@dzhechkov/skills-feature-adr 1.5.13 → 1.5.15

package/.dz-manifest.json

@@ -5,7 +5,7 @@
     "files": [
       {
         "path": "CHANGELOG.md",
-        "sha256": "88e0389016f697b90731ee8743a35e523f0ccb079426d59861d894f33e516392"
+        "sha256": "26fa03f89862d03ebfc6447c866cb5d4dcbd5591ae4f8fb3a3ea08b6c586e19d"
       },
       {
         "path": "LICENSE",
@@ -13,7 +13,7 @@
       },
       {
         "path": "README.md",
-        "sha256": "619a231352a602c736ce6eec1e5fa79a056dc2f5f80f6b35eed236f9670f7337"
+        "sha256": "66f3b2cf7fdcf9efbf64c7d5cc36b34ffc82b151a9bc07e6fda5fc47d4fba72e"
       },
       {
         "path": "bin/cli.js",
@@ -25,7 +25,7 @@
       },
       {
         "path": "package.json",
-        "sha256": "07537ebbd424c668a37a42b670c033a4d3134a9e7e13803f426aa31149a85b58"
+        "sha256": "e74b1a22e0260a61a3ab37512366402c4e6231b85625028a86219ee3f5c21e12"
       },
       {
         "path": "src/cli.js",
@@ -37,7 +37,7 @@
       },
       {
         "path": "src/commands/init.js",
-        "sha256": "fbc3853435048c5b632119bf37cc702c4c305cae7de656548891295f6b9e8947"
+        "sha256": "07ff9d955422979ea67aa5aa454c13303f5ae644efb32e5cc6f359a509638347"
       },
       {
         "path": "src/commands/list.js",
@@ -113,7 +113,7 @@
       },
       {
         "path": "templates/.claude/skills/feature-adr/SKILL.md",
-        "sha256": "67eb730c815e6b8f9b9b61084c67ceb85440548f00b21f52187d16e81e3d7af8"
+        "sha256": "28c265603481f976a52ea4c627e11194f3c0283801ea93d02cbb9f4248b43c68"
       },
       {
         "path": "templates/.claude/skills/feature-adr/examples/sample-feature-output.md",
@@ -149,15 +149,15 @@
       },
       {
         "path": "templates/.claude/skills/feature-adr/modules/06-implementation-plan.md",
-        "sha256": "5c8d4c79d5329702b8c75b9afb36ae335aa63c99e74045c0afa98e12658ee975"
+        "sha256": "d7a2a4e56b24451da234b1ac4ef440ba9b0370617e1c0cde8ad5f3d575a53209"
       },
       {
         "path": "templates/.claude/skills/feature-adr/modules/07-code.md",
-        "sha256": "fb054591e86d55cee184eab96b5f0440053ada8fa9edf619e4a66ec5ac371587"
+        "sha256": "db79f8d026cc47edd1e5d2f30d0e1455c55c570dd022d549d440985a3e5940a5"
       },
       {
         "path": "templates/.claude/skills/feature-adr/modules/08-qe.md",
-        "sha256": "7ab9b7b1523289986a1f51b4cbbebf4410d96ad4a153f22fa8613e8a0f4f5fa7"
+        "sha256": "d7a967926feec9b0ec173ac7042ae3abd368644d8e1dba9a81a770794417fcb3"
       },
       {
         "path": "templates/.claude/skills/feature-adr/modules/09-fleet-qe.md",
@@ -247,9 +247,13 @@
         "path": "templates/.claude/skills/feature-adr/references/qe-checklist.md",
         "sha256": "238d8896996dc53559f58b24aa7eca966cc8845c3b9dda6d982c717c657b91e4"
       },
+      {
+        "path": "templates/.claude/skills/feature-adr/scripts/build-coder-context.mjs",
+        "sha256": "19c3991bfb5e88205c9ff6d4d44037c136a190581bab199d3928196e4616fe15"
+      },
       {
         "path": "templates/.claude/skills/feature-adr/scripts/check-plan-completeness.mjs",
-        "sha256": "08e69d2e63350fe2dc64483226fcd30901b77a8e930ca6280ef71181c67675b4"
+        "sha256": "8b93949ce4f671d932c5389050db3a3e2750efcbed69681d69f392ccf4d2a168"
       },
       {
         "path": "templates/.claude/skills/feature-adr/scripts/markdown-masker.mjs",
@@ -317,7 +321,7 @@
       },
       {
         "path": "templates/.claude/workflows/feature-adr.js",
-        "sha256": "e96c5280ad21b604036cc912418c9627b0ff0ae8ca1436c58dcff951ba9a034b"
+        "sha256": "7928cc90575eaef4f4626849f4490631676bc4ce6eb80e8f94c99d9e215bbd77"
       },
       {
         "path": "templates/lib/memory-protocol.md",
@@ -329,5 +333,5 @@
       }
     ]
   },
-  "signature": "CfTJgsBbekrZE+IbzaQFLofuRDZ3B2CJTqPk6gjeOiLXhChXiPGkqxaWKCT+hdQTy6ORB7fu6vfMx37rwJf6Cg=="
+  "signature": "wSJpaUNRsFXAD5Ye7V5y8FPgcvrLDEq6peQpK/RadbFsD518KglZDn5GNzV4DJE6i+mkHbH9tIwr+HIYXKN4BA=="
 }

