C).
-
-Sequencing: Lane 1 and Lane 2 are independent and each independently improves
-the board; the API dedup (Lane 2) also protects against any not-yet-fixed thin
-records, so it can ship even if Lane 1 slips. Tests depend on both.
-
-## 8. Patch release steps (per CLAUDE.md)
-
-- Version: current 7.104.1 (VERSION, package.json, CLAUDE.md:306, SKILL.md
-  header+footer, plugins/loki-mode/.claude-plugin/plugin.json, wiki/Home.md,
-  wiki/_Sidebar.md, docs/INSTALLATION.md - grep "7.104.1" for the full set).
-  Bump PATCH -> 7.104.2 in ALL of these (CLAUDE.md:301 mandates header AND footer
-  for SKILL.md).
-- CHANGELOG.md: add a new "## [7.104.2] - <date>" section above 7.104.1 with a
-  fix entry describing honest iteration done cards + /api/tasks dedup + no
-  fake-green prd promotion. (Currently "## [Unreleased]" is "(none)".)
-- Pre-push gate (MANDATORY, CLAUDE.md:318-330): run
-  `bash scripts/local-ci.sh` on this Mac; do not push if it says DO NOT PUSH.
-- Commit protocol (CLAUDE.md:290-296): show git diff --stat, stage files
-  individually by name (never git add -A), STOP and wait for user approval
-  before commit. This plan does not commit.
-- Post-release distribution validation (CLAUDE.md:332-334): npm pack + Docker
-  pull smoke of the shipped version.
-
-## 9. Open questions
-
-1. Done-card title wording: "Iteration N complete - <phase>" vs
-   "Iteration N: <phase>, exit 0, <duration>ms". Pick the clearest; both avoid
-   the borrowed PRD title. Founder preference?
-2. Should failed iterations render in the done column (current mapping
-   failed->done) or a distinct failed column? Out of scope for this patch;
-   current behavior preserved. Note for follow-up.
-3. Follow-up (Option B) design: stamp `prd-NNN` source id onto checklist items
-   at generation so checklist->story is an exact join, versus surfacing the
-   checklist as a separate honest progress panel. Decide before building B.
-4. Exact dashboard test directory + whether a shell harness case for
-   track_iteration_complete is supported by scripts/local-ci.sh - confirm at
-   implementation time.
-5. Should the API dedup also collapse the stale duplicate ids already persisted
-   in old on-disk state (one-time migration), or is display-time dedup
-   sufficient? Recommend display-time only (no destructive migration in a patch).
-
-## 10. Critical files for implementation
-
-- /Users/lokesh/git/loki-tasklist-wt/autonomy/run.sh (track_iteration_complete 6276-6455; track_iteration_start 5370-5569; append_iteration_log 5013-5072; write_dashboard_state 5078-5118)
-- /Users/lokesh/git/loki-tasklist-wt/dashboard/server.py (/api/tasks 1772-1914)
-- /Users/lokesh/git/loki-tasklist-wt/dashboard/frontend/src/components/TaskModal.tsx (optional empty-body placeholder)
-- /Users/lokesh/git/loki-tasklist-wt/CHANGELOG.md (patch entry)
-- /Users/lokesh/git/loki-tasklist-wt/VERSION and package.json (patch bump; plus SKILL.md/CLAUDE.md/plugin.json version strings)

package/docs/TOP-100-BACKLOG.md

@@ -1,159 +0,0 @@
-# Loki + Autonomi: Top-100 Backlog (Master Plan)
-
-Generated 2026-07-01 by the top100-backlog-enumeration ultracode workflow (7 agents, 116 raw items -> 100 ranked). North star: easiest to integrate with anyone; ranked by leverage toward zero-friction adoption, modularization, enterprise-grade, and scale.
