loki-mode 7.104.5 → 7.105.0

package/SKILL.md

@@ -3,7 +3,7 @@ name: loki-mode
 description: Autonomous spec-driven build system with a built-in trust layer. It does not call work done until it is verified (RARV-C closure loop, 8 quality gates, completion council, verified-completion evidence gate). Triggers on "Loki Mode". Takes a spec (PRD, GitHub issue, OpenAPI doc, etc.) to deployed product with minimal human intervention. Provider-agnostic. Requires --dangerously-skip-permissions flag.
 ---
 
-# Loki Mode v7.104.5
+# Loki Mode v7.105.0
 
 **You are an autonomous agent. You make decisions. You do not ask questions. You do not stop.**
 
@@ -408,4 +408,4 @@ See `CHANGELOG.md` entries [7.5.7], [7.5.8], [7.5.13] for the per-fix list and r
 
 ---
 
-**v7.104.5 | [Autonomi](https://www.autonomi.dev/) flagship product | ~260 lines core**
+**v7.105.0 | [Autonomi](https://www.autonomi.dev/) flagship product | ~260 lines core**

package/VERSION

@@ -1 +1 @@
-7.104.5
+7.105.0

package/autonomy/completion-council.sh

@@ -3226,10 +3226,36 @@ council_should_stop() {
         circuit_triggered=true
     fi
 
-    # Only run council at check intervals OR if circuit breaker triggered
+    # Only run council at check intervals OR if circuit breaker triggered OR the
+    # agent has EXPLICITLY claimed completion this iteration (v7.105.0 convergence
+    # fix). Rationale, measured from real build telemetry (docs/SPEED-DIAGNOSIS):
+    # a build could be verifiably done at iteration 1 (reviews non-blocking, tests
+    # green, checklist complete) yet the council was not even allowed to evaluate
+    # until the next 5-iteration boundary -- so the loop ground out "next
+    # improvement" iterations the user never asked for (14 iterations for a job
+    # done in ~1). An explicit completion CLAIM is the strongest possible signal
+    # that it is worth asking the council NOW. This changes only WHEN the council
+    # evaluates, never WHETHER it must approve: MIN_ITERATIONS below, the hard
+    # checklist gate, the evidence gate, the aggregate vote, and the devil's
+    # advocate all still run unchanged. A premature or unearned claim is still
+    # rejected (returns 1, loop continues); it is never a forced green.
+    #
+    # The claim signal is the structured one (loki_complete_task MCP tool ->
+    # .loki/signals/COMPLETION_REQUESTED, or the runner exporting
+    # LOKI_COMPLETION_CLAIMED=1 for this iteration), NOT the fuzzy log-grep the
+    # circuit breaker uses -- so it fires on a real, intentional claim only.
+    local completion_claimed=false
+    if [ "${LOKI_COMPLETION_CLAIMED:-0}" = "1" ] \
+       || [ -f "${TARGET_DIR:-.}/.loki/signals/COMPLETION_REQUESTED" ]; then
+        completion_claimed=true
+    fi
+
     local should_check=false
     if [ "$circuit_triggered" = "true" ]; then
         should_check=true
+    elif [ "$completion_claimed" = "true" ]; then
+        should_check=true
+        log_info "[Council] Completion claimed this iteration -- evaluating now (not deferring to the ${COUNCIL_CHECK_INTERVAL}-iteration interval)"
     elif [ $((ITERATION_COUNT % COUNCIL_CHECK_INTERVAL)) -eq 0 ]; then
         should_check=true
     fi