package/CHANGELOG.md

@@ -1,5 +1,19 @@
 # Changelog
 
+## [Unreleased]
+
+### Fixed — `init --force` больше не стирает локальную правку молча
+
+- Из трёх путей записи `init --force` был ЕДИНСТВЕННЫМ, который перезаписывал локально
+  изменённый файл без резервной копии и без строки в отчёте — при том, что баннер успеха
+  рекомендует именно эту команду. `update` сохраняет правку по трёхсторонней сверке,
+  `update --force` кладёт `.bak` с первого дня; расходился только `init --force`.
+- Теперь `init --force` копирует изменённый файл в `<file>.bak` ПЕРЕД перезаписью и называет
+  его в отчёте. Копия делается только если байты ОТЛИЧАЮТСЯ от шаблона: `.bak`, совпадающий
+  с шаблоном, — чистый мусор. `--dry-run --force` называет будущую копию и ничего не пишет.
+- Поведение самого `--force` не смягчено: файл по-прежнему перезаписывается, это его смысл.
+  Менялось только то, что правка перестала исчезать бесследно.
+
 ## [1.5.1] - 2026-08-21
 
 ### Changed — the Step-8 amendment gate is a COMMAND, and the durable writers are witnessed

package/README.md

@@ -1,5 +1,9 @@
 # @dzhechkov/skills-feature-adr
 
+Current package version: `1.5.15`. <!-- dz:version -->
+
+Site: https://aicoding.space · Source: https://github.com/djd1m/dz-harness/tree/main/packages/@dzhechkov/skills-feature-adr
+
 **Spec-Driven Development pipeline for AI coding agents (Claude Code, Codex, …)**
 
 An 11-step, complexity-routed pipeline that makes an AI coding agent build a feature the way a
@@ -37,6 +41,21 @@ npx @dzhechkov/skills-feature-adr init
 
 After installation, open Claude Code in your project directory and use `/feature-adr`.
 
+Plain usage guidance joins the existing stage writer to `usage --by-stage --project` with explicit FA/
+Wf source selection and observed receipt IDs. It preserves unknown splits/prices, caller estimates and
+separate conservation/inventory/source verification. No billing inference, new ledger or paid replay.
+
+Plain Step 8 bridge guidance now passes current `--round`, `--round-run` and `--task` from the existing
+round receipt with the execution `--project`. Explicit conflicts refuse before reviewer work; the
+bridge's invocation `runId` stays distinct from pipeline identity. Native Workflow QE remains its own
+review path. Historical window correlation is disclosed as lower assurance, without guessed identity.
+
+Step 7 uses the installed `scripts/build-coder-context.mjs` helper to include literal requirements,
+plan tasks and ADR Decision/Confirmation. Workflow reads current inputs before code checkpoint
+lookup; plain coding runs the same helper and reads or embeds its successful `promptBlock`.
+Missing required sections, invalid files and exceeded UTF-8 bounds refuse coding instead of trimming
+the context. Existing decision recall and code-wrapper routing remain in place.
+
 ---
 