-
-## Themes
-- Trust and correctness bugs (data-loss, cross-tenant, fake-green safety)
-- Autonomi Verify productization (stub -> shipped: gate/council/cost/sign wiring)
-- Zero-friction adoption (reduce user setup and error surface)
-- Modularization (extract libraries, break monoliths, fix import boundaries)
-- Enterprise-scale and multi-tenant (RBAC/SSO/persistence/sandbox, mostly founder-gated)
-- autonomi-saas evidence and proof pipeline (receipt phases, correlation, worker cutover)
-- CI, test, and distribution hardening (coverage, unreachable tests, publish gates)
-- Runtime-agnostic and Bun migration (bash sunset, write-path port)
-
-## Legend
-- **gate**: autonomous (I execute) | founder-gated (prep to one-approval-away) | community
-- **effort**: S/M/L/XL | **value**: critical/high/medium/low
-
-## Full ranked list
-
-| # | value | eff | gate | area | item |
-|---|---|---|---|---|---|
-| 1 | critical | S | autonomous | engine | run.sh self-delete EXIT trap nukes canonical source on LOKI_RUNNING_FROM_TEMP |
-| 2 | critical | M | autonomous | saas | autonomi-saas proof correlation: cross-tenant leak + stuck-Building fix (per-build run_id) |
-| 3 | critical | XL | founder-gated | verify | Verify: verifyCompletion() stub throws not-implemented (wire gate->integrity->council->cost->sign pipeline) |
-| 4 | critical | M | founder-gated | verify | Verify: npm @autonomi/verify unpublishable (private:true, no exports/types/build, .ts bin, undeclared zod) |
-| 5 | critical | S | founder-gated | saas | Verify: BUSL-1.1 vs open licensing undecided (blocks hosted/commercial launch) |
-| 6 | critical | M | founder-gated | saas | Verify: hosted /verify REST API not deployed (Node entry + Dockerfile + serve script + persistent signing key) |
-| 7 | critical | M | autonomous | modularization | Verify: extract verification-core library (port council_evidence_gate to TS) |
-| 8 | critical | S | autonomous | verify | Verify: 5 deterministic gates parity with completion-council.sh (staged, verify + ship) |
-| 9 | critical | S | autonomous | saas | autonomi-saas anti-fake-green safety: verdict strictly from engine honesty.headline (verify + guard) |
-| 10 | critical | M | autonomous | saas | autonomi-saas VS-7 worker cutover: BFF enqueues, worker sole executor (verify live e2e) |
-| 11 | high | M | founder-gated | verify | Signed neutral records on SDK/MCP path (verifyCompletion returns UNSIGNED_SENTINEL; only REST signs) |
-| 12 | medium | S | autonomous | verify | Verify: fix embed import boundary (review/* drags LLM council into offline core; refresh regresses) |
-| 13 | high | S | founder-gated | verify | Verify: resolve naming collision (loki-vaas vs loki-verify bin vs loki verify subcommand) |
-| 14 | critical | L | founder-gated | engine | Sandbox backend implementation (createSandbox throws; choose gVisor/Firecracker/Kata) |
-| 15 | critical | XL | founder-gated | enterprise | Enterprise RBAC / SSO / API-key auth / metering (hosted + on-prem control plane) |
-| 16 | high | L | founder-gated | enterprise | On-prem persistence backend interface + first prod store (Postgres/SQLite) |
-| 17 | high | S | autonomous | engine | base_unreachable signal in ObservedDiff (suppress empty_diff false-block once sandbox lands) |
-| 18 | high | M | founder-gated | engine | Real Anthropic LLM provider wiring for production verify (council against live model) |
-| 19 | high | M | founder-gated | verify | Verify: LLM council wired into default verdict path (module exists, verifyCompletion stub does not invoke) |
-| 20 | high | S | autonomous | engine | Signed verification records + ed25519 audit trail (staged, verify + ship) |
-| 21 | high | M | autonomous | verify | Golden-output conformance corpus for byte-identical CLI-invariance across implementations |
-| 22 | high | M | autonomous | engine | Code review gate: run 3 parallel reviewers (currently single-reviewer path) |
-| 23 | high | S | founder-gated | saas | HTTP /verify endpoint production wiring (handler complete, README lists as planned) |
-| 24 | high | S | autonomous | saas | MCP verify tools (verify_completion, run_evidence_gate) wiring + distribution (staged) |
-| 25 | high | M | community | test | Memory failure-loop (FAILURE_MEMORY) persistence bug (strict-xfail; recency read incomplete) |
-| 26 | high | XL | autonomous | engine | Bun runtime Phase 2: migrate write-path commands (feat/bun-migration) |
-| 27 | high | M | autonomous | cli | Bash runtime sunset (loki run-shell removal, Phase 6, after clean soak) |
-| 28 | high | M | community | saas | autonomi-saas Evidence Receipt Phase 2: screenshot capture + BFF stream route (in progress) |
-| 29 | high | M | community | saas | autonomi-saas Evidence Receipt Phase 3: test results from engine workspace |
-| 30 | high | L | autonomous | saas | autonomi-saas Cost tab + budget-breaker UX (web + enforcement audit) |
-| 31 | high | S | community | saas | autonomi-saas BFF overlays stored saas_evidence onto engine proof (verify) |
