loki-mode 8.32.0 → 8.34.0

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 v8.32.0
+# Loki Mode v8.34.0
 
 **You are an autonomous agent. You make decisions. You do not ask questions. You do not stop.**
 
@@ -469,4 +469,4 @@ See `CHANGELOG.md` entries [7.5.7], [7.5.8], [7.5.13] for the per-fix list and r
 
 ---
 
-**v8.32.0 | [Autonomi](https://www.autonomi.dev/) flagship product | ~410 lines core**
+**v8.34.0 | [Autonomi](https://www.autonomi.dev/) flagship product | ~410 lines core**

package/VERSION

@@ -1 +1 @@
-8.32.0
+8.34.0

package/autonomy/run.sh

@@ -14330,6 +14330,43 @@ reviewers = mandatory + [
         }
         for name in installed_selected
     ]
+# TOTAL council cap (LOKI_REVIEW_MAX_REVIEWERS). The tier map sizes the
+# SPECIALIST slots ({simple:2, standard:2, complex:4}), but installed agents and
+# the dependency-analyst append AFTER that sizing, so nothing bounded the total.
+# Measured consequence, from real code_review_start/complete pairs:
+#
+#     3 reviewers ->  31s        6 reviewers -> 177s
+#     7 reviewers -> 280s        7 reviewers -> 502s
+#
+# Dispatch is already concurrent, so this superlinearity is the max-of-N tail
+# plus contention on one provider -- a scoped issue was drawing a 7-member
+# council (including two overlapping security reviewers) and paying 9-16x the
+# 3-member wall clock for it.
+#
+# TRIMMING ORDER IS A SAFETY PROPERTY. Mandatory reviewers
+# (requirements-verifier, architecture-strategist, maintainer-mergeability) are
+# NEVER dropped: each carries a mandate no keyword-selected specialist has, and
+# shrinking a council must never be able to manufacture an approval. Only the
+# appended tail (installed agents, then keyword specialists beyond the floor) is
+# trimmed, and the cap can never cut below the mandatory set.
+#
+# Default 0 = uncapped, preserving today's behaviour exactly. This is a knob to
+# be turned on deliberately per route, not a silent change to every council.
+try:
+    _cap = int(os.environ.get("LOKI_REVIEW_MAX_REVIEWERS", "0") or "0")
+except ValueError:
+    _cap = 0
+if _cap > 0 and len(reviewers) > _cap:
+    _mandatory_names = {r["name"] for r in mandatory}
+    _keep = [r for r in reviewers if r["name"] in _mandatory_names]
+    for _r in reviewers:
+        if len(_keep) >= _cap:
+            break
+        if _r["name"] not in _mandatory_names:
+            _keep.append(_r)
+    # Never below the mandatory set, even if the cap is set lower than it.
+    reviewers = _keep if len(_keep) >= len(_mandatory_names) else reviewers
+
 if os.environ.get("LOKI_REVIEW_REQUIREMENTS_ONLY") == "1":
     reviewers = [
         reviewer for reviewer in reviewers
@@ -22839,8 +22876,37 @@ if __name__ == "__main__":
                     log_warn "Invariant gate FAILED ($inv_count consecutive) - CRITICAL/HIGH invariant/property violations (advisory; surfaced to next iteration)"
                 fi
             fi
+            # SKIP THE COUNCIL WHEN A DETERMINISTIC GATE ALREADY FAILED
+            # (LOKI_REVIEW_SKIP_ON_GATE_FAIL, default off).
+            #
+            # Measured: the council costs 31s at 3 reviewers and 280-502s at 6-7.
+            # The gates above cost ~6s COMBINED (static_analysis 5s, security_scan
+            # 1s, lsp_diagnostics 1s, test_suite <1s). When one of them has already
+            # failed, the iteration cannot be accepted no matter what the council
+            # says -- gate_failures is non-empty and feeds the same completion
+            # decision -- so the review is spending 280-502s to produce advice on
+            # code that is already going back for another pass.
+            #
+            # WHAT THIS IS NOT. It does not weaken any gate: a skipped review is
+            # recorded as skipped, never as a PASS, and the failing gate still
+            # blocks exactly as before. It cannot turn a rejection into an
+            # approval -- it only declines to spend five minutes describing a
+            # rejection that is already decided.
+            #
+            # DEFAULT OFF. Review findings are also next-iteration STEERING
+            # (LOKI_INJECT_FINDINGS), so skipping trades some guidance for a large
+            # latency win. That trade is a per-route decision, not a silent
+            # global one.
+            local _skip_review=false
+            if [ "${LOKI_REVIEW_SKIP_ON_GATE_FAIL:-false}" = "true" ] \
+               && [ -n "${gate_failures:-}" ]; then
+                _skip_review=true
+            fi
+            if [ "$_skip_review" = "true" ]; then
+                log_warn "Code review SKIPPED: deterministic gates already failed (${gate_failures%,}). The iteration is already going back; not spending a full council on it. Unset LOKI_REVIEW_SKIP_ON_GATE_FAIL to always review."
+                emit_stage_complete "code_review" "skipped" "$(date +%s 2>/dev/null)"
             # Code review gate (upgraded from advisory, with escalation)
-            if [ "$PHASE_CODE_REVIEW" = "true" ] && [ "$ITERATION_COUNT" -gt 0 ]; then
+            elif [ "$PHASE_CODE_REVIEW" = "true" ] && [ "$ITERATION_COUNT" -gt 0 ]; then
                 log_info "Quality gate: code review..."
                 local _stg_t0=$(date +%s 2>/dev/null); local _stg_ok=pass
                 if run_code_review; then

package/dashboard/__init__.py

@@ -7,7 +7,7 @@ Modules:
     control: Session control API (start/stop/pause/resume)
 """
 