 ## What You Get
@@ -63,7 +82,9 @@ npx @dzhechkov/skills-feature-adr init               # Install core components
 npx @dzhechkov/skills-feature-adr init --with-learning                  # + reward learning
 npx @dzhechkov/skills-feature-adr init --knowledge-extractor            # + knowledge extractor
 npx @dzhechkov/skills-feature-adr init --with-learning --knowledge-extractor  # + both
-npx @dzhechkov/skills-feature-adr init --force       # Overwrite existing files
+npx @dzhechkov/skills-feature-adr init --force       # Overwrite existing files (a locally
+                                                     # changed file is copied to <file>.bak first,
+                                                     # and the copy is named in the report)
 npx @dzhechkov/skills-feature-adr init --dry-run     # Preview without making changes
 npx @dzhechkov/skills-feature-adr update             # Update to latest version
 npx @dzhechkov/skills-feature-adr remove             # Clean uninstall
@@ -106,6 +127,13 @@ ARCHITECTURE → IMPLEMENTATION →  CODE   →   QE    →  FLEET QE
 # Full protocols + 6 extra skills, up to 7 fleet QE agents
 ```
 
+### Checkpoint reads have one source (v1.5.14)
+
+`loadCheckpoints` in the bundled `feature-adr.js` now builds its read command with the checkpoints blob's own
+`checkpointReadCmd(FDIR)` instead of a hand-written duplicate, so the helper with the speaking name is the one the pipeline
+runs. Behaviour is unchanged: for directory names with spaces, quotes and missing directories the shell output is
+byte-identical (proved against the old command). A wiring test fails if the duplicate is re-inlined.
+
 ### Advisory micro-recall at two decision points (v1.5.9, staged)
 
 The workflow makes one bounded decision-local recall attempt immediately before the live Step 3
@@ -1354,3 +1382,22 @@ harness-core's `src/markdown-masker.ts`. It runs without a core build. Amendment
 and K2 share the parser while retaining their existing unclosed-block and indentation policies.
 The four-space indented-code gap remains open for amendment checks and K2; swarm briefs retain their
 existing masking of indented code. Versions are unchanged in this staged change.
+
+### Codex companion for feature-adr
+
+`dz statusline --watch --project "/path/to/worktree" --brain "/path/to/shared-brain"
+--slug "feature-slug" --run-id "stable-run-id"` adds an explicitly launched adjacent terminal
+companion, including Plain runs. Until installed, invoke the worktree-built
+`node packages/@dzhechkov/harness-cli/dist/bin.js statusline --watch ...`. This extends the existing
+command inventory. Canonical feature-adr guidance supplies the quoted producer/observer recipe: record
+a stable run ID and actual tier at the START of each step, then `done` on real completion; recall/teach
+remain scoped to the shared brain. Project run state and brain counts are separate. One slug retains
+one latest run; report freshness is not process liveness and stage position is not passed gates.
+
+The readonly companion requires a dedicated stdout TTY, uses serial 2-second refreshes (0.25–60
+allowed), sanitizes/clips text and handles resize. Below 40x8 it shows a size warning. Ctrl-C/SIGTERM
+exit 0; output failure 1; invalid/piped watch 2. Failed/absent counts and optional values are explicit;
+watch v1 ETA is unavailable and global source inventory omitted. No stdin/raw mode, models, logical
+store writes or automatic terminal/settings changes. Normal SQLite ephemeral WAL/SHM sidecars are
+permitted. One-shot Claude text/JSON/ETA remain unchanged. Codex native footer capability is not
+asserted; parity names manual `dz statusline --watch` access.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@dzhechkov/skills-feature-adr",
-  "version": "1.5.13",
+  "version": "1.5.15",
   "description": "Adaptive Feature Development skill pack for Claude Code — 11-step pipeline with Complexity Router (S/M/L/XL), ADR-driven architecture, 15 agentic-qe skills, multi-agent fleet QE. Supports --full-qe, --full-qe-extended, --with-learning, and --knowledge-extractor modes.",
   "bin": {
     "skills-feature-adr": "./bin/cli.js"
@@ -49,7 +49,7 @@
     "url": "git+https://github.com/djd1m/dz-harness.git",
     "directory": "packages/@dzhechkov/skills-feature-adr"
   },
-  "homepage": "https://github.com/djd1m/dz-harness/tree/main/packages/@dzhechkov/skills-feature-adr#readme",
+  "homepage": "https://aicoding.space",
   "bugs": {
     "url": "https://github.com/djd1m/dz-harness/issues"
   },

package/sbom.json

@@ -15,7 +15,7 @@
       "hashes": [
         {
           "alg": "SHA-256",
-          "content": "88e0389016f697b90731ee8743a35e523f0ccb079426d59861d894f33e516392"
+          "content": "26fa03f89862d03ebfc6447c866cb5d4dcbd5591ae4f8fb3a3ea08b6c586e19d"
         }
       ]
     },
@@ -35,7 +35,7 @@
       "hashes": [
         {
           "alg": "SHA-256",
-          "content": "619a231352a602c736ce6eec1e5fa79a056dc2f5f80f6b35eed236f9670f7337"
+          "content": "66f3b2cf7fdcf9efbf64c7d5cc36b34ffc82b151a9bc07e6fda5fc47d4fba72e"
         }
       ]
     },
@@ -69,7 +69,7 @@
         },
         {
           "name": "dz:canonical-json-sha256-v2",