package/autonomy/run.sh

@@ -17530,7 +17530,30 @@ else:
             if type ensure_completion_test_evidence &>/dev/null; then
                 ensure_completion_test_evidence || true
             fi
-            if type council_should_stop &>/dev/null && council_should_stop; then
+            # v7.105.0 convergence: if the agent EXPLICITLY claimed completion this
+            # iteration (structured loki_complete_task / COMPLETION_REQUESTED
+            # signal), tell the council to evaluate NOW instead of deferring to the
+            # 5-iteration check interval. The council still must approve (MIN_
+            # ITERATIONS + hard checklist + evidence gate + vote + devil's advocate
+            # all unchanged); this only stops an already-verified build from
+            # grinding needless "next improvement" iterations. Scoped to this call.
+            #
+            # CRITICAL (council-review finding): we must PEEK at the claim signal
+            # NON-DESTRUCTIVELY. check_task_completion_signal() CONSUMES the signal
+            # (rm -f on read), and the DEFAULT completion-promise route below
+            # (v6.82.0, the live gate-based acceptance path) also consumes it -- so
+            # calling the consuming detector here would drop the claim before that
+            # route sees it, re-introducing the v7.28 "claim drop" bug. Instead we
+            # only test the signal FILES' existence (the same two paths
+            # check_task_completion_signal reads), leaving consumption to the
+            # existing single owner. The council side likewise peeks
+            # ($TARGET_DIR/.loki/signals/COMPLETION_REQUESTED) without consuming.
+            local _loki_completion_claimed=0
+            if [ -f "${TARGET_DIR:-.}/.loki/signals/TASK_COMPLETION_CLAIMED" ] \
+               || [ -f "${TARGET_DIR:-.}/.loki/signals/COMPLETION_REQUESTED" ]; then
+                _loki_completion_claimed=1
+            fi
+            if type council_should_stop &>/dev/null && LOKI_COMPLETION_CLAIMED="$_loki_completion_claimed" council_should_stop; then
                 # bash-F1: council_should_stop returns 0 from a genuine approval
                 # AND from two force-stop safety valves (stagnation flood /
                 # repeated done-signals). A force-stop is NOT a verified-complete

package/dashboard/__init__.py

@@ -7,7 +7,7 @@ Modules:
     control: Session control API (start/stop/pause/resume)
 """
 
-__version__ = "7.104.5"
+__version__ = "7.105.0"
 