-__version__ = "8.32.0"
+__version__ = "8.34.0"
 
 # Expose the control app for easy import
 try:

package/docs/MODEL-REGISTRY-PLAN.md

@@ -0,0 +1,155 @@
+# One JSON file to update when a provider ships a model
+
+Founder ask (2026-08-01): a simple JSON we define at packaging time, so
+updating for new model releases is easy.
+
+## The shape
+
+`providers/model_registry.json` -- ONE file, flat, no cross-references. To adopt
+a new model you edit one line.
+
+```json
+{
+  "registry_version": "2026-08-01.1",
+  "_howto": "Edit a value below and ship. Nothing else needs to change.",
+
+  "tiers": {
+    "frontier": {
+      "anthropic": "opus",
+      "openai":    "gpt-5.6",
+      "google":    "gemini-3.1-pro-preview"
+    },
+    "balanced": {
+      "anthropic": "sonnet",
+      "openai":    "gpt-5.6-terra",
+      "google":    "gemini-3.6-flash"
+    },
+    "fast": {
+      "anthropic": "haiku",
+      "openai":    "gpt-5.6-luna",
+      "google":    "gemini-flash-latest"
+    }
+  },
+
+  "pinned": {
+    "_comment": "Optional. Set to override an alias with an exact snapshot.",
+    "anthropic": {},
+    "openai": {},
+    "google": {}
+  }
+}
+```
+
+That is the whole contract. Three tiers x three providers.
+
+## The one design decision that matters: ALIAS, not snapshot
+
+The `anthropic` column holds `opus` / `sonnet` / `haiku`, not
+`claude-opus-5` / `claude-sonnet-5` / `claude-haiku-4-5`.
+
+**Verified against the live CLI on 2026-08-01:**
+
+| passed | resolves to | note |
+|---|---|---|
+| `opus` | `claude-opus-5` | tracks latest automatically |
+| `best` | `claude-fable-5` | Anthropic's most capable GA model |
+| `haiku` | `claude-haiku-4-5-20251001` | |
+| `claude-haiku-4-6` | **REJECTED** | "may not exist or you may not have access" |
+
+Two facts fall out of that table, and both argue for aliases:
+
+1. **Aliases self-update.** When Anthropic ships Opus 6, `opus` follows it with
+   zero edits to this file. A snapshot id would need a release to track it --
+   which is exactly how our catalog ended up pinned to `claude-opus-4-8` while
+   Opus 5 shipped (fixed in v8.31.0).
+2. **Snapshot ids are account-scoped.** `claude-haiku-4-6` may exist on the API
+   and still be unreachable on a given account/CLI. Hardcoding it ships a
+   broken default to whoever lacks access.
+
+That second point is not theoretical. We already have a scar:
+`tests/test-codex-model-trusted.sh` records that pinning `gpt-5.3-codex` broke
+**every ChatGPT-account user**, because codex-cli rejects it on that tier -- and
+Codex ships free with every ChatGPT plan. The safe landing spot was "send no
+model and let the provider choose."
+
+**Rule: prefer a provider alias. Use an exact id only in `pinned`, only
+deliberately.**
+
+## Why `pinned` exists
+
+Anthropic's docs are explicit that Claude 4.6-generation dateless identifiers
+are still **pinned snapshots, not evergreen pointers**. And Claude Code's
+`best` / `opus` / `sonnet` are *Claude Code routing aliases* -- not a portable
+contract in the raw Messages API.
+
+So: aliases are right for the CLI providers we drive, and `pinned` is the escape
+hatch for anyone who needs reproducibility or is calling an API directly.
+Empty by default.
+
+## What "frontier" maps to for Anthropic
+
+`best` resolves to `claude-fable-5` today, and Anthropic positions Fable as most
+capable generally, with `claude-opus-5` aimed specifically at complex agentic
+coding and enterprise work.
+
+**We map `frontier` -> `opus`, not `best`.** Our workload IS agentic coding, and
+`fable` is advisory-only in this codebase: the runner collapses it to opus, so
+offering it as an execution model would be a cost surprise. That reasoning
+predates this file and is unchanged.
+
+## What this replaces, and what it does not
+
+`providers/model_catalog.json` stays. It carries per-provider metadata this file
+deliberately does not (validation prefixes, tier fallbacks, aider's litellm
+strings, the `models[]` ordering that the resolver depends on).
+
+`model_registry.json` is the **editable surface**: the file a human opens when a
+provider ships something. The catalog becomes derived where they overlap, and
+`tests/test-model-catalog-single-source.sh` already fails when derived mirrors
+disagree with their source -- that mechanism extends to cover this.
+
+## Rollout, matching the lifecycle in the founder's note
+
+The note's recommended lifecycle is right, and one step already exists here:
+
+```
+provider release detected      <- tools/probe-model-catalog.py (exists, reads live docs)
+        v
+add as frontier candidate      <- edit model_registry.json
+        v
+run coding-agent eval suite    <- benchmarks/ (exists)
+        v
+canary                         <- NOT built; needs traffic splitting we do not have
+        v
+promote alias                  <- edit one line
+        v
+keep previous as fallback      <- pinned{} holds the old id
+```
+
+**Honest gap:** we have no traffic-splitting layer, so 5% canary is not
+implementable today. Do not put it in a release note as though it were.
+Aliases blunt the need -- the provider is the one rolling the model forward,
+and our eval suite plus the previous id in `pinned` is the realistic control.
+
+## Resolution order (must stay identical on both routes)
+
+```
+LOKI_<PROVIDER>_MODEL        (trusted verbatim -- fine-tunes, org models)
+  -> pinned[provider][tier]  (deliberate snapshot)
+  -> tiers[tier][provider]   (the alias -- the normal path)
+  -> provider default        (may legitimately be EMPTY: see codex/ChatGPT)
+```
+
+The last line is load-bearing. Empty means "send no `--model` and let the
+provider pick," which is correct whenever we cannot know an id is valid for that
+account. Never replace it with a guess.
+
+## Acceptance
+
+- Adding a model is a one-line edit to `model_registry.json`.
+- `tests/test-model-catalog-current-flagship.sh` (v8.31.0) already fails when a
+  tier points at a superseded same-family model; extend it to this file.
+- A mutation flipping any alias back to a hardcoded snapshot must turn a test
+  red.
+- No test may assert an exact snapshot id that this account cannot dispatch --
+  the `claude-haiku-4-6` result above is the reason.

