loki-mode 8.94.0 → 8.96.0

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 v8.94.0
+# Loki Mode v8.96.0
 
 **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
 
 ---
 
-**v8.94.0 | [Autonomi](https://www.autonomi.dev/) flagship product | ~410 lines core**
+**v8.96.0 | [Autonomi](https://www.autonomi.dev/) flagship product | ~410 lines core**

package/VERSION

@@ -1 +1 @@
-8.94.0
+8.96.0

package/dashboard/__init__.py

@@ -7,7 +7,7 @@ Modules:
     control: Session control API (start/stop/pause/resume)
 """
 
-__version__ = "8.94.0"
+__version__ = "8.96.0"
 
 # Expose the control app for easy import
 try:

package/loki-ts/dist/loki.js

@@ -1,5 +1,5 @@
 // @bun
-var m_=Object.create;var{getPrototypeOf:u_,defineProperty:eK,getOwnPropertyNames:p_}=Object;var d_=Object.prototype.hasOwnProperty;function c_(Z){return this[Z]}var l_,i_,a_=(Z,X,Q)=>{var Y=Z!=null&&typeof Z==="object";if(Y){var J=X?l_??=new WeakMap:i_??=new WeakMap,z=J.get(Z);if(z)return z}Q=Z!=null?m_(u_(Z)):{};let K=X||!Z||!Z.__esModule?eK(Q,"default",{value:Z,enumerable:!0}):Q;for(let $ of p_(Z))if(!d_.call(K,$))eK(K,$,{get:c_.bind(Z,$),enumerable:!0});if(Y)J.set(Z,K);return K};var HQ=(Z,X)=>()=>(X||Z((X={exports:{}}).exports,X),X.exports);var s_=(Z)=>Z;function n_(Z,X){this[Z]=s_.bind(null,X)}var l0=(Z,X)=>{for(var Q in X)eK(Z,Q,{get:X[Q],enumerable:!0,configurable:!0,set:n_.bind(X,Q)})};var p=(Z,X)=>()=>(Z&&(X=Z(Z=0)),X);var e0=import.meta.require;var kO={};l0(kO,{lokiDir:()=>j0,homeLokiDir:()=>R4,findRepoRootForVersion:()=>X$,REPO_ROOT:()=>i0});import{resolve as n7,dirname as Z$}from"path";import{fileURLToPath as o_}from"url";import{existsSync as UQ}from"fs";import{homedir as r_}from"os";function t_(){let Z=RO;for(let X=0;X<6;X++){if(UQ(n7(Z,"VERSION"))&&UQ(n7(Z,"autonomy/run.sh")))return Z;let Q=Z$(Z);if(Q===Z)break;Z=Q}return n7(RO,"..","..","..")}function X$(Z){let X=Z;for(let Q=0;Q<6;Q++){if(UQ(n7(X,"VERSION"))&&UQ(n7(X,"autonomy/run.sh")))return X;let Y=Z$(X);if(Y===X)break;X=Y}return n7(Z,"..","..","..")}function j0(){return process.env.LOKI_DIR??n7(process.cwd(),".loki")}function R4(){return n7(r_(),".loki")}var RO,i0;var H8=p(()=>{RO=Z$(o_(import.meta.url));i0=t_()});import{readFileSync as e_}from"fs";import{resolve as Zf,dirname as Xf}from"path";import{fileURLToPath as Qf}from"url";function h3(){if(h5!==null)return h5;let Z="8.94.0";if(typeof Z==="string"&&Z.length>0)return h5=Z,h5;try{let X=Xf(Qf(import.meta.url)),Q=X$(X);h5=e_(Zf(Q,"VERSION"),"utf-8").trim()}catch{h5="unknown"}return h5}var h5=null;var BQ=p(()=>{H8()});var bO={};l0(bO,{runOrThrow:()=>jf,run:()=>E0,readStreamCapped:()=>NQ,commandVersion:()=>Tf,commandExists:()=>X9,ShellError:()=>Q$,MAX_STDOUT_BYTES:()=>yO});async function NQ(Z,X=yO){let Q=Z.getReader(),Y=new TextDecoder,J="",z=0;try{while(z<X){let{done:K,value:$}=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([NQ(Q.stdout),new Response(Q.stderr).text(),Q.exited]);return{stdout:z,stderr:K,exitCode:$}}finally{if(Y)clearTimeout(Y);if(J)clearTimeout(J)}}async function jf(Z,X={}){let Q=await E0(Z,X);if(Q.exitCode!