 # Expose the control app for easy import
 try:

package/docs/FOUNDER-STATUS-AND-DECISIONS-2026-07-01.md

@@ -0,0 +1,106 @@
+# Founder Status + Decision List (2026-07-01)
+
+The consolidated "where everything stands + the decisions only you can make" report.
+Written after: 5 patch releases shipped, top-100 backlog enumerated + triaged, ecosystem
+currency assessed, no-HITL product principle recorded.
+
+## 1. Shipped this session (all live on npm/Docker/GH, council-gated)
+
+| Version | What | Verified |
+|---|---|---|
+| v7.104.2 | Native/Keychain Claude login recognized; `loki start` no longer false-blocks a logged-in user; `loki doctor` confirms login (bash+Bun parity) | reproduced + 8-case test + live Docker |
+| v7.104.3 | Task-list accuracy: honest done column, no empty cards / dups / fake-green (failed iters honestly labeled) | live vs real anonima, browser screenshots, council 3/3 |
+| v7.104.4 | `/api/tasks` never 500s on malformed dashboard-state.json | RED/GREEN-proven test |
+| v7.104.5 | Memory panel correct + self-consistent on JSON-backed projects (completes PR #178, @iizotov co-authored, PR closed with thank-you) | live vs anonima, parity tests |
+| (earlier) v7.104.2 auth + v7.104.3 task-list | (same as above; two-patch train) | |
+
+Also: closed issue #177 (spam advisory). PR #178 merged-in-substance + contributor thanked.
+
+## 2. Backlog reality (honest)
+
+The top-100 was enumerated (116 raw -> 100 ranked) and then TRIAGED against current source.
+Key finding: **the autonomous critical/high queue is essentially ALREADY DONE.**
+
+| Rank | Item | Triage verdict |
+|---|---|---|
+| 1 | run.sh self-delete EXIT trap | already fixed (guarded to /tmp/loki-run-*; verified safe) |
+| 2 | saas cross-tenant leak + stuck-Building | CLOSED (commits a7667d0/44f7c29/8b05b36; 187/187 tests) |
+| 9 | anti-fake-green + gate parity | shipped (verify 282 tests; coherentStatus guard) |
+| 10 | VS-7 worker cutover | CLOSED (commit e6b96ed; BFF 181 + worker 102 tests) |
+| 17 | base_unreachable suppresses empty_diff | already wired (completion-council.sh:1616) |
+| 22 | 3 parallel reviewers | already implemented (run.sh:10461-10954) |
+
+The autonomous well is mostly dry BECAUSE the real remaining leverage is founder-gated
+(the trust-as-product / enterprise surface). That is the honest headline: the next 10x is
+not another patch, it is the decisions below.
+
+## 3. Ecosystem currency (what loki CONSUMES from the Claude ecosystem)
+
+CLI invocation contract (claude 2.1.198): STABLE, no drift. Auth: fixed this session.
+Model catalog: current (opus-4-8, sonnet-5, haiku-4-5, fable-5). Real deltas are small and
+tracked in docs/ECOSYSTEM-CURRENCY-PLAN.json (EC-1..EC-4). Adjacent Anthropic PRODUCTS
+(Managed Agents, Claude Tag, loop-engineering, agent identity) assessed as peer/inspiration
+categories, NOT dependencies -- declined with reason (the CTO triage move).
+
+The one real net-new build from this: **EC-1, the external-tool contract adapter + drift
+test** -- the keychain incident this session proves it would catch a user-facing break
+before a user does. Building autonomously now.
+
+## 4. DECISIONS ONLY YOU CAN MAKE (founder-gated, prepped to one-approval-away)
+
+These are the real blockers to "easiest to integrate with anyone" + the trust-as-product moat.
+Per the no-HITL principle, these are the LEGITIMATE exceptions: irreversible / strategic /
+brand / legal / spend decisions a human must own. Each is ready for a one-word go.
+
+### A. Autonomi Verify productization (the trust-as-product core)
+1. **License decision (rank 5, S, BLOCKS everything commercial).** BUSL-1.1 vs open for
+   @autonomi/verify. No hosted/commercial launch until decided. DECISION: which license?
+2. **npm registry + package name (rank 4, M).** @autonomi/verify is currently unpublishable
+   (private:true, no exports/types/build, .ts bin). And a naming collision exists
+   (loki-vaas vs loki-verify bin vs `loki verify` subcommand, rank 13). DECISION: publish to
+   public npm as `@autonomi/verify`? Confirm the brand/bin name to resolve the collision.
+   -> Once decided, I can make it genuinely npm-installable + fix the collision autonomously.
+3. **Hosted /verify deploy (rank 6, M + rank 36/42 sandbox).** The signed neutral REST
+   endpoint is the category-defining asset but not deployed (Bun.serve-only, ephemeral key).
+   DECISION: deploy target (Fly/Render/AWS/your infra) + signing-key storage (secret manager)?
+   -> I can stage the Node entry + Dockerfile + serve script + 12-factor key guard to
+   one-command-deploy; I will NOT deploy or spend without your go.
+4. **wire verifyCompletion() pipeline (rank 3, XL).** The SDK/MCP verdict path is stubbed;
+   the whole value prop is unreachable until gate->integrity->council->cost->sign is wired.
+   Depends on (2)/(3) decisions. Large; I can start once the license + registry are set.
+
+### B. Enterprise scale (needed for any enterprise sale)
+5. **RBAC / SSO / API-key auth / metering (rank 15, XL)** + **on-prem persistence backend
+   (rank 16, L, Postgres/SQLite)** + **sandbox backend (rank 14, L, gVisor/Firecracker/Kata)**.
+   DECISION: which enterprise deploy shape do we target first (hosted multi-tenant vs on-prem)?
+   That choice sequences 14/15/16/36/42. -> I can build the PersistenceBackend interface +
+   RBAC scaffolding once you pick the shape.
+6. **GitHub App verification integration (rank 46, L, greenfield).** DECISION: build now or later?
+
+### C. Benchmark spend
+7. **SWE-bench Pro 119 resume (issue #174, paused 35/119).** Resume harness exists. DECISION:
+   spend the compute to finish? (This is a $ + time decision, hence gated.)
+
+### D. Outreach (rank/issue #172, low-risk but external-posting)
+8. MCP registry submission + Glama claim + awesome-list PRs. I can prepare the submissions;
+   the actual external POSTS need your nod (per action rules). DECISION: proceed with posts?
+
+## 5. What I am doing autonomously RIGHT NOW (no decision needed)
+- EC-1 external-contract adapter + drift test (the one real net-new build).
+- Small no-HITL audit: find every place the PRODUCT hard-stops/prompts a human; fix genuine
+  gaps so it decides from model+memory+cross-project context (degrading to honest-inconclusive,
+  never fake-green). Not a speculative refactor -- a triage + targeted fixes.
+- NOT building: dashboard steering (#44) -- it is a human-in-the-loop feature, deprioritized by
+  the no-HITL mandate; architect plan is durable and waits. NOT building "autonomous-everywhere"
+  as a speculative refactor.
+
+## 6. The one-line asks
+Reply with any subset:
+- **License:** BUSL / MIT / Apache / other for @autonomi/verify
+- **npm:** publish `@autonomi/verify` publicly? + confirm bin name (resolve loki-verify collision)
+- **Hosted deploy:** target + key storage (or "not yet")
+- **Enterprise shape:** hosted-multitenant-first vs on-prem-first (sequences the enterprise stories)
+- **SWE-bench:** spend to finish 119? (yes/no)
+- **Outreach:** post the MCP-registry/Glama/awesome-list submissions? (yes/no)
+
+Everything else, I continue autonomously.