package/docs/SPEED-AND-FIRST-PASS-PLAN.md

@@ -0,0 +1,239 @@
+# Fastest First-Pass Completion: measured plan
+
+Goal (founder, 2026-08-01): **fastest first-pass full completion** -- highest
+quality output, no waiting through a second iteration, faster than Cursor /
+Cognition / Replit on the axis a user actually feels.
+
+This plan is built on measurement first, competitor research second, and it
+DISCARDS the intuition I started with.
+
+---
+
+## 1. What the competition actually does (researched, not assumed)
+
+| Vendor | Speed mechanism | Reachable for us? |
+|---|---|---|
+| **Cursor** | Composer: in-house MoE model, ~250 tok/s (~4x GPT-5/Sonnet). MXFP8 training, no post-train quant. Compaction-in-the-loop RL cuts context errors 50%. | **No.** Custom frontier model. |
+| **Cognition** | SWE-1.6 at ~950 tok/s; SWE-bench 51.5%. | **No.** Custom model. |
+| **Replit** | Agent 3: ~2 min to visual preview, ~10 min to first app version. Design Mode <2 min. 2-3x gains across 2025. | **Partly.** Their edge is product/lifecycle, not model. |
+
+**The strategic conclusion.** Cursor and Cognition bought speed by TRAINING
+THEIR OWN MODELS. We cannot copy that and should stop pretending the gap is
+closable that way. Replit's edge is time-to-first-visible-thing, which IS
+reachable, because it is a product and lifecycle property.
+
+The industry-wide numbers that matter for us, from the latency research:
+
+- **Tool execution is 35-61% of total agent request time.** Harness, not model.
+- Parallel tool calling / speculative execution deliver **2-5x latency
+  reduction while preserving correctness**.
+- Prompt caching turns repeated-prefix attention from O(n^2) to O(n).
+
+That 35-61% is our lane. It is won by architecture, and it does not require a
+frontier lab.
+
+## 2. What we actually cost, MEASURED (this is the surprise)
+
+Real `stage_complete` events from builds on this machine (n=35):
+
+| stage | n | median s | max s | total s |
+|---|---:|---:|---:|---:|
+| **code_review** | 3 | **281** | **504** | **1055** |
+| static_analysis | 5 | 5 | 14 | 26 |
+| security_scan | 5 | 1 | 2 | 4 |
+| lsp_diagnostics | 4 | 1 | 1 | 2 |
+| test_suite | 4 | 0 | 1 | 1 |
+| mutation_integrity | 4 | 0 | 1 | 1 |
+| mock_integrity | 4 | 0 | 0 | 0 |
+| doc_coverage | 3 | 0 | 0 | 0 |
+| magic_debate | 3 | 0 | 0 | 0 |
+| **TOTAL** | | | | **1089** |
+
+**`code_review` is 97% of all measured gate time.** Everything else combined is
+34 seconds.
+
+### Two intuitions this killed
+
+**First:** I was about to propose parallelizing the seven sequential gates at
+`run.sh:22616-22980`. They ARE sequential and mostly independent. The change
+would have been clean, defensible, and worth **about 20 seconds**.
+
+**Second, and worse:** I then proposed parallelizing the reviewer council --
+which was **already parallel**. I had grepped the wrong line range, got no
+match, and treated that silence as evidence. A plan whose headline item was a
+no-op.
+
+Both were caught by going back to data instead of trusting the story. The
+standing rule for this document:
+
+> **No optimization ships without a before-number from this table, and no
+> claim about how the code behaves ships without reading the code that does
+> it.**
+
+An absent grep match is not evidence of absence -- it is evidence the grep did
+not match.
+
+### Where the real time goes
+
+Code review dominates, and the driver is **council size**, not sequential
+dispatch (see the correction in P0 -- my first reading of this was wrong).
+
+Measured: 3 reviewers finish in 31s; 6-7 reviewers take 177-502s. The council
+is already concurrent, so what grows is the max-of-N tail plus contention on a
+single provider.
+
+This explains the founder's "21 minutes for a simple GitHub issue" better than
+any other measurement taken today: a scoped issue was drawing a 7-member
+council containing two overlapping security reviewers.
+
+---
+
+## 3. The plan, ranked by measured seconds returned
+
+### P0 -- CORRECTED: the council is ALREADY parallel. The cost is its SIZE.
+
+**This section originally claimed reviewers ran sequentially and proposed
+parallelizing them. That was wrong, and the correction is recorded here rather
+than quietly edited out.**
+
+`run_code_review` forks every reviewer with `) &` (run.sh:14803), collects PIDs,
+and `wait`s on each (run.sh:14880). It has been parallel all along. My first
+grep searched the wrong line range, returned nothing, and I read that silence as
+proof of absence -- the same mistake class this codebase has been punishing all
+session.
+
+**What the data actually says.** Pairing `code_review_start` with
+`code_review_complete` across every recorded review:
+
+| reviewers | seconds | verdict |
+|---:|---:|---|
+| 7 | 502 | 0 pass / 7 fail |
+| 7 | 280 | 1 pass / 6 fail |
+| 6 | 177 | 0 pass / 5 fail |
+| **3** | **31** | 2 pass / 1 fail |
+
+**3 reviewers = 31s. 7 reviewers = 280-502s.** Roughly 2x the council for 9-16x
+the wall clock. Since dispatch is already concurrent, that superlinearity is not
+the count itself -- it is that a larger council pulls in slower reviewers and
+contends for the same provider, so the max-of-N tail dominates.
+
+**And the 7-member council is partly redundant:**
+
+```
+architecture-strategist, maintainer-mergeability,
+security-sentinel, review-security,      <- TWO security reviewers
+performance-oracle, eng-qa, dependency-analyst
+```
+