-          "value": "07537ebbd424c668a37a42b670c033a4d3134a9e7e13803f426aa31149a85b58"
+          "value": "e74b1a22e0260a61a3ab37512366402c4e6231b85625028a86219ee3f5c21e12"
         }
       ]
     },
@@ -99,7 +99,7 @@
       "hashes": [
         {
           "alg": "SHA-256",
-          "content": "fbc3853435048c5b632119bf37cc702c4c305cae7de656548891295f6b9e8947"
+          "content": "07ff9d955422979ea67aa5aa454c13303f5ae644efb32e5cc6f359a509638347"
         }
       ]
     },
@@ -289,7 +289,7 @@
       "hashes": [
         {
           "alg": "SHA-256",
-          "content": "67eb730c815e6b8f9b9b61084c67ceb85440548f00b21f52187d16e81e3d7af8"
+          "content": "28c265603481f976a52ea4c627e11194f3c0283801ea93d02cbb9f4248b43c68"
         }
       ]
     },
@@ -379,7 +379,7 @@
       "hashes": [
         {
           "alg": "SHA-256",
-          "content": "5c8d4c79d5329702b8c75b9afb36ae335aa63c99e74045c0afa98e12658ee975"
+          "content": "d7a2a4e56b24451da234b1ac4ef440ba9b0370617e1c0cde8ad5f3d575a53209"
         }
       ]
     },
@@ -389,7 +389,7 @@
       "hashes": [
         {
           "alg": "SHA-256",
-          "content": "fb054591e86d55cee184eab96b5f0440053ada8fa9edf619e4a66ec5ac371587"
+          "content": "db79f8d026cc47edd1e5d2f30d0e1455c55c570dd022d549d440985a3e5940a5"
         }
       ]
     },
@@ -399,7 +399,7 @@
       "hashes": [
         {
           "alg": "SHA-256",
-          "content": "7ab9b7b1523289986a1f51b4cbbebf4410d96ad4a153f22fa8613e8a0f4f5fa7"
+          "content": "d7a967926feec9b0ec173ac7042ae3abd368644d8e1dba9a81a770794417fcb3"
         }
       ]
     },
@@ -623,13 +623,23 @@
         }
       ]
     },
+    {
+      "type": "file",
+      "name": "templates/.claude/skills/feature-adr/scripts/build-coder-context.mjs",
+      "hashes": [
+        {
+          "alg": "SHA-256",
+          "content": "19c3991bfb5e88205c9ff6d4d44037c136a190581bab199d3928196e4616fe15"
+        }
+      ]
+    },
     {
       "type": "file",
       "name": "templates/.claude/skills/feature-adr/scripts/check-plan-completeness.mjs",
       "hashes": [
         {
           "alg": "SHA-256",
-          "content": "08e69d2e63350fe2dc64483226fcd30901b77a8e930ca6280ef71181c67675b4"
+          "content": "8b93949ce4f671d932c5389050db3a3e2750efcbed69681d69f392ccf4d2a168"
         }
       ]
     },
@@ -799,7 +809,7 @@
       "hashes": [
         {
           "alg": "SHA-256",
-          "content": "e96c5280ad21b604036cc912418c9627b0ff0ae8ca1436c58dcff951ba9a034b"
+          "content": "7928cc90575eaef4f4626849f4490631676bc4ce6eb80e8f94c99d9e215bbd77"
         }
       ]
     },

package/src/commands/init.js

@@ -54,14 +54,15 @@ function showKeysariumIntegration(keysariumManifest) {
 
 // Copies a component file-by-file with per-file overwrite protection:
 //   - destination missing            -> write, record in `written`
-//   - destination exists + --force   -> overwrite, record in `written`
+//   - destination exists + --force   -> BACK UP to a .bak sibling if the bytes differ,
+//                                       then overwrite; record in `written` (+ `backedUp`)
 //   - destination exists, no --force -> do NOT write, record in `preserved`
 // In dry-run mode nothing is written, but the same written/preserved
 // classification is produced.
 // Returns { missing, fileCount } where `missing` means the template source
 // was absent on disk and `fileCount` is how many files the template provides.
 function installComponent(key, comp, templatesDir, targetDir, opts) {
-  const { force, dryRun, written, preserved, hashes } = opts;
+  const { force, dryRun, written, preserved, backedUp, hashes } = opts;
   const src = path.join(templatesDir, comp.src);
   const destRoot = path.join(targetDir, comp.src);
 
@@ -89,12 +90,20 @@ function installComponent(key, comp, templatesDir, targetDir, opts) {
   }
 
   for (const entry of entries) {
-    if (fileExists(entry.destFile) && !force) {
+    const existed = fileExists(entry.destFile);
+    if (existed && !force) {
       preserved.push(entry.rel);
       continue;
     }
+    // `update --force` has backed edits up to a `.bak` sibling since day one; `init --force` was
+    // the ONE path that overwrote silently — and the success banner recommends exactly that
+    // command, so a locally-edited workflow could vanish with no notice and no copy.
+    // Only DIFFERING bytes are backed up: a `.bak` identical to the template is pure litter.
+    const differs = existed && hashFile(entry.destFile) !== hashFile(entry.srcFile);
+    if (differs && backedUp) backedUp.push(entry.rel);
     if (!dryRun) {
       ensureDir(path.dirname(entry.destFile));
+      if (differs) fs.copyFileSync(entry.destFile, `${entry.destFile}.bak`);
       fs.copyFileSync(entry.srcFile, entry.destFile);
       // Record the SHA-256 of the TEMPLATE bytes we just installed (not the
       // dest) as this file's baseline. This makes baseline == mine immediately
@@ -110,6 +119,23 @@ function installComponent(key, comp, templatesDir, targetDir, opts) {
   return { missing: false, fileCount: entries.length };
 }
 
+// Print the block of locally-changed files that were (or would be) backed up before --force