==0)throw new Q$(`command failed (${Q.exitCode}): ${Z.join(" ")}`,Q.exitCode,Q.stdout,Q.stderr);return Q}async function X9(Z){let X=Mf(Z),Q=await E0(["sh","-c",`command -v ${X}`],{timeoutMs:5000});if(Q.exitCode===0)return Q.stdout.trim()||null;return null}function Mf(Z){if(!/^[A-Za-z0-9._/-]+$/.test(Z))throw Error(`refused to shell-escape suspect token: ${Z}`);return Z}async function Tf(Z,X="--version"){if(!await X9(Z))return null;let Y=await E0([Z,X],{timeoutMs:5000});if(Y.exitCode!==0)return null;return((Y.stdout||Y.stderr).split(/\r?\n/)[0]?.trim()??"")||null}var yO=16777216,Q$;var x9=p(()=>{Q$=class Q$ 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 wf?"":Z}var wf,L0,F8,p0,zV0,a0,W8,Q9,v;var S6=p(()=>{wf=(process.env.NO_COLOR??"").length>0;L0=o7("\x1B[0;31m"),F8=o7("\x1B[0;32m"),p0=o7("\x1B[1;33m"),zV0=o7("\x1B[0;34m"),a0=o7("\x1B[0;36m"),W8=o7("\x1B[1m"),Q9=o7("\x1B[2m"),v=o7("\x1B[0m")});import{existsSync as bf}from"fs";async function E7(){if(x4!==void 0)return x4;let Z="/opt/homebrew/bin/python3.12";if(bf(Z))return x4=Z,Z;let X=await X9("python3.12");if(X)return x4=X,X;let Q=await X9("python3");return x4=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 x4;var r7=p(()=>{x9()});var ZL={};l0(ZL,{runStatus:()=>Kh});import{existsSync as Y9,readFileSync as g3,readdirSync as iO,statSync as aO}from"fs";import{resolve as h8,basename as rf}from"path";import{homedir as tf}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*LQ/X);if(J>LQ)J=LQ;let z=LQ-J,K=F8;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`  ${W8}${Q}${v} ${K}[${$}]${v} ${Y}% (${W} / ${V})`}async function Zh(){if(await X9("jq"))return!0;return process.stdout.write(`${L0}Error: jq is required but not installed.${v}
+var m_=Object.create;var{getPrototypeOf:u_,defineProperty:eK,getOwnPropertyNames:p_}=Object;var d_=Object.prototype.hasOwnProperty;function c_(Z){return this[Z]}var l_,i_,a_=(Z,X,Q)=>{var Y=Z!=null&&typeof Z==="object";if(Y){var J=X?l_??=new WeakMap:i_??=new WeakMap,z=J.get(Z);if(z)return z}Q=Z!=null?m_(u_(Z)):{};let K=X||!Z||!Z.__esModule?eK(Q,"default",{value:Z,enumerable:!0}):Q;for(let $ of p_(Z))if(!d_.call(K,$))eK(K,$,{get:c_.bind(Z,$),enumerable:!0});if(Y)J.set(Z,K);return K};var HQ=(Z,X)=>()=>(X||Z((X={exports:{}}).exports,X),X.exports);var s_=(Z)=>Z;function n_(Z,X){this[Z]=s_.bind(null,X)}var l0=(Z,X)=>{for(var Q in X)eK(Z,Q,{get:X[Q],enumerable:!0,configurable:!0,set:n_.bind(X,Q)})};var p=(Z,X)=>()=>(Z&&(X=Z(Z=0)),X);var e0=import.meta.require;var kO={};l0(kO,{lokiDir:()=>j0,homeLokiDir:()=>R4,findRepoRootForVersion:()=>X$,REPO_ROOT:()=>i0});import{resolve as n7,dirname as Z$}from"path";import{fileURLToPath as o_}from"url";import{existsSync as UQ}from"fs";import{homedir as r_}from"os";function t_(){let Z=RO;for(let X=0;X<6;X++){if(UQ(n7(Z,"VERSION"))&&UQ(n7(Z,"autonomy/run.sh")))return Z;let Q=Z$(Z);if(Q===Z)break;Z=Q}return n7(RO,"..","..","..")}function X$(Z){let X=Z;for(let Q=0;Q<6;Q++){if(UQ(n7(X,"VERSION"))&&UQ(n7(X,"autonomy/run.sh")))return X;let Y=Z$(X);if(Y===X)break;X=Y}return n7(Z,"..","..","..")