+`security-sentinel` and `review-security` overlap. `eng-qa` and
+`dependency-analyst` are appended agents, not part of the sized battery.
+
+**The real P0, in priority order:**
+
+1. **Deduplicate overlapping reviewers.** Two security reviewers on one diff is
+   paying the tail cost twice for one signal. Collapse by mandate, not by name.
+2. **Cap the effective council for scoped changes.** The tier map is
+   `{simple: 2, standard: 2, complex: 4}` specialists + 2 mandatory. Appended
+   agents bypass that sizing entirely, which is how 4 becomes 7. Bound the
+   TOTAL, not just the specialist slots.
+3. **Bound the tail, not the mean.** One slow reviewer sets the whole council's
+   latency. A per-reviewer deadline that records a non-vote (never a silent
+   pass) converts a 502s worst case into a bounded one.
+
+Expected: the 6-7 member case moves toward the measured 3-member behaviour
+(31s) for scoped work, while `complex` keeps its deeper battery.
+
+**Fail-safe direction is load-bearing:** a dropped or deadlined reviewer must
+count as a NON-VOTE exactly as today, never as a pass. Shrinking a council must
+never be able to manufacture approval. Guard with a test asserting the verdict
+is identical for the same fixture at any council size, and that a deadlined
+reviewer never contributes a PASS.
+
+### P1 -- Do not run the full council on iteration 1 of a scoped change
+
+The 8-gate council exists to catch regressions in a large build. On a scoped
+GitHub-issue fix, running a 3-reviewer council before the change is even
+verified is spending 281s to review something a test run would disprove in 1s.
+
+- Order: cheap deterministic gates FIRST (test_suite, static_analysis,
+  lsp_diagnostics -- 6s combined), council only if those pass.
+- A failing test means the council would have rejected anyway; running it first
+  is strictly wasted wall-clock.
+- Expected: on a failing first pass, ~281s saved outright.
+
+### P2 -- Time-to-first-signal (the Replit lesson)
+
+Replit shows a visual preview in ~2 minutes. We show nothing until an iteration
+completes. Even when our total time is competitive, the FELT time is worse.
+
+- Emit a first-signal event as soon as the agent's first file write lands.
+- `.loki/app-runner/first-preview.json` already exists (write-once, bash route)
+  -- surface it, and extend to the non-preview case as "first artifact".
+- This is perception, not throughput, and it is cheap.
+
+### P3 -- First-pass correctness (the actual "no second iteration" ask)
+
+Research finding: success now hinges on **specification quality and dynamic
+context**, not static upfront planning. Notably, auto-generated context files
+REDUCED success ~3% while human-written ones improved it ~4%, both raising cost
+20%+. So "generate a big context file" is measurably the wrong move.
+
+What the evidence supports:
+- ACE (ICLR 2026): generate -> reflect -> curate as an evolving context loop:
+  **+10.6% on coding benchmarks**.
+- We already have the seam: `LOKI_INJECT_FINDINGS` feeds structured per-finding
+  records into the next iteration.
+- The lever is making iteration 1's prompt carry what iteration 2 would have
+  learned -- which the existing first-pass-excellence work began and which
+  measured 2.8x cheaper with correctness held.
+
+**Do not** add a static generated context file. The data says it hurts.
+
+### P4 -- Prompt-cache discipline (already partly built, verify it holds)
+
+`[CACHE_BREAKPOINT]` splits a cache-stable prefix from a volatile tail. Cache
+reads price at 0.1x input. Any always-on instruction added to the wrong side
+busts the cache every iteration.
+
+- Add a regression test asserting no volatile content crosses the breakpoint.
+- This is a cost lever primarily, and a TTFT lever secondarily.
+
+---
+
+## 4. What we do NOT do
+
+- **Do not train a model.** Cursor and Cognition's speed is bought with custom
+  MoE models at 250-950 tok/s. That is not a gap we close with harness work,
+  and claiming otherwise would be dishonest.
+- **Do not parallelize the seven cheap gates as a headline.** Measured worth:
+  ~20s. Do it opportunistically inside P1's reordering, never sold as the win.
+- **Do not add static generated context files.** Measured -3% success, +20%
+  cost.
+
+## 5. How we know it worked
+
+Every item ships with a before/after from the same `stage_complete` telemetry
+that produced the table above. The acceptance number for the founder's
+complaint:
+
+> a scoped GitHub issue completes in **under 5 minutes**, first pass,
+> with the council verdict intact.
+
+Current measured critical path for that case is dominated by 281s of
+sequential code review. P0 + P1 target exactly that.
+
+## 6. Honest competitive position after this plan
+
+- **vs Cursor** on raw interactive latency: still behind. Different category.
+- **vs Replit** on time-to-first-visible: reachable with P2.
+- **vs Cognition** on trust: ahead, and that is the moat -- the Evidence
+  Receipt, now that the gates behind it actually run (v8.24.0-v8.31.0).
+
+The wedge is **verified-and-fast**, not fastest-in-absolute. We should say that
+plainly rather than claim a speed crown we cannot hold.
+
+---
+
+Sources: Cursor Composer blog, VentureBeat Composer coverage, Cognition
+SWE-1.6 reporting, Replit Agent 3 reviews, Zylos speculative-execution/parallel
+tool-calling research, ACE (ICLR 2026), 2026 context-engineering surveys.