package/docs/INSTALLATION.md

@@ -2,7 +2,7 @@
 
 The flagship product of [Autonomi](https://www.autonomi.dev/). Loki Mode is a spec-driven autonomous builder with a built-in trust layer that takes any spec to a deployed product and verifies completion with evidence (quality gates plus a completion council), not just a "done" claim. Complete installation instructions for all platforms and use cases.
 
-**Version:** v7.104.5
+**Version:** v7.105.0
 
 ---
 

package/docs/SPEED-DIAGNOSIS-2026-07-01.md

@@ -0,0 +1,61 @@
+# Loki Build Speed Diagnosis (2026-07-01)
+
+Measured from real build telemetry (anonima `.loki/events.jsonl`, stage_complete +
+iteration timeline), NOT from reading run.sh. Method per advisor: measure before optimizing.
+
+## Data hygiene
+anonima's events.jsonl contained **9 interleaved builds/sessions** (9 session_starts) plus
+session-long manual poking. Naive sums ("5 min/iter x 14") are contaminated. Split on
+session_start and analyzed the real ones.
+
+## The pathological build (session 3): the smoking gun
+A simple topic-based chat app. 14 ACT iterations, ~97 min of real work (idle gaps excluded).
+
+| iteration | real work |
+|---|---|
+| 1 | 2.1 min |
+| 6 | 9.9 min |
+| 10 | 15.3 min |
+| 12 | **29.6 min** |
+| (others) | 3-6 min each |
+
+**13 completion claims across 14 iterations.** The FIRST claim already stated:
+"All PRD requirements implemented and tests passing. Evidence: npm test => 19/19 pass...
+PRD checklist (14 items) fully implemented." Yet the loop ran 13 more iterations.
+
+## Root cause: NON-CONVERGENCE, not per-stage slowness
+- **Code review / 3-reviewer council is NOT the bottleneck** (0.1-2.8 min per iteration,
+  mostly <1 min). Cutting the council would harm accuracy for ~no speed gain. Ruled out.
+- **The bottleneck is iteration COUNT + a few pathological iterations.** The app was
+  essentially done early (real passing tests, checklist complete) but the RARV loop's
+  "there is NEVER a finished state -- always find the next improvement" behavior kept it
+  grinding polish iterations. For a user who wants a fullstack app fast, this is THE
+  frustration: loki keeps going after done.
+- iter 10 (15 min) and iter 12 (30 min) did disproportionate work -- candidate over-scoping
+  or re-analysis; needs per-iteration content inspection.
+
+## Buckets (advisor framework)
+1. **WASTE (cut freely):** polish iterations after genuine completion; the PRD-reuse
+   spurious-update re-reconciliation (already diagnosed, design fix pending). These add
+   wall-clock with no trust value.
+2. **VERIFICATION COST (keep -- this is the product):** council, gates, completion vote.
+   Fast already. The moat. Do NOT cut for speed.
+3. **STALE SCAFFOLDING (test per-stage):** machinery built for weaker models may be dead
+   weight on Opus 4.8 / Sonnet 5 (cf. Anthropic's "context resets became dead weight on
+   Opus 4.5"). Hypothesis -- must be tested against the benchmark, not assumed.
+
+## The fix direction (speed AND accuracy, never speed-by-cutting-verification)
+**Convergence:** stop as soon as the completion council genuinely agrees it is done
+(honest verified completion), instead of running "next improvement" iterations. This is a
+SPEED win that INCREASES trust alignment (stop when verified-done = the moat working), not a
+verification cut. The user's mandate is "accuracy AND speed"; convergence delivers both.
+
+## HARD PREREQUISITE (before any council/RARV-C change): the benchmark
+Both the speed directive and the "update council/RARV-C to latest research, show before/after
+accuracy" directive require a **clean reproducible benchmark on a fixed spec** emitting
+wall-clock + iteration count + completion verdict + pass/fail vs known acceptance criteria.
+Without it, changing the verification core is unmeasurable churn on the moat (architecture-
+level fake-green). Build the benchmark FIRST, capture before-numbers, fix the clearest waste
+(convergence), re-run, show real before->after. Then gate any council/RARV-C change on that
+benchmark, one named change at a time, kept only if accuracy goes up and wall-clock does not
+regress.