+// overwrote them. Silence here is what the field report caught: the operator had no way to learn
+// that their edit was gone, let alone where the copy is.
+function printBackedUpBlock(backedUp, dryRun) {
+  if (backedUp.length === 0) return;
+  const MAX_SHOWN = 10;   // `printPreservedBlock` держит свою копию — она объявлена в его теле
+  console.log('');
+  const verb = dryRun ? 'would be backed up' : 'backed up';
+  warn(`${backedUp.length} locally-changed file(s) ${verb} to a .bak sibling before being overwritten:`);
+  for (const rel of backedUp.slice(0, MAX_SHOWN)) {
+    console.log(dim(`  ${rel} -> ${rel}.bak`));
+  }
+  if (backedUp.length > MAX_SHOWN) {
+    console.log(dim(`  …and ${backedUp.length - MAX_SHOWN} more`));
+  }
+}
+
 // Print the block of pre-existing files that were (or would be) preserved
 function printPreservedBlock(preserved, dryRun) {
   console.log('');
@@ -233,6 +259,7 @@ async function run(options) {
   const installedFiles = [];   // files written (or would-write in dry-run)
   const installedHashes = {};  // rel -> sha256 of the TEMPLATE bytes installed (baseline)
   const preservedFiles = [];   // pre-existing files NOT overwritten (no --force)
+  const backedUpFiles = [];    // locally-changed files copied to .bak before --force overwrote them
   const completedKeys = [];    // component keys processed so far (for partial manifest)
   const installedOptionalKeys = [];
   let stepNum = 0;
@@ -246,7 +273,7 @@ async function run(options) {
 
       installComponent(key, comp, templatesDir, targetDir, {
         force, dryRun, written: installedFiles, preserved: preservedFiles,
-        hashes: installedHashes,
+        backedUp: backedUpFiles, hashes: installedHashes,
       });
       completedKeys.push(key);
     }
@@ -262,7 +289,7 @@ async function run(options) {
 
       const res = installComponent(key, comp, templatesDir, targetDir, {
         force, dryRun, written: installedFiles, preserved: preservedFiles,
-        hashes: installedHashes,
+        backedUp: backedUpFiles, hashes: installedHashes,
       });
       if (!res.missing && res.fileCount > 0) {
         installedOptionalKeys.push(key);
@@ -296,6 +323,10 @@ async function run(options) {
   if (dryRun) {
     console.log('');
     info(`Dry run: ${installedFiles.length} file(s) would be written, ${preservedFiles.length} pre-existing file(s) would be preserved.`);
+    // Отчёт о копиях стоит СНАРУЖИ стража сохранённых: при --force сохранять нечего, список
+    // пуст, и внутри стража блок был бы недостижим ровно в том случае, ради которого написан.
+    // Пустой список функция отсекает сама.
+    printBackedUpBlock(backedUpFiles, true);
     if (preservedFiles.length > 0) {
       printPreservedBlock(preservedFiles, true);
     }
@@ -318,10 +349,11 @@ async function run(options) {
     process.exit(0);
   }
 
+  printBackedUpBlock(backedUpFiles, false);   // снаружи: при --force preservedFiles пуст
   if (preservedFiles.length > 0) {
     printPreservedBlock(preservedFiles, false);
-    console.log('');
   }
+  if (preservedFiles.length > 0 || backedUpFiles.length > 0) console.log('');
 
   // ── f) Write manifest ──────────────────────────────────────────────────
   const pkgPath = path.resolve(__dirname, '../../package.json');

package/templates/.claude/skills/feature-adr/SKILL.md

@@ -562,13 +562,91 @@ only the statusline:
 
 **Record the panel at the START of every step, not only at Steps 0/8/9.** The panel shows the last step
 that reported; a pipeline that reports three times per run shows a stale step for most of its life. Emit
-`dz statusline --fa-record --slug <slug> --step "<Step N Name>" --recalled <n> --stored <n>` as the first
+`dz statusline --fa-record --project "<worktree>" --slug "<slug>" --run-id "<stable-run-id>" --tier M --step "<Step N Name>" --recalled <n> --stored <n>` as the first
 action of each step. The recall/teach counts only change at Steps 0/8/9; the *step label* changes at every
 one of them.
 
 *Honesty note:* the panel is live only insofar as the pipeline records state — it reflects what the
 pipeline actually did with the loop (recalls that ran, stores that landed), not an aspirational count.
 
+### Plain observed usage receipts
+
+Use the existing witnessed writer for usage you actually observed. At a real stage boundary,
+capture the execution project, stable run/task, verbatim stage, actual model/family/role, attempt,
+tier/mode and source window/IDs available to this host. Missing fields remain null with their reason;
+do not infer input/cache from total, price from a model family, or tokens from invocation budgets.
+
+```bash
+dz feature-adr-record --kind ledger --stage "$CURRENT_STAGE" --project "$EXECUTION_PROJECT" \
+  --row "$OBSERVED_STAGE_ROW_JSON" --rollout-id "$ACTUAL_SESSION_ID" --turn-id "$ACTUAL_TURN_ID" --json
+dz usage --by-stage --project "$EXECUTION_PROJECT" --source fa-ledger --run "$STABLE_RUN_ID" --json
+```
+
+`OBSERVED_STAGE_ROW_JSON` is real host metadata: `runId`, `taskId`, `stage`, `model`, `family`, `role`,
+`attempt`, `tier`, `mode`, `tokens`/dimensions when observed, and optional separate `estimate` with
+tokens/costUsd/method/source/capturedAt. Omit selectors not known; window/cwd/model-only correlation