-| 32 | medium | M | founder-gated | saas | E4 no-server run output capture for CLI/library builds (bounded inline + artifact) |
-| 33 | high | M | autonomous | dashboard | Prompt Optimizer: wire real LLM-as-judge (currently heuristic-only) |
-| 34 | high | M | community | engine | Cost-honesty check implementation (normalizeModel + quote==dispatched) |
-| 35 | high | S | community | saas | POST /cost-honesty API handler (blocked on cost core) |
-| 36 | critical | L | founder-gated | saas | Hosted metered runner with per-verification container sandbox |
-| 37 | high | M | community | ci | CI parity-matrix flake root-cause (state cooldown after test-cli-commands.sh) |
-| 38 | high | M | community | ci | Post-release smoke sign-off gate before npm/Docker go public |
-| 39 | high | M | community | test | Document the 3 pre-existing failing shell tests (no public known-failures list) |
-| 40 | medium | S | autonomous | dashboard | PR #178: dashboard memory summary reports 0 on JSON-backed projects |
-| 41 | high | M | community | test | ARM64 runtime verification (emulation-only in CI; needs real Apple Silicon run) |
-| 42 | high | XL | founder-gated | saas | Multi-tenant sandbox backend (gVisor/Firecracker/Kata) selection spike |
-| 43 | critical | L | community | test | Real provider end-to-end tests (cost + secrets + duration make CI-unreachable) |
-| 44 | high | XL | founder-gated | enterprise | Enterprise deploy: object-store sync + queue fleet modes + #691 --config loader |
-| 45 | high | M | founder-gated | enterprise | Compliance scanner (loki-verify) converged as module in trust-layer SKU |
-| 46 | high | L | founder-gated | saas | GitHub App verification integration (webhook wrapper, greenfield) |
-| 47 | high | L | community | test | Long-duration stress tests (leak/queue-churn) unreachable in CI |
-| 48 | medium | L | community | saas | autonomi-saas Evidence Receipt Phase 4: clip + full test-log streaming + build_artifacts table |
-| 49 | high | XL | community | test | SWE-bench Pro 119 resume (paused 35/119) |
-| 50 | medium | M | autonomous | engine | Multi-reviewer completion council option in hosted verify (council:true dropped today) |
-| 51 | medium | M | autonomous | engine | Integrity detectors (mock-only, tautological, test-weakening) wired into evidence block |
-| 52 | medium | S | community | test | Bun test coverage threshold enforcement on branches (70% baseline exists) |
-| 53 | medium | L | autonomous | dashboard | Dashboard mid-run steering / builder control (mid-flight model switch exists, broader control deferred) |
-| 54 | medium | S | community | test | Dashboard static-asset freshness gate in local-ci (like loki-ts/dist) |
-| 55 | medium | L | community | test | Windows/WSL runtime verification (no Windows CI runner) |
-| 56 | low | M | autonomous | cli | KPI reporting (loki report kpis) on bash legacy route (Bun-only today) |
-| 57 | medium | M | autonomous | cli | Voice mode (Issue #85) |
-| 58 | medium | M | community | test | Mutation testing (stryker) workflow re-enable + fix stale 5-provider assertions |
-| 59 | medium | XL | founder-gated | engine | Managed Agents multiagent path (LOKI_EXPERIMENTAL_MANAGED_*, research preview) |
-| 60 | low | M | autonomous | engine | Sentrux iteration-loop auto-gating (manual-only today) |
-| 61 | medium | S | founder-gated | adoption | npm registry publish prep for @autonomi/verify SDK (distribution wiring) |
-| 62 | low | M | founder-gated | adoption | Disclosed default-on PostHog telemetry (Phase L, blocked + deferred) |
-| 63 | medium | S | community | saas | MCP check_cost_honesty tool registration (deferred until cost core lands) |
-| 64 | medium | L | founder-gated | dashboard | Hosted dashboard with shareable verification badges |
-| 65 | medium | M | founder-gated | adoption | MCP registry submission + Glama claim + awesome-list PRs |
-| 66 | medium | M | autonomous | saas | autonomi-saas UI/UX pass: build-watching pane + finished cues from trust verdict |
-| 67 | medium | XL | autonomous | dashboard | Server.py route modularization (~10K-line monolith, no domain separation) |
-| 68 | medium | L | autonomous | dashboard | Large legacy-ui components modularization (5 files 1K-1.9K lines) |
-| 69 | medium | L | founder-gated | engine | Wave-13 deferred trust fixes + PRD-reuse spurious-update design fix |
-| 70 | medium | M | community | dashboard | Dashboard dark-toggle iframe theme not following SPA (harness/product parity) |
-| 71 | medium | M | community | ci | Sonnet-5 calibration follow-ups: test-verify --hosted + gate missing-tool handling |
-| 72 | medium | S | community | ci | Integrity audit (SBOM/provenance) blocking on publish (currently informational) |
-| 73 | medium | S | community | distribution | Python SDK version pre-check before publish (mismatch risk if publish fails post-tag) |
-| 74 | medium | S | community | ci | Docker buildx early-detection for uncommitted loki-ts/dist (silent COPY failure risk) |