package/loki-ts/dist/loki.js

@@ -1,5 +1,5 @@
 // @bun
-var h_=Object.create;var{getPrototypeOf:v_,defineProperty:rK,getOwnPropertyNames:g_}=Object;var m_=Object.prototype.hasOwnProperty;function u_(Z){return this[Z]}var p_,d_,c_=(Z,X,Q)=>{var Y=Z!=null&&typeof Z==="object";if(Y){var J=X?p_??=new WeakMap:d_??=new WeakMap,z=J.get(Z);if(z)return z}Q=Z!=null?h_(v_(Z)):{};let K=X||!Z||!Z.__esModule?rK(Q,"default",{value:Z,enumerable:!0}):Q;for(let $ of g_(Z))if(!m_.call(K,$))rK(K,$,{get:u_.bind(Z,$),enumerable:!0});if(Y)J.set(Z,K);return K};var qQ=(Z,X)=>()=>(X||Z((X={exports:{}}).exports,X),X.exports);var l_=(Z)=>Z;function i_(Z,X){this[Z]=l_.bind(null,X)}var l0=(Z,X)=>{for(var Q in X)rK(Z,Q,{get:X[Q],enumerable:!0,configurable:!0,set:i_.bind(X,Q)})};var p=(Z,X)=>()=>(Z&&(X=Z(Z=0)),X);var e0=import.meta.require;var kO={};l0(kO,{lokiDir:()=>j0,homeLokiDir:()=>P4,findRepoRootForVersion:()=>eK,REPO_ROOT:()=>r0});import{resolve as n7,dirname as tK}from"path";import{fileURLToPath as a_}from"url";import{existsSync as GQ}from"fs";import{homedir as s_}from"os";function n_(){let Z=RO;for(let X=0;X<6;X++){if(GQ(n7(Z,"VERSION"))&&GQ(n7(Z,"autonomy/run.sh")))return Z;let Q=tK(Z);if(Q===Z)break;Z=Q}return n7(RO,"..","..","..")}function eK(Z){let X=Z;for(let Q=0;Q<6;Q++){if(GQ(n7(X,"VERSION"))&&GQ(n7(X,"autonomy/run.sh")))return X;let Y=tK(X);if(Y===X)break;X=Y}return n7(Z,"..","..","..")}function j0(){return process.env.LOKI_DIR??n7(process.cwd(),".loki")}function P4(){return n7(s_(),".loki")}var RO,r0;var G8=p(()=>{RO=tK(a_(import.meta.url));r0=n_()});import{readFileSync as o_}from"fs";import{resolve as r_,dirname as t_}from"path";import{fileURLToPath as e_}from"url";function f3(){if(h5!==null)return h5;let Z="8.32.0";if(typeof Z==="string"&&Z.length>0)return h5=Z,h5;try{let X=t_(e_(import.meta.url)),Q=eK(X);h5=o_(r_(Q,"VERSION"),"utf-8").trim()}catch{h5="unknown"}return h5}var h5=null;var HQ=p(()=>{G8()});var bO={};l0(bO,{runOrThrow:()=>Of,run:()=>E0,readStreamCapped:()=>UQ,commandVersion:()=>Af,commandExists:()=>X9,ShellError:()=>Z$,MAX_STDOUT_BYTES:()=>yO});async function UQ(Z,X=yO){let Q=Z.getReader(),Y=new TextDecoder,J="",z=0;try{while(z<X){let{done:K,value:$}=await Q.read();if(K)break;if(!$)continue;if(z+=$.byteLength,z>X){let W=$.byteLength-(z-X);J+=Y.decode($.subarray(0,W),{stream:!0});break}J+=Y.decode($,{stream:!0})}J+=Y.decode()}finally{try{await Q.cancel()}catch{}Q.releaseLock()}return J}async function E0(Z,X={}){let Q=Bun.spawn({cmd:[...Z],stdout:"pipe",stderr:"pipe",env:X.env?{...process.env,...X.env}:process.env,cwd:X.cwd}),Y,J;if(X.timeoutMs&&X.timeoutMs>0)Y=setTimeout(()=>{try{Q.kill("SIGTERM")}catch{}J=setTimeout(()=>{try{Q.kill("SIGKILL")}catch{}},2000)},X.timeoutMs);try{let[z,K,$]=await Promise.all([UQ(Q.stdout),new Response(Q.stderr).text(),Q.exited]);return{stdout:z,stderr:K,exitCode:$}}finally{if(Y)clearTimeout(Y);if(J)clearTimeout(J)}}async function Of(Z,X={}){let Q=await E0(Z,X);if(Q.exitCode!