+is labelled legacy-window and cannot claim exact source verification. Source roots can be named with
+`--codex-sessions`. Exact IDs are validated against existing receipts; no new IDs or recall/teach occur.
+The source receipt subset is captured once; later source append cannot enlarge the old row.
+Captured payload integrity and every pricing-bearing dimension must match the original scoped source.
+A reported monetary amount belongs to one observation, not each expanded token receipt. Preserve an
+actual observation ID/scope/basis in optional `reportedCostObservation: { id, scope, basis }` within
+`--row` when known; otherwise a captured source scope supplies a stable identity and unrelated money
+attribution stays unavailable. Reimports of the same observation count once; conflicting amounts are
+diagnosed. Missing money differs from zero. Exports must avoid every selected authoritative source,
+including custom run directories and symlink aliases. Invalid Claude counters retain nulls/diagnostics;
+a declared invalid total cannot derive a replacement, and accounting validity does not rewrite the
+actual generation outcome.
+
+For Wf, use `--source workflow-budget --run <id>` and optional `--run-dir <dir>` for its existing
+budget/trace/state. Auto source collisions require an explicit source; joined Wf summary projections
+never add another cost. Preserve reported total basis and cache/reasoning subsets. Wf budget.spent
+counts dispatch units; native Workflow's existing budget delta is output-only. Neither is raw total.
+
+Reports separate conservation, expected inventory and independent amount verification. Without
+a same-scope witness, verified totals are null even when reported values conserve. Unknown rates or
+split keep primary estimated USD null; static family estimates, provider-reported USD and caller
+pre-run estimates are separate, never current exact prices or billed amounts. Billing remains unobserved.
+
+### Codex companion terminal (Plain included)
+
+Open an adjacent terminal or a manual tmux split and launch the observer explicitly. Codex does not
+have a dz native command-provider footer. Keep the producer's project, slug and stable run ID equal to
+the observer's; use the actual complexity tier at the START of every active step. Brain is the shared
+learning store for recall/teach, while project is the worktree containing run slots and branch.
+
+```bash
+# Set these to your actual absolute paths. Use the worktree build until the change is installed.
+PANEL_CLI="/path/to/worktree/packages/@dzhechkov/harness-cli/dist/bin.js"
+PANEL_PROJECT="/path/to/worktree"
+PANEL_BRAIN="/path/to/canonical-brain"
+PANEL_SLUG="feature-slug"
+PANEL_RUN="feature-20261002-1"  # choose once per invocation, retain at every step
+node "$PANEL_CLI" statusline --watch --project "$PANEL_PROJECT" --brain "$PANEL_BRAIN" \
+  --slug "$PANEL_SLUG" --run-id "$PANEL_RUN" --interval 2
+# In the producer terminal, at the START of each real step (example: tier M):
+node "$PANEL_CLI" statusline --fa-record --project "$PANEL_PROJECT" --slug "$PANEL_SLUG" \
+  --run-id "$PANEL_RUN" --tier M --step "Step 7 Code" --recalled 3 --stored 0
+# Only after the run actually completes; this is not a claim that QE passed:
+node "$PANEL_CLI" statusline --fa-record --project "$PANEL_PROJECT" --slug "$PANEL_SLUG" \
+  --run-id "$PANEL_RUN" --tier M --step "done" --recalled 3 --stored 0
+```
+
+Use `dz recall ... --project "$PANEL_BRAIN"` and `dz teach ... --project "$PANEL_BRAIN"` for actual
+learning. Supply measured cumulative counters, not the example numbers. The observer counts its brain
+source directly; the slot's producer pool is not a shared-brain inventory. One slug holds one latest
+run, so a replacement makes an exact old selection missing. No selectors means visibly automatic
+selection. Freshness measures producer-report age, not process liveness: fresh <30 minutes, stale
+30–<90, expired >=90; completed remains completed. Stage position is not completed gates.
+
+Watch v1 displays ETA unavailable, unknown optional values and unavailable failed/absent learning
+sources, and omits ambiguous global source inventory. It uses escaped ASCII text, a dedicated stdout
+TTY, 0.25–60 second intervals (default 2), and a size warning below 40 columns/8 rows. Ctrl-C/SIGTERM
+exit 0; output failures exit 1; piped output and watch+JSON/install/record combinations exit 2.
+Observation never reads stdin, runs models or writes logical store state. SQLite-managed ephemeral
+WAL/SHM files are permitted; no application locks, repair or schema changes occur. One-shot Claude
+statusline/JSON and its ETA keep their existing behavior. Do not start a terminal automatically.
+
 ### What changes with `--full-qe`
 
 Full agentic-qe protocols for the same 9 core skills. No new agents, just deeper methodology.

package/templates/.claude/skills/feature-adr/modules/06-implementation-plan.md

@@ -235,6 +235,16 @@ C2 recognises JS/TS, pytest, Go, Rust, JVM and .NET test paths, extensible per p
 
 Never proceed on a non-zero exit, and never treat empty output as a pass — the last line
 (`K2 plan-completeness: PASS|FAIL|NOT-ESTABLISHED`) is the verdict, and its absence is not one.