}function j0(){return process.env.LOKI_DIR??n7(process.cwd(),".loki")}function R4(){return n7(r_(),".loki")}var RO,i0;var H8=p(()=>{RO=Z$(o_(import.meta.url));i0=t_()});import{readFileSync as e_}from"fs";import{resolve as Zf,dirname as Xf}from"path";import{fileURLToPath as Qf}from"url";function h3(){if(h5!==null)return h5;let Z="8.96.0";if(typeof Z==="string"&&Z.length>0)return h5=Z,h5;try{let X=Xf(Qf(import.meta.url)),Q=X$(X);h5=e_(Zf(Q,"VERSION"),"utf-8").trim()}catch{h5="unknown"}return h5}var h5=null;var BQ=p(()=>{H8()});var bO={};l0(bO,{runOrThrow:()=>jf,run:()=>E0,readStreamCapped:()=>NQ,commandVersion:()=>Tf,commandExists:()=>X9,ShellError:()=>Q$,MAX_STDOUT_BYTES:()=>yO});async function NQ(Z,X=yO){let Q=Z.getReader(),Y=new TextDecoder,J="",z=0;try{while(z<X){let{done:K,value:$}=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([NQ(Q.stdout),new Response(Q.stderr).text(),Q.exited]);return{stdout:z,stderr:K,exitCode:$}}finally{if(Y)clearTimeout(Y);if(J)clearTimeout(J)}}async function jf(Z,X={}){let Q=await E0(Z,X);if(Q.exitCode!==0)throw new Q$(`command failed (${Q.exitCode}): ${Z.join(" ")}`,Q.exitCode,Q.stdout,Q.stderr);return Q}async function X9(Z){let X=Mf(Z),Q=await E0(["sh","-c",`command -v ${X}`],{timeoutMs:5000});if(Q.exitCode===0)return Q.stdout.trim()||null;return null}function Mf(Z){if(!/^[A-Za-z0-9._/-]+$/.test(Z))throw Error(`refused to shell-escape suspect token: ${Z}`);return Z}async function Tf(Z,X="--version"){if(!await X9(Z))return null;let Y=await E0([Z,X],{timeoutMs:5000});if(Y.exitCode!==0)return null;return((Y.stdout||Y.stderr).split(/\r?\n/)[0]?.trim()??"")||null}var yO=16777216,Q$;var x9=p(()=>{Q$=class Q$ 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 wf?"":Z}var wf,L0,F8,p0,zV0,a0,W8,Q9,v;var S6=p(()=>{wf=(process.env.NO_COLOR??"").length>0;L0=o7("\x1B[0;31m"),F8=o7("\x1B[0;32m"),p0=o7("\x1B[1;33m"),zV0=o7("\x1B[0;34m"),a0=o7("\x1B[0;36m"),W8=o7("\x1B[1m"),Q9=o7("\x1B[2m"),v=o7("\x1B[0m")});import{existsSync as bf}from"fs";async function E7(){if(x4!==void 0)return x4;let Z="/opt/homebrew/bin/python3.12";if(bf(Z))return x4=Z,Z;let X=await X9("python3.12");if(X)return x4=X,X;let Q=await X9("python3");return x4=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 x4;var r7=p(()=>{x9()});var ZL={};l0(ZL,{runStatus:()=>Kh});import{existsSync as Y9,readFileSync as g3,readdirSync as iO,statSync as aO}from"fs";import{resolve as h8,basename as rf}from"path";import{homedir as tf}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*LQ/X);if(J>LQ)J=LQ;let z=LQ-J,K=F8;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`  ${W8}${Q}${v} ${K}[${$}]${v} ${Y}% (${W} / ${V})`}async function Zh(){if(await X9("jq"))return!0;return process.stdout.write(`${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)
@@ -1232,4 +1232,4 @@ Set LOKI_LEGACY_BASH=1 to force the bash CLI for every command.
 `),2}case"start":{let{runStart:Y}=await Promise.resolve().then(() => (h_(),f_));return Y(Q)}default:return process.stderr.write(`Unknown command: ${X}
 `),process.stderr.write(v_),2}}cO();process.on("SIGINT",()=>process.exit(130));process.on("SIGTERM",()=>process.exit(143));var uW0=await mW0(Bun.argv.slice(2));process.exit(uW0);
 