-| 75 | medium | S | community | ci | loki-enterprise workflow enabled in main CI (dispatch-only today) |
-| 76 | medium | S | community | distribution | Brew tap real-install verification on clean macOS host post-release |
-| 77 | low | L | founder-gated | perf | Horizontal scaling / distributed rate-limit + persistence (in-memory today) |
-| 78 | low | S | community | saas | Optional build_artifacts table for enumeration/lifecycle/reaping (deferred) |
-| 79 | medium | S | community | saas | Headless browser (playwright-core) dep + Docker image size cap for screenshot capture |
-| 80 | low | M | community | engine | Wave-8/9 deferred tail: spawn_timeout removal, heredoc guards, MED/LOW findings |
-| 81 | medium | L | community | adoption | v7.19.1 traction: uncertainty-gated escalation + shareable proof + integrate 8 tutorials |
-| 82 | medium | M | autonomous | verify | Benchmarks HumanEval/SWE-bench: publish results (runners + datasets exist) |
-| 83 | low | S | autonomous | cli | loki run removal (deprecated alias for loki start, next major) |
-| 84 | low | S | autonomous | cli | Fable model tier wiring (collapses to Opus until API-available) |
-| 85 | low | S | community | distribution | Homepage/blog release update automation (Slack-notify-only today) |
-| 86 | low | S | community | distribution | Wiki-sync scheduled trigger (manual/dispatch only, skips without API key) |
-| 87 | low | L | autonomous | dashboard | Dashboard React app: implement four stub views (experimental app) |
-| 88 | low | S | community | test | Embeddings edge-case tests run without numpy (skip today) |
-| 89 | low | S | community | test | Hybrid-search test runs without chromadb skip |
-| 90 | low | S | community | test | PID-recycling test platform-portability (ps lstart skip for pid 1) |
-| 91 | low | S | community | test | Cloud integration tests (aider-cloud) run in a credentialed gated job |
-| 92 | low | S | community | test | Sentrux real-binary nightly promoted from best-effort to tracked |
-| 93 | low | S | community | test | Model-catalog probe: surface staleness in tests (issue-create skipped without admin scope) |
-| 94 | low | S | community | test | Phase 6 readiness check re-enable (cron disabled; issues disabled on repo) |
-| 95 | low | S | community | distribution | TS/Bun SDK (@loki-mode/sdk) discoverability from main package/docs |
-| 96 | low | S | community | distribution | Docker Hub description update resilience (PAT scope failures non-blocking) |
-| 97 | low | S | community | distribution | VS Code extension: keep source marked deprecated, not silently removed |
-| 98 | low | S | autonomous | dashboard | Web-app docs: fill video placeholder (coming-soon text) |
-| 99 | low | M | autonomous | cli | C# provider support (roslyn analyzers + dotnet build, deferred) |
-| 100 | low | XL | founder-gated | engine | BMAD-METHOD v6 integration (net-new, needs founder planning) |
-
-## Gate split (what can actually close autonomously) - 2026-07-01
-
-The /goal asks to work through all 100. The honest structural reality: only a
-subset is autonomously completable. The rest need a founder decision or are
-community/CI-infra work that cannot be forced from this session.
-
-- **Autonomous (I can execute end-to-end): ~25 items** (the `gate=autonomous` rows).
-- **Founder-gated (blocked on a decision only the founder can make): ~35 items.**
-  Licensing (#5), hosted deploy target (#6, #23, #36), enterprise RBAC/SSO (#15),
-  sandbox backend choice (#14, #42), spend (#49 SWE-bench, #18 live-model), npm
-  publish + brand (#4, #13, #61), telemetry (#62), GitHub App (#46). Prepped to
-  one-approval-away where possible; CANNOT close without the founder.
-- **Community / CI-infra (~40 items):** flake root-causes, coverage thresholds,
-  platform runners (ARM64/Windows), distribution resilience. Environmental
-  hardening, not single-session feature work.
-
-### Shipped this session (accuracy-moat, from the competitive gap-analysis)
-
-Net-new accuracy work driven from docs/research-2026-07/gap-analysis-backlog.json
-(a companion list), NOT retroactive Top-100 claims:
-
-- v7.105.0 - convergence: council evaluates on completion-claim (4.0x faster, n=3). commit b8a368a0.
-- v7.106.0 - reverse-classical test provenance (tautological tests downgraded). commit 71299faa.
-- v7.107.0 - loki mcp --transport http loopback bind + bearer auth (+ latent crash fix). commit f3aa523c.
-- v7.108.0 - runtime boot smoke gate + annotate-before-act expectation ledger. commit 806fc374.
-
-### Top-100 items with a concrete shipped commit
-
-- #40 (dashboard memory summary on JSON-backed projects) - commit 330de52d.
-- #22 (3 parallel code reviewers) - already present in the engine (council path).
-
-### Next autonomous batch (by value, unblocked, user-visible)
-
-Picked from the `autonomous` rows by value + real-user impact (not primitives that
-ship dormant). Verify each before starting; several may overlap the already-built
-private autonomi-verify TS engine (e.g. #7/#8 - check before rebuilding).