==0)throw new Z$(`command failed (${Q.exitCode}): ${Z.join(" ")}`,Q.exitCode,Q.stdout,Q.stderr);return Q}async function X9(Z){let X=Lf(Z),Q=await E0(["sh","-c",`command -v ${X}`],{timeoutMs:5000});if(Q.exitCode===0)return Q.stdout.trim()||null;return null}function Lf(Z){if(!/^[A-Za-z0-9._/-]+$/.test(Z))throw Error(`refused to shell-escape suspect token: ${Z}`);return Z}async function Af(Z,X="--version"){if(!await X9(Z))return null;let Y=await E0([Z,X],{timeoutMs:5000});if(Y.exitCode!==0)return null;return((Y.stdout||Y.stderr).split(/\r?\n/)[0]?.trim()??"")||null}var yO=16777216,Z$;var x9=p(()=>{Z$=class Z$ extends Error{message;exitCode;stdout;stderr;constructor(Z,X,Q,Y){super(Z);this.message=Z;this.exitCode=X;this.stdout=Q;this.stderr=Y;this.name="ShellError"}}});function o7(Z){return jf?"":Z}var jf,L0,k8,p0,ZV0,i0,H8,Q9,v;var S6=p(()=>{jf=(process.env.NO_COLOR??"").length>0;L0=o7("\x1B[0;31m"),k8=o7("\x1B[0;32m"),p0=o7("\x1B[1;33m"),ZV0=o7("\x1B[0;34m"),i0=o7("\x1B[0;36m"),H8=o7("\x1B[1m"),Q9=o7("\x1B[2m"),v=o7("\x1B[0m")});import{existsSync as xf}from"fs";async function E7(){if(k4!==void 0)return k4;let Z="/opt/homebrew/bin/python3.12";if(xf(Z))return k4=Z,Z;let X=await X9("python3.12");if(X)return k4=X,X;let Q=await X9("python3");return k4=Q,Q}async function Y7(Z,X={}){let Q=await E7();if(!Q)return{stdout:"",stderr:"python3 not found",exitCode:127};return E0([Q,"-c",Z],X)}var k4;var r7=p(()=>{x9()});var ZL={};l0(ZL,{runStatus:()=>Yh});import{existsSync as Y9,readFileSync as v3,readdirSync as iO,statSync as aO}from"fs";import{resolve as h8,basename as sf}from"path";import{homedir as nf}from"os";function sO(Z){let X=Math.trunc(Z);if(X>=1e6)return`${(Math.trunc(X/1e6*10)/10).toFixed(1)}M`;if(X>=1000)return`${(Math.trunc(X/1000*10)/10).toFixed(1)}K`;return String(X)}function nO(Z,X,Q){if(X===0)return null;let Y=Math.trunc(Z*100/X),J=Math.trunc(Z*NQ/X);if(J>NQ)J=NQ;let z=NQ-J,K=k8;if(Y>=80)K=L0;else if(Y>=50)K=p0;let $="=".repeat(Math.max(0,J))+" ".repeat(Math.max(0,z)),W=sO(Z),V=sO(X);return`  ${H8}${Q}${v} ${K}[${$}]${v} ${Y}% (${W} / ${V})`}async function rf(){if(await X9("jq"))return!0;return process.stdout.write(`${L0}Error: jq is required but not installed.${v}
+var h_=Object.create;var{getPrototypeOf:v_,defineProperty:rK,getOwnPropertyNames:g_}=Object;var m_=Object.prototype.hasOwnProperty;function u_(Z){return this[Z]}var p_,d_,c_=(Z,X,Q)=>{var Y=Z!=null&&typeof Z==="object";if(Y){var J=X?p_??=new WeakMap:d_??=new WeakMap,z=J.get(Z);if(z)return z}Q=Z!=null?h_(v_(Z)):{};let K=X||!Z||!Z.