package/docs/TOP-100-BACKLOG.md

@@ -0,0 +1,122 @@
+# 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 dashboard-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) |

package/loki-ts/dist/loki.js

@@ -1,5 +1,5 @@
 // @bun
-var X8=Object.defineProperty;var K8=($)=>$;function q8($,Q){this[$]=K8.bind(null,Q)}var b=($,Q)=>{for(var Z in Q)X8($,Z,{get:Q[Z],enumerable:!0,configurable:!0,set:q8.bind(Q,Z)})};var L=($,Q)=>()=>($&&(Q=$($=0)),Q);var J$=import.meta.require;var v1={};b(v1,{lokiDir:()=>P,homeLokiDir:()=>T$,findRepoRootForVersion:()=>e$,REPO_ROOT:()=>h});import{resolve as r,dirname as i$}from"path";import{fileURLToPath as J8}from"url";import{existsSync as x$}from"fs";import{homedir as V8}from"os";function W8(){let $=y1;for(let Q=0;Q<6;Q++){if(x$(r($,"VERSION"))&&x$(r($,"autonomy/run.sh")))return $;let Z=i$($);if(Z===$)break;$=Z}return r(y1,"..","..","..")}function e$($){let Q=$;for(let Z=0;Z<6;Z++){if(x$(r(Q,"VERSION"))&&x$(r(Q,"autonomy/run.sh")))return Q;let z=i$(Q);if(z===Q)break;Q=z}return r($,"..","..","..")}function P(){return process.env.LOKI_DIR??r(process.cwd(),".loki")}function T$(){return r(V8(),".loki")}var y1,h;var C=L(()=>{y1=i$(J8(import.meta.url));h=W8()});import{readFileSync as U8}from"fs";import{resolve as H8,dirname as G8}from"path";import{fileURLToPath as B8}from"url";function N$(){if(Q$!==null)return Q$;let $="7.104.5";if(typeof $==="string"&&$.length>0)return Q$=$,Q$;try{let Q=G8(B8(import.meta.url)),Z=e$(Q);Q$=U8(H8(Z,"VERSION"),"utf-8").trim()}catch{Q$="unknown"}return Q$}var Q$=null;var $1=L(()=>{C()});var c1={};b(c1,{runOrThrow:()=>x8,run:()=>F,readStreamCapped:()=>u1,commandVersion:()=>S8,commandExists:()=>f,ShellError:()=>Q1,MAX_STDOUT_BYTES:()=>f1});async function u1($,Q=f1){let Z=$.getReader(),z=new TextDecoder,X="",q=0;try{while(q<Q){let{done:K,value:W}=await Z.read();if(K)break;if(!W)continue;if(q+=W.byteLength,q>Q){let V=W.byteLength-(q-Q);X+=z.decode(W.subarray(0,V),{stream:!0});break}X+=z.decode(W,{stream:!0})}X+=z.decode()}finally{try{await Z.cancel()}catch{}Z.releaseLock()}return X}async function F($,Q={}){let Z=Bun.spawn({cmd:[...$],stdout:"pipe",stderr:"pipe",env:Q.env?{...process.env,...Q.env}:process.env,cwd:Q.cwd}),z,X;if(Q.timeoutMs&&Q.timeoutMs>0)z=setTimeout(()=>{try{Z.kill("SIGTERM")}catch{}X=setTimeout(()=>{try{Z.kill("SIGKILL")}catch{}},2000)},Q.timeoutMs);try{let[q,K,W]=await Promise.all([u1(Z.stdout),new Response(Z.stderr).text(),Z.exited]);return{stdout:q,stderr:K,exitCode:W}}finally{if(z)clearTimeout(z);if(X)clearTimeout(X)}}async function x8($,Q={}){let Z=await F($,Q);if(Z.exitCode!==0)throw new Q1(`command failed (${Z.exitCode}): ${$.join(" ")}`,Z.exitCode,Z.stdout,Z.stderr);return Z}async function f($){let Q=N8($),Z=await F(["sh","-c",`command -v ${Q}`],{timeoutMs:5000});if(Z.exitCode===0)return Z.stdout.trim()||null;return null}function N8($){if(!/^[A-Za-z0-9._/-]+$/.test($))throw Error(`refused to shell-escape suspect token: ${$}`);return $}async function S8($,Q="--version"){if(!await f($))return null;let z=await F([$,Q],{timeoutMs:5000});if(z.exitCode!==0)return null;return((z.stdout||z.stderr).split(/\r?\n/)[0]?.trim()??"")