-//# debugId=8BB2CE8D47221C8564756E2164756E21
+//# debugId=3B5AE9CBA3FA521064756E2164756E21

package/mcp/__init__.py

@@ -75,4 +75,4 @@ try:
 except ImportError:
     __all__ = ['mcp']
 
-__version__ = '8.94.0'
+__version__ = '8.96.0'

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "loki-mode",
   "mcpName": "io.github.asklokesh/loki-mode",
-  "version": "8.94.0",
+  "version": "8.96.0",
   "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": "8.94.0",
+  "version": "8.96.0",
   "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",

package/tools/cost-history.py

@@ -0,0 +1,317 @@
+#!/usr/bin/env python3
+"""Track agent cost across MANY runs and report the trend.
+
+cost-guard.py gates ONE run against a ceiling. receipt-diff.py compares TWO
+runs. Neither answers the question a team actually asks at the end of a month:
+is our agent spend trending up? That needs a durable history, so this appends
+one line per run to a JSONL file and reports over the whole series.
+
+THE FIVE PROPERTIES THAT MAKE THE TREND WORTH BELIEVING:
+
+1. AN UNMEASURED RUN IS NEVER RECORDED AS 0. It is appended with usd=null and
+   excluded from the trend. Recording zero would drag the median down and
+   understate real spend -- the exact "$0.00 means free" lie this repo spent
+   thirteen surfaces removing. Whether a number counts as measured is
+   record_is_measured() in autonomy/lib/efficiency_cost.py, imported and never
+   restated; receipts store the figure under cost.usd while the predicate reads
+   cost_usd, so the KEY is mapped here. Mapping a key name is not restating the
+   rule.
+
+   RECORD-NULL rather than REFUSE, deliberately. Refusing is cost-guard.py's
+   job and it already does it, per run, with an exit code. Here the history IS
+   the product, and a refused append leaves no trace: a month with four broken
+   instrumentation runs would look identical to a month with eight clean ones.
+   The null row is countable, so the report can say "12 runs, 8 measured" and
+   the operator can see the measurement gap instead of inferring it from a
+   short file.
+
+2. A TREND FROM ONE POINT IS NOT A TREND. Fewer than 2 measured runs reads
+   INSUFFICIENT DATA. "Flat" is a claim about change over time and one
+   observation cannot support it.
+
+3. APPEND IS ATOMIC PER LINE. See _append().
+
+4. A CORRUPT LINE IS COUNTED AND REPORTED, never silently skipped. A history
+   that quietly drops rows reports a cleaner trend than reality, and drops them
+   most often when something upstream is broken -- precisely when the number
+   matters.
+
+5. AN EMPTY HISTORY IS NOT A FLAT TREND. It exits non-zero. Zero runs is an
+   absent measurement, not evidence of stability.
+
+Usage:
+  tools/cost-history.py record [workspace] [--file .loki/cost-history.jsonl]
+  tools/cost-history.py report [--file ...] [--json]
+
+Exit: 0 report produced / run recorded, 1 nothing to report (empty or no
+measured runs), 2 cannot record.
+"""
+
+import argparse
+import json
+import os
+import sys
+import time
+
+_HERE = os.path.dirname(os.path.abspath(__file__))
+sys.path.insert(0, os.path.join(os.path.dirname(_HERE), "autonomy", "lib"))
+
+from efficiency_cost import collect_efficiency, record_is_measured  # noqa: E402
+
+OK, NOTHING, CANNOT = 0, 1, 2
+
+DEFAULT_FILE = os.path.join(".loki", "cost-history.jsonl")
+
+# Below this the two halves are the same number to the cent, and calling that
+# a direction is noise dressed as signal.
+FLAT_PCT = 5.0
+
+
+def _num(v):
+    """A number as itself; None, "", or a bool as None."""
+    if isinstance(v, bool) or not isinstance(v, (int, float)):
+        return None
+    return v
+
+
+def measured_usd(cost):
+    """The USD figure from a cost block, or None when it was never measured.
+
+    `available` is NOT trusted on its own: a real receipt shipped
+    available=true with every field zero (v8.52.0), so the flag is a claim and
+    the VALUES are the evidence.
+    """
+    if not isinstance(cost, dict):
+        return None
+    rec = {
+        "cost_usd": _num(cost.get("usd")),
+        "input_tokens": _num(cost.get("input_tokens")),
+        "output_tokens": _num(cost.get("output_tokens")),
+        "cache_read_tokens": _num(cost.get("cache_read_tokens")),
+        "cache_creation_tokens": _num(cost.get("cache_creation_tokens")),
+    }
+    if not record_is_measured(rec):
+        return None
+    return _num(rec["cost_usd"])
+
+
+def _loki_dir(workspace):