__esModule?rK(Q,"default",{value:Z,enumerable:!0}):Q;for(let $ of g_(Z))if(!m_.call(K,$))rK(K,$,{get:u_.bind(Z,$),enumerable:!0});if(Y)J.set(Z,K);return K};var qQ=(Z,X)=>()=>(X||Z((X={exports:{}}).exports,X),X.exports);var l_=(Z)=>Z;function i_(Z,X){this[Z]=l_.bind(null,X)}var l0=(Z,X)=>{for(var Q in X)rK(Z,Q,{get:X[Q],enumerable:!0,configurable:!0,set:i_.bind(X,Q)})};var p=(Z,X)=>()=>(Z&&(X=Z(Z=0)),X);var e0=import.meta.require;var kO={};l0(kO,{lokiDir:()=>j0,homeLokiDir:()=>P4,findRepoRootForVersion:()=>eK,REPO_ROOT:()=>r0});import{resolve as n7,dirname as tK}from"path";import{fileURLToPath as a_}from"url";import{existsSync as GQ}from"fs";import{homedir as s_}from"os";function n_(){let Z=RO;for(let X=0;X<6;X++){if(GQ(n7(Z,"VERSION"))&&GQ(n7(Z,"autonomy/run.sh")))return Z;let Q=tK(Z);if(Q===Z)break;Z=Q}return n7(RO,"..","..","..")}function eK(Z){let X=Z;for(let Q=0;Q<6;Q++){if(GQ(n7(X,"VERSION"))&&GQ(n7(X,"autonomy/run.sh")))return X;let Y=tK(X);if(Y===X)break;X=Y}return n7(Z,"..","..","..")}function j0(){return process.env.LOKI_DIR??n7(process.cwd(),".loki")}function P4(){return n7(s_(),".loki")}var RO,r0;var G8=p(()=>{RO=tK(a_(import.meta.url));r0=n_()});import{readFileSync as o_}from"fs";import{resolve as r_,dirname as t_}from"path";import{fileURLToPath as e_}from"url";function f3(){if(h5!==null)return h5;let Z="8.34.0";if(typeof Z==="string"&&Z.length>0)return h5=Z,h5;try{let X=t_(e_(import.meta.url)),Q=eK(X);h5=o_(r_(Q,"VERSION"),"utf-8").trim()}catch{h5="unknown"}return h5}var h5=null;var HQ=p(()=>{G8()});var bO={};l0(bO,{runOrThrow:()=>Of,run:()=>E0,readStreamCapped:()=>UQ,commandVersion:()=>Af,commandExists:()=>X9,ShellError:()=>Z$,MAX_STDOUT_BYTES:()=>yO});async function UQ(Z,X=yO){let Q=Z.getReader(),Y=new TextDecoder,J="",z=0;try{while(z<X){let{done:K,value:$}=await Q.read();if(K)break;if(!$)continue;if(z+=$.byteLength,z>X){let W=$.byteLength-(z-X);J+=Y.decode($.subarray(0,W),{stream:!0});break}J+=Y.decode($,{stream:!0})}J+=Y.decode()}finally{try{await Q.cancel()}catch{}Q.releaseLock()}return J}async function E0(Z,X={}){let Q=Bun.spawn({cmd:[...Z],stdout:"pipe",stderr:"pipe",env:X.env?{...process.env,...X.env}:process.env,cwd:X.cwd}),Y,J;if(X.timeoutMs&&X.timeoutMs>0)Y=setTimeout(()=>{try{Q.kill("SIGTERM")}catch{}J=setTimeout(()=>{try{Q.kill("SIGKILL")}catch{}},2000)},X.timeoutMs);try{let[z,K,$]=await Promise.all([UQ(Q.stdout),new Response(Q.stderr).text(),Q.exited]);return{stdout:z,stderr:K,exitCode:$}}finally{if(Y)clearTimeout(Y);if(J)clearTimeout(J)}}async function Of(Z,X={}){let Q=await E0(Z,X);if(Q.exitCode!==0)throw new Z$(`command failed (${Q.exitCode}): ${Z.join(" ")}`,Q.exitCode,Q.stdout,Q.stderr);return Q}async function X9(Z){let X=Lf(Z),Q=await E0(["sh","-c",`command -v ${X}`],{timeoutMs:5000});if(Q.exitCode===0)return Q.stdout.trim()||null;return null}function Lf(Z){if(!/^[A-Za-z0-9._/-]+$/.test(Z))throw Error(`refused to shell-escape suspect token: ${Z}`);return Z}async function Af(Z,X="--version"){if(!await X9(Z))return null;let Y=await E0([Z,X],{timeoutMs:5000});if(Y.exitCode!