||null}var f1=16777216,Q1;var d=L(()=>{Q1=class Q1 extends Error{message;exitCode;stdout;stderr;constructor($,Q,Z,z){super($);this.message=$;this.exitCode=Q;this.stdout=Z;this.stderr=z;this.name="ShellError"}}});function t($){return D8?"":$}var D8,T,S,_,nZ,I,k,y,J;var c=L(()=>{D8=(process.env.NO_COLOR??"").length>0;T=t("\x1B[0;31m"),S=t("\x1B[0;32m"),_=t("\x1B[1;33m"),nZ=t("\x1B[0;34m"),I=t("\x1B[0;36m"),k=t("\x1B[1m"),y=t("\x1B[2m"),J=t("\x1B[0m")});import{existsSync as c8}from"fs";async function Z$(){if(A$!==void 0)return A$;let $="/opt/homebrew/bin/python3.12";if(c8($))return A$=$,$;let Q=await f("python3.12");if(Q)return A$=Q,Q;let Z=await f("python3");return A$=Z,Z}async function z$($,Q={}){let Z=await Z$();if(!Z)return{stdout:"",stderr:"python3 not found",exitCode:127};return F([Z,"-c",$],Q)}var A$;var V$=L(()=>{d()});var J0={};b(J0,{runStatus:()=>U3});import{existsSync as v,readFileSync as U$,readdirSync as e1,statSync as $0}from"fs";import{resolve as D,basename as Q3}from"path";import{homedir as Z3}from"os";function Q0($){let Q=Math.trunc($);if(Q>=1e6)return`${(Math.trunc(Q/1e6*10)/10).toFixed(1)}M`;if(Q>=1000)return`${(Math.trunc(Q/1000*10)/10).toFixed(1)}K`;return String(Q)}function Z0($,Q,Z){if(Q===0)return null;let z=Math.trunc($*100/Q),X=Math.trunc($*D$/Q);if(X>D$)X=D$;let q=D$-X,K=S;if(z>=80)K=T;else if(z>=50)K=_;let W="=".repeat(Math.max(0,X))+" ".repeat(Math.max(0,q)),V=Q0($),U=Q0(Q);return`  ${k}${Z}${J} ${K}[${W}]${J} ${z}% (${V} / ${U})`}async function X3(){if(await f("jq"))return!0;return process.stdout.write(`${T}Error: jq is required but not installed.${J}
+var X8=Object.defineProperty;var K8=($)=>$;function q8($,Q){this[$]=K8.bind(null,Q)}var b=($,Q)=>{for(var Z in Q)X8($,Z,{get:Q[Z],enumerable:!0,configurable:!0,set:q8.bind(Q,Z)})};var L=($,Q)=>()=>($&&(Q=$($=0)),Q);var J$=import.meta.require;var v1={};b(v1,{lokiDir:()=>P,homeLokiDir:()=>T$,findRepoRootForVersion:()=>e$,REPO_ROOT:()=>h});import{resolve as r,dirname as i$}from"path";import{fileURLToPath as J8}from"url";import{existsSync as x$}from"fs";import{homedir as V8}from"os";function W8(){let $=y1;for(let Q=0;Q<6;Q++){if(x$(r($,"VERSION"))&&x$(r($,"autonomy/run.sh")))return $;let Z=i$($);if(Z===$)break;$=Z}return r(y1,"..","..","..")}function e$($){let Q=$;for(let Z=0;Z<6;Z++){if(x$(r(Q,"VERSION"))&&x$(r(Q,"autonomy/run.sh")))return Q;let z=i$(Q);if(z===Q)break;Q=z}return r($,"..","..","..")}function P(){return process.env.LOKI_DIR??r(process.cwd(),".loki")}function T$(){return r(V8(),".loki")}var y1,h;var C=L(()=>{y1=i$(J8(import.meta.url));h=W8()});import{readFileSync as U8}from"fs";import{resolve as H8,dirname as G8}from"path";import{fileURLToPath as B8}from"url";function N$(){if(Q$!==null)return Q$;let $="7.105.0";if(typeof $==="string"&&$.length>0)return Q$=$,Q$;try{let Q=G8(B8(import.meta.url)),Z=e$(Q);Q$=U8(H8(Z,"VERSION"),"utf-8").trim()}catch{Q$="unknown"}return Q$}var Q$=null;var $1=L(()=>{C()});var c1={};b(c1,{runOrThrow:()=>x8,run:()=>F,readStreamCapped:()=>u1,commandVersion:()=>S8,commandExists:()=>f,ShellError:()=>Q1,MAX_STDOUT_BYTES:()=>f1});async function u1($,Q=f1){let Z=$.getReader(),z=new TextDecoder,X="",q=0;try{while(q<Q){let{done:K,value:W}=await Z.read();if(K)break;if(!W)continue;if(q+=W.byteLength,q>Q){let V=W.byteLength-(q-Q);X+=z.decode(W.subarray(0,V),{stream:!0});break}X+=z.decode(W,{stream:!0})}X+=z.decode()}finally{try{await Z.cancel()}catch{}Z.releaseLock()}return X}async function F($,Q={}){let Z=Bun.spawn({cmd:[...