+    """Accept either a workspace root or a .loki dir; collect_ wants .loki."""
+    if os.path.basename(os.path.normpath(workspace)) == ".loki":
+        return workspace
+    return os.path.join(workspace, ".loki")
+
+
+def _append(path, entry):
+    """Append ONE newline-terminated line with ONE write() call, in "a" mode.
+
+    WHY THAT IS SUFFICIENT HERE, rather than the usual write-temp-and-rename.
+    POSIX gives an O_APPEND write a seek-to-end that cannot be interleaved, so
+    concurrent appenders cannot overwrite each other, and a line small enough
+    to land in a single write is not torn between two of them. Rename-based
+    atomicity would be strictly WORSE for this file: it rewrites the whole
+    history every time, which turns a crash into the loss of every prior run
+    instead of a partial final line.
+
+    That partial final line is the residual risk and it is handled rather than
+    prevented: a torn or truncated row is COUNTED as corrupt by load() and
+    named in the report. This is why the newline terminator is load-bearing --
+    it is what makes a complete row distinguishable from a truncated one.
+
+    ponytail: no locking. O_APPEND is the lock. Add one only if this ever needs
+    to write multi-line entries, which would break the single-write property.
+    """
+    parent = os.path.dirname(os.path.abspath(path))
+    if parent:
+        os.makedirs(parent, exist_ok=True)
+    line = json.dumps(entry, sort_keys=True) + "\n"
+    with open(path, "a", encoding="utf-8") as handle:
+        handle.write(line)
+
+
+def load(path):
+    """Read the history. Returns (entries, corrupt_count).
+
+    A line that is not JSON, or is JSON but not an object, is corrupt and
+    COUNTED. It is never dropped on the floor.
+    """
+    entries, corrupt = [], 0
+    try:
+        with open(path, "r", encoding="utf-8") as handle:
+            raw = handle.read()
+    except OSError:
+        return None, 0
+    for line in raw.splitlines():
+        if not line.strip():
+            continue
+        try:
+            obj = json.loads(line)
+        except ValueError:
+            corrupt += 1
+            continue
+        if not isinstance(obj, dict) or "usd" not in obj:
+            corrupt += 1
+            continue
+        entries.append(obj)
+    return entries, corrupt
+
+
+def median(values):
+    """Median of a non-empty list. Even length averages the middle pair."""
+    s = sorted(values)
+    n = len(s)
+    mid = n // 2
+    return s[mid] if n % 2 else (s[mid - 1] + s[mid]) / 2.0
+
+
+def direction(values):
+    """(label, explanation) for a series oldest to newest.
+
+    Compares the median of the older half against the median of the newer
+    half, not first-vs-last: a single expensive outlier at either end should
+    not name the trend for a whole quarter. With 2 or 3 points the halves are
+    small, so the comparison it used is always printed alongside the verdict --
+    the operator can see how thin the evidence is.
+    """
+    if len(values) < 2:
+        return "INSUFFICIENT DATA", (
+            "%d measured run(s); a trend needs at least 2. One observation "
+            "cannot show change over time, and 'flat' would be a claim."
+            % len(values))
+    half = len(values) // 2
+    older = median(values[:half])
+    newer = median(values[len(values) - half:])
+    basis = ("median of the oldest %d ($%.4f) vs the newest %d ($%.4f)"
+             % (half, older, half, newer))
+    if older == 0:
+        if newer == 0:
+            return "flat", basis + "; both zero"
+        return "rising", basis + "; percent change is undefined against zero"
+    pct = (newer - older) / older * 100.0
+    if abs(pct) < FLAT_PCT:
+        return "flat", "%s; %+.1f%%, within the %.0f%% flat band" % (
+            basis, pct, FLAT_PCT)
+    return ("rising" if pct > 0 else "falling"), "%s; %+.1f%%" % (basis, pct)
+
+
+def record(workspace, path, now=None):
+    """Append this workspace's run to the history. Returns a verdict dict."""
+    cost, model = collect_efficiency(_loki_dir(workspace))
+    usd = measured_usd(cost)
+    entry = {
+        "ts": now if now is not None else time.time(),
+        "workspace": os.path.abspath(workspace),
+        "usd": usd,          # None when unmeasured. NEVER 0 as a stand-in.
+        "measured": usd is not None,
+        "model": model or None,
+    }
+    try:
+        _append(path, entry)