==0)return null;return((Y.stdout||Y.stderr).split(/\r?\n/)[0]?.trim()??"")||null}var yO=16777216,Z$;var x9=p(()=>{Z$=class Z$ extends Error{message;exitCode;stdout;stderr;constructor(Z,X,Q,Y){super(Z);this.message=Z;this.exitCode=X;this.stdout=Q;this.stderr=Y;this.name="ShellError"}}});function o7(Z){return jf?"":Z}var jf,L0,k8,p0,ZV0,i0,H8,Q9,v;var S6=p(()=>{jf=(process.env.NO_COLOR??"").length>0;L0=o7("\x1B[0;31m"),k8=o7("\x1B[0;32m"),p0=o7("\x1B[1;33m"),ZV0=o7("\x1B[0;34m"),i0=o7("\x1B[0;36m"),H8=o7("\x1B[1m"),Q9=o7("\x1B[2m"),v=o7("\x1B[0m")});import{existsSync as xf}from"fs";async function E7(){if(k4!==void 0)return k4;let Z="/opt/homebrew/bin/python3.12";if(xf(Z))return k4=Z,Z;let X=await X9("python3.12");if(X)return k4=X,X;let Q=await X9("python3");return k4=Q,Q}async function Y7(Z,X={}){let Q=await E7();if(!Q)return{stdout:"",stderr:"python3 not found",exitCode:127};return E0([Q,"-c",Z],X)}var k4;var r7=p(()=>{x9()});var ZL={};l0(ZL,{runStatus:()=>Yh});import{existsSync as Y9,readFileSync as v3,readdirSync as iO,statSync as aO}from"fs";import{resolve as h8,basename as sf}from"path";import{homedir as nf}from"os";function sO(Z){let X=Math.trunc(Z);if(X>=1e6)return`${(Math.trunc(X/1e6*10)/10).toFixed(1)}M`;if(X>=1000)return`${(Math.trunc(X/1000*10)/10).toFixed(1)}K`;return String(X)}function nO(Z,X,Q){if(X===0)return null;let Y=Math.trunc(Z*100/X),J=Math.trunc(Z*NQ/X);if(J>NQ)J=NQ;let z=NQ-J,K=k8;if(Y>=80)K=L0;else if(Y>=50)K=p0;let $="=".repeat(Math.max(0,J))+" ".repeat(Math.max(0,z)),W=sO(Z),V=sO(X);return`  ${H8}${Q}${v} ${K}[${$}]${v} ${Y}% (${W} / ${V})`}async function rf(){if(await X9("jq"))return!0;return process.stdout.write(`${L0}Error: jq is required but not installed.${v}
 `),process.stdout.write(`Install with:
 `),process.stdout.write(`  brew install jq    (macOS)
 `),process.stdout.write(`  apt install jq     (Debian/Ubuntu)
@@ -1222,4 +1222,4 @@ Set LOKI_LEGACY_BASH=1 to force the bash CLI for every command.
 `),2}case"start":{let{runStart:Y}=await Promise.resolve().then(() => (b_(),y_));return Y(Q)}default:return process.stderr.write(`Unknown command: ${X}
 `),process.stderr.write(__),2}}cO();process.on("SIGINT",()=>process.exit(130));process.on("SIGTERM",()=>process.exit(143));var fW0=await _W0(Bun.argv.slice(2));process.exit(fW0);
 
-//# debugId=E73CE71EDB93AD7A64756E2164756E21
+//# debugId=651C52085A3FC1C964756E2164756E21

package/mcp/__init__.py

@@ -75,4 +75,4 @@ try:
 except ImportError:
     __all__ = ['mcp']
 
-__version__ = '8.32.0'
+__version__ = '8.34.0'

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "loki-mode",
   "mcpName": "io.github.asklokesh/loki-mode",
-  "version": "8.32.0",
+  "version": "8.34.0",
   "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": "8.32.0",
+  "version": "8.34.0",
   "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",