@dzhechkov/p-replicator 1.10.3 → 1.12.0

package/templates/.claude/rules/feature-lifecycle.md

@@ -89,13 +89,22 @@ After 3 retries with 🔴, halt and surface to user.
 **Strategy:** maximum parallelism via `Task` tool.
 
 1. Identify independent work units from Phase 1's Architecture
-2. Spawn one Task per unit
-3. Each Task: read SPARC sections + implement + test + commit
-4. Coordinator merges/integrates after all Tasks complete
+2. Allocate one `RUN_ID`, a unique `WORK_UNIT_ID`, and an absolute `TRACE_PATH` per unit
+3. Spawn one Task per unit; each writes its substantive trace before its one-line pointer
+4. Coordinator validates every trace before merge/integration
 5. Run full test suite
 
 **Quality gate:** tests pass, lint clean, build succeeds.
 
+### Positive file receipt (required)
+
+Each worker must write a substantive body ending in `Status: completed` or `Status: failed` to its unique
+`TRACE_PATH` before returning a one-line report. Before integration, the coordinator verifies every
+path is a regular non-symlink file, non-whitespace, post-launch, and terminal. Narrative output or
+silence is not a receipt. Name missing, stale, partial, unreadable, duplicate, failed, dead-PID, or
+probe-error units and refuse merge/completion unless every required receipt is valid and completed.
+See [`swarm-file-evidence`](./swarm-file-evidence.md) for the write protocol and bounded exception.
+
 ## Phase 4: REVIEW (brutal-honesty-review)
 
 **Skill:** `.claude/skills/brutal-honesty-review/SKILL.md`
@@ -118,6 +127,8 @@ After 3 retries with 🔴, halt and surface to user.
 | Tests + lint + build | 3 | Re-run, max 3 attempts |
 | No `blocker` findings | 4 | Loop Phase 4 until clean |
 
+Place each gate on the strongest reliable enforcement layer; see [`cost-of-detection-ladder`](./cost-of-detection-ladder.md).
+
 ## Commit Discipline
 
 - After Phase 1: `docs(<feature>): SPARC plan`

package/templates/.claude/rules/honest-configuration.md

@@ -0,0 +1,54 @@
+# Honest Configuration
+
+## Rule
+
+A value that controls external output, access, limits, routing, or an authoritative measurement must
+not become a plausible or permissive result when its meaning is absent or unproven. Refuse or expose
+unknown; never manufacture health.
+
+## Mechanics
+
+### Substitution axis
+
+Absence is not permission to invent a runtime value. Validate required values at the boundary and
+derive degradation from the value actually obtained, even when no failure-only sentinel was set.
+
+### Interpretation axis
+
+Keep `undefined` distinct from `''`. Validate empty, misspelled, unmapped, and authority-derived values
+against a closed set in versioned code. Environment variables may select a code-owned variant; they
+must not define the allowlist. A declared input that no decision reads is fail-open-by-omission.
+
+| Case | Observable signal | Required response |
+|---|---|---|
+| CFG-S1 | Required runtime value is absent | REFUSE and name the external consequence; no plausible default. |
+| CFG-S2 | Obtained value is invalid although the failure sentinel is unset | REFUSE from the obtained value. |
+| CFG-I1 | Value is `undefined` | REFUSE or UNKNOWN; preserve the absent state. |
+| CFG-I2 | Value is the empty string `''` | REFUSE; do not collapse it into `undefined` or unrestricted. |
+| CFG-I3 | Variant is misspelled, unknown, or unmapped | REFUSE; list the code-owned recognized variants. |
+| CFG-I