+    except OSError as exc:
+        return {"status": "cannot_record", "exit_code": CANNOT,
+                "why": "could not append to %s: %s" % (path, exc),
+                "entry": None}
+    return {"status": "recorded", "exit_code": OK, "why": None,
+            "entry": entry}
+
+
+def report(path):
+    """Summarize the history. Returns a verdict dict."""
+    entries, corrupt = load(path)
+    if entries is None:
+        return {"status": "no_history", "exit_code": NOTHING,
+                "why": "no history file at %s; record a run first." % path,
+                "runs": 0, "measured": 0, "corrupt_lines": 0,
+                "costs": [], "median_usd": None,
+                "direction": "INSUFFICIENT DATA", "basis": None}
+
+    costs = [e["usd"] for e in entries
+             if e.get("measured") and _num(e.get("usd")) is not None]
+
+    if not entries:
+        # An empty file is not a flat trend. Zero runs is an absent
+        # measurement, and reporting stability from it would be a claim made
+        # out of nothing.
+        return {"status": "empty", "exit_code": NOTHING,
+                "why": "history %s is empty (%d corrupt line(s)); zero runs "
+                       "is not a flat trend." % (path, corrupt),
+                "runs": 0, "measured": 0, "corrupt_lines": corrupt,
+                "costs": [], "median_usd": None,
+                "direction": "INSUFFICIENT DATA", "basis": None}
+
+    label, basis = direction(costs)
+    return {
+        "status": "ok" if costs else "no_measured_runs",
+        "exit_code": OK if costs else NOTHING,
+        "why": None if costs else (
+            "%d run(s) recorded but none carried a measured cost; there is "
+            "nothing to trend. Unmeasured runs are kept as null, not 0."
+            % len(entries)),
+        "runs": len(entries),
+        "measured": len(costs),
+        "corrupt_lines": corrupt,
+        "costs": costs,
+        "median_usd": median(costs) if costs else None,
+        "direction": label,
+        "basis": basis,
+    }
+
+
+def render(d):
+    if d["status"] in ("no_history", "empty"):
+        return "NO TREND: %s" % d["why"]
+    if d["status"] == "cannot_record":
+        return "CANNOT RECORD: %s" % d["why"]
+    if d["status"] == "recorded":
+        e = d["entry"]
+        return ("recorded %s: %s" % (
+            e["workspace"],
+            "$%.4f" % e["usd"] if e["measured"]
+            else "UNMEASURED (stored as null, excluded from the trend)"))
+
+    lines = ["%d run(s), %d measured" % (d["runs"], d["measured"])]
+    if d["corrupt_lines"]:
+        # Named, not swallowed. A dropped row makes the trend look cleaner
+        # than reality.
+        lines.append("%d CORRUPT line(s) in the history -- counted, not "
+                     "skipped; the trend below omits them."
+                     % d["corrupt_lines"])
+    if d["status"] == "no_measured_runs":
+        lines.append("NO TREND: %s" % d["why"])
+        return "\n".join(lines)
+    lines.append("costs oldest to newest: "
+                 + ", ".join("$%.4f" % c for c in d["costs"]))
+    lines.append("median: $%.4f" % d["median_usd"])
+    lines.append("direction: %s (%s)" % (d["direction"], d["basis"]))
+    return "\n".join(lines)
+
+
+def main(argv=None):
+    ap = argparse.ArgumentParser(
+        description="Track agent cost across many runs and report the trend.")
+    sub = ap.add_subparsers(dest="cmd", required=True)
+
+    rec = sub.add_parser("record", help="append this run's cost to the history")
+    rec.add_argument("workspace", nargs="?", default=".",
+                     help="workspace root (or its .loki dir); default .")
+    rec.add_argument("--file", default=DEFAULT_FILE,
+                     help="history JSONL (default %s)" % DEFAULT_FILE)
+
+    rep = sub.add_parser("report", help="summarize the recorded history")
+    rep.add_argument("--file", default=DEFAULT_FILE,
+                     help="history JSONL (default %s)" % DEFAULT_FILE)
+    rep.add_argument("--json", action="store_true", dest="as_json",
+                     help="emit the report as JSON")
+
+    args = ap.parse_args(argv)
+    if args.cmd == "record":
+        d = record(args.workspace, args.file)
+        print(render(d))
+    else:
+        d = report(args.file)
+        print(json.dumps(d, indent=2) if args.as_json else render(d))
+    return d["exit_code"]
+
+
+if __name__ == "__main__":
+    sys.exit(main())