+
+**Where an id counts (the default since 2026-09-27, owner decision).** C1 and C8 read the plan's
+TASK LINES only: a heading, a list item or a table row. An `ADR-<n>` or `FR-<n>` that appears only
+in a prose paragraph, a fenced code block (the SPARC-GOAP ```yaml goal state included), an HTML
+comment, the `## Amendments` section or the `EXPECTED_CODE_TARGETS:` block is a mention, not a task,
+and the gate FAILs it with `(cited only outside task lines)`. Write the id on the FIRST line of the
+task that implements it: a wrapped continuation line of a list item is not read either. Measured on the archive when the default changed: 56 of 102 plans that had
+passed would fail this reader, so a plan written before that date may be red on a re-check — move
+the citation onto a task line, or re-check that one plan with `--no-require-task-lines` and say so
+in the checkpoint banner.
 What it checks: C1 every ADR **decision** (a `# ADR-NNN` / `## ADR-NNN` heading INSIDE the file, not
 just the filename prefix — a file with several headings owes several plan citations) has a plan task
 citing it · C2 every ADR Confirmation test path is named in the plan · C3 the `EXPECTED_CODE_TARGETS:`

package/templates/.claude/skills/feature-adr/modules/07-code.md

@@ -26,6 +26,35 @@ opus (complex code generation)
 
 ### 1. Pre-Implementation Checklist
 
+### Current literal context (plain and delegated coding)
+
+Before coding or delegating, run the installed helper beside this module. Set `CONTEXT_HELPER` to
+the absolute `scripts/build-coder-context.mjs` path of the skill installation you are reading;
+set `FEATURE_DIR` to the absolute target `features/<slug>` directory and use the actual tier:
+
+```bash
+node "$CONTEXT_HELPER" "$FEATURE_DIR" --tier=M
+```
+
+The command emits exactly one JSON envelope and exits 0 only for `status: "complete"`. On any
+nonzero exit, unavailable/incomplete status, malformed JSON or required missing/empty section,
+stop before coding and repair the named input. Never paste a partial result as complete. A missing
+helper requires restoring this skill installation, not inventing a replacement block.
+
+For every delegated coder assignment, paste the successful envelope's literal `promptBlock` into
+the actual prompt, then append the existing advisory decision-recall block once. Keep the source
+paths below for deeper reading. For single-agent in-session coding, read this generated block
+directly before implementing. Requirements and plan are included in full; each ADR supplies its
+Decision and Confirmation with source labels. M/L/XL require at least one ADR; S can have none.
+
+The same canonical helper is used by the programmatic Workflow before code checkpoint lookup.
+It fingerprints full current inputs and binds the prompt separately, so changed documents cannot
+reuse old code. The plain mode boundary is an executable helper command plus these required
+read/embedding instructions; there is no separately automated plain dispatcher. Pure/fixture tests
+do not establish a live model relay's authenticity. Helper bounds are 64 documents, 256 KiB/file,
+1 MiB read and 96 KiB UTF-8 for the entire labelled block; exceeded bounds refuse without trimming.
+These are document limits, not a new limit on the existing coder wrapper's final prompt.
+
 Before writing any code:
 - [ ] Read existing similar implementations in codebase
 - [ ] Identify naming conventions (files, classes, functions, variables)

package/templates/.claude/skills/feature-adr/modules/08-qe.md

@@ -23,6 +23,30 @@ sonnet (analytical evaluation)
 
 ## Protocol
 
+### Bridge identity for plain Step 8
+
+When the plain host invokes the existing Claude review bridge, take round number, pipeline run and
+task from the current round receipt/state in the execution project. Pass the fields actually present:
+
+```bash
+dz qe-bridge --family claude --slug "$FEATURE_SLUG" --project "$EXECUTION_PROJECT" \
+  --round "$CURRENT_ROUND" --round-run "$CURRENT_ROUND_RUN" --task "$CURRENT_TASK" \
+  --coder-family codex
+```
+
+Do not mint missing identifiers or point `--project` at a separate learning brain. Explicit identity
+must agree with one readable open round before any probe or review child; conflicts refuse with exit 2.
+The bridge freezes that snapshot for the signoff. Signoff `roundRun` is the pipeline run; its existing
+`runId` identifies the separate bridge invocation and audit filenames. No-flags standalone reviews
+retain best-effort lookup and visibly unbound provenance when authority is absent or unavailable.
+
+Round close validates every present identity field before selection. Full, partial and genuinely
+identity-free window assurance appear as `round-run-task`, `partial-identity` and `legacy-window`.
+Malformed or foreign claims cannot become a manual-close fallback. This contract governs the
+existing plain bridge and control-review caller; native Workflow QE already runs its own review and
+must not invoke a second bridge to manufacture this receipt. Local fake-child tests do not establish
+a live Claude model roundtrip or secure every native Codex QE receipt.
+
 ### 1. Smoke Tests (All tiers)
 