$],stdout:"pipe",stderr:"pipe",env:Q.env?{...process.env,...Q.env}:process.env,cwd:Q.cwd}),z,X;if(Q.timeoutMs&&Q.timeoutMs>0)z=setTimeout(()=>{try{Z.kill("SIGTERM")}catch{}X=setTimeout(()=>{try{Z.kill("SIGKILL")}catch{}},2000)},Q.timeoutMs);try{let[q,K,W]=await Promise.all([u1(Z.stdout),new Response(Z.stderr).text(),Z.exited]);return{stdout:q,stderr:K,exitCode:W}}finally{if(z)clearTimeout(z);if(X)clearTimeout(X)}}async function x8($,Q={}){let Z=await F($,Q);if(Z.exitCode!==0)throw new Q1(`command failed (${Z.exitCode}): ${$.join(" ")}`,Z.exitCode,Z.stdout,Z.stderr);return Z}async function f($){let Q=N8($),Z=await F(["sh","-c",`command -v ${Q}`],{timeoutMs:5000});if(Z.exitCode===0)return Z.stdout.trim()||null;return null}function N8($){if(!/^[A-Za-z0-9._/-]+$/.test($))throw Error(`refused to shell-escape suspect token: ${$}`);return $}async function S8($,Q="--version"){if(!await f($))return null;let z=await F([$,Q],{timeoutMs:5000});if(z.exitCode!==0)return null;return((z.stdout||z.stderr).split(/\r?\n/)[0]?.trim()??"")||null}var f1=16777216,Q1;var d=L(()=>{Q1=class Q1 extends Error{message;exitCode;stdout;stderr;constructor($,Q,Z,z){super($);this.message=$;this.exitCode=Q;this.stdout=Z;this.stderr=z;this.name="ShellError"}}});function t($){return D8?"":$}var D8,T,S,_,nZ,I,k,y,J;var c=L(()=>{D8=(process.env.NO_COLOR??"").length>0;T=t("\x1B[0;31m"),S=t("\x1B[0;32m"),_=t("\x1B[1;33m"),nZ=t("\x1B[0;34m"),I=t("\x1B[0;36m"),k=t("\x1B[1m"),y=t("\x1B[2m"),J=t("\x1B[0m")});import{existsSync as c8}from"fs";async function Z$(){if(A$!==void 0)return A$;let $="/opt/homebrew/bin/python3.12";if(c8($))return A$=$,$;let Q=await f("python3.12");if(Q)return A$=Q,Q;let Z=await f("python3");return A$=Z,Z}async function z$($,Q={}){let Z=await Z$();if(!Z)return{stdout:"",stderr:"python3 not found",exitCode:127};return F([Z,"-c",$],Q)}var A$;var V$=L(()=>{d()});var J0={};b(J0,{runStatus:()=>U3});import{existsSync as v,readFileSync as U$,readdirSync as e1,statSync as $0}from"fs";import{resolve as D,basename as Q3}from"path";import{homedir as Z3}from"os";function Q0($){let Q=Math.trunc($);if(Q>=1e6)return`${(Math.trunc(Q/1e6*10)/10).toFixed(1)}M`;if(Q>=1000)return`${(Math.trunc(Q/1000*10)/10).toFixed(1)}K`;return String(Q)}function Z0($,Q,Z){if(Q===0)return null;let z=Math.trunc($*100/Q),X=Math.trunc($*D$/Q);if(X>D$)X=D$;let q=D$-X,K=S;if(z>=80)K=T;else if(z>=50)K=_;let W="=".repeat(Math.max(0,X))+" ".repeat(Math.max(0,q)),V=Q0($),U=Q0(Q);return`  ${k}${Z}${J} ${K}[${W}]${J} ${z}% (${V} / ${U})`}async function X3(){if(await f("jq"))return!0;return process.stdout.write(`${T}Error: jq is required but not installed.${J}
 `),process.stdout.write(`Install with:
 `),process.stdout.write(`  brew install jq    (macOS)
 `),process.stdout.write(`  apt install jq     (Debian/Ubuntu)
@@ -805,4 +805,4 @@ Set LOKI_LEGACY_BASH=1 to force the bash CLI for every command.
 `),2}default:return process.stderr.write(`Unknown command: ${Q}
 `),process.stderr.write(z8),2}}i1();process.on("SIGINT",()=>process.exit(130));process.on("SIGTERM",()=>process.exit(143));var EZ=await RZ(Bun.argv.slice(2));process.exit(EZ);
 