package/tools/model-advisor.py

@@ -484,10 +484,21 @@ def main(argv=None):
     ap.add_argument("--json", action="store_true")
     args = ap.parse_args(argv)
 
+    # A workspace that does not EXIST is not the same fact as a workspace with
+    # no cost history, and exit 0 collapsed them. A CI job doing
+    # `model-advisor.py "$WS" && ...` on a mistyped or unmounted path saw green
+    # and carried on. Every sibling tool distinguishes these (run-replay 66,
+    # cost-guard 2, receipt-bundle 3); this one did not.
+    if not os.path.isdir(args.workspace):
+        sys.stderr.write(
+            "cannot advise: workspace does not exist: %s\n" % args.workspace)
+        return 66
+
     adv = advise(args.workspace)
     print(json.dumps(adv, indent=2) if args.json else render(adv))
-    # Exit 0 either way: "no basis" is a successful, honest answer, not a tool
-    # failure. Callers read has_basis.
+    # Exit 0 for a real workspace with no basis: that IS a successful, honest
+    # answer rather than a tool failure, and the output says so in words.
+    # Callers read has_basis.
     return 0
 
 

package/tools/receipt-attest.py

@@ -334,11 +334,50 @@ def render(record):
     return "\n".join(lines)
 
 
+_USAGE = """usage: receipt-attest.py <proof.json> [--json]
+
+Turn a receipt into a portable ATTESTATION a third party can check without the
+original workspace.
+
+  --json    emit the attestation record as JSON
+  --help    show this message
+
+States, which are never collapsed into one another:
+  VERIFIED      the axis was checked and passed
+  FAILED        the axis was checked and failed
+  UNVERIFIABLE  the axis could NOT be checked here, with the reason
+
+Exit: 0 all scored axes verified, 1 something FAILED, 2 something was
+UNVERIFIABLE, 64 usage error.
+"""
+
+
 def main(argv):
-    args = [a for a in argv[1:] if a != "--json"]
-    as_json = "--json" in argv[1:]
+    rest = argv[1:]
+
+    # --help must print usage, NOT be read as a filename. It previously fell
+    # through to the positional slot and produced
+    # "UNVERIFIABLE -- proof file not found: --help", which is a verification
+    # verdict about a file the user never named: a fabricated answer to a
+    # question they did not ask.
+    if "--help" in rest or "-h" in rest:
+        sys.stdout.write(_USAGE)
+        return 0
+
+    as_json = "--json" in rest
+    args = [a for a in rest if a != "--json"]
+
+    # An unrecognized flag is an ERROR, never a path. Treating "--jsonn" as a
+    # proof file reports UNVERIFIABLE for a typo, which reads as a finding
+    # about the receipt rather than about the command line.
+    unknown = [a for a in args if a.startswith("-")]
+    if unknown:
+        sys.stderr.write("unknown option(s): %s\n" % " ".join(unknown))
+        sys.stderr.write(_USAGE)
+        return 64
+
     if len(args) != 1:
-        sys.stderr.write("usage: receipt-attest.py <proof.json> [--json]\n")
+        sys.stderr.write(_USAGE)
         return 64
     record = attest(args[0])
     print(json.dumps(record, indent=2, sort_keys=True) if as_json