 ```
@@ -205,7 +229,9 @@ list), grep for unfinished-stub markers: `TODO` / `FIXME` / `HACK` / `XXX` / `PL
 file:line, UNLESS the line carries an inline `no-stubs: <reason>` waiver WITH a non-empty reason, or
 `.dz/guard.json` `stubWaivers` lists the path WITH a reason. A REASONLESS waiver is itself a HIGH gap,
 never an exemption. Cross-check mechanically: `dz guard check --op publish --json` runs the same scan
-as the SOFT `no-stubs` rule over the working-tree diff. When you QUOTE a marker in `08_qe_report.md`,
+as the SOFT `no-stubs` rule over the working-tree diff.
+After the Step-7 code has landed, run `dz guard check --op code --json` and treat a HARD `block` verdict
+as a HIGH finding naming the drifted file. When you QUOTE a marker in `08_qe_report.md`,
 backtick it so the report itself scans clean (the claim-check forbidden-phrase convention). Record the
 verdict in the ADR Fitness section.
 
@@ -311,6 +337,9 @@ Compile all findings into a structured report:
 ✅ READY FOR MERGE | ❌ NEEDS FIXES | ⚠️ CONDITIONAL APPROVAL
 ```
 
+Settle a cross-family re-QE debt with `dz reqe --slug <s> --done --report <08b>`.
+For `dz reqe --done`, exit 3 means the review settled but named BLOCKER/HIGH findings — stop and surface them to the owner.
+
 ### 7.1 Findings ledger (machine-readable)
 
 Prose is for people; `dz score`/`dz recap` need a machine-readable surface too (qe-findings-record,
@@ -373,9 +402,13 @@ resume guard at all before this).
 Otherwise — QE ran fresh in this pass — after `08_qe_report.md` is written and the grade is final, run:
 
 ```bash
-dz feature-adr-record --kind ledger --stage qe --slug <slug> --row '<json>' --auto --json
+dz feature-adr-record --kind ledger --stage qe --slug <slug> --row '<json>' --json
 ```
 
+fix-round-1 (Codex r1 HIGH finding 4, ADR-001 D5): no `--auto` here. `--auto` is the trusted marker
+of an AUTOMATED pipeline run and requires an experiment envelope that the manual path does not have;
+this plain-mode row is written without it.
+
 `<json>` carries the same fields the ultracode pipeline writes for this row: `reviewer` (the model
 that reviewed, or `null` if unknown — never guessed), `reviewerFamily` (`claude`|`codex`|`null` when
 the reviewer identity is not one of the two known families — never guessed as `claude`), `qeRole`