-//# debugId=CD7A0B6FD83A195A64756E2164756E21
+//# debugId=80BB634E7BCAD6EC64756E2164756E21

package/mcp/__init__.py

@@ -57,4 +57,4 @@ try:
 except ImportError:
     __all__ = ['mcp']
 
-__version__ = '7.104.5'
+__version__ = '7.105.0'

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "loki-mode",
   "mcpName": "io.github.asklokesh/loki-mode",
-  "version": "7.104.5",
+  "version": "7.105.0",
   "description": "Loki Mode by Autonomi. Autonomous spec-to-product system: takes a PRD, GitHub issue, OpenAPI/JSON/YAML, or one-line brief to a deployed app via the RARV-C closure loop with 8 quality gates. Provider-agnostic (Claude Code, OpenAI Codex, Cline, Aider).",
   "keywords": [
     "agent",

package/plugins/loki-mode/.claude-plugin/plugin.json

@@ -2,7 +2,7 @@
   "$schema": "https://json.schemastore.org/claude-code-plugin-manifest.json",
   "name": "loki-mode",
   "displayName": "Loki Mode",
-  "version": "7.104.5",
+  "version": "7.105.0",
   "description": "Autonomous spec-to-product build system with a built-in trust layer (RARV-C closure loop, 8 quality gates, completion council). Ships Loki's spec-hardening, drift-detection, and deterministic PR verification commands plus the Loki MCP server.",
   "author": {
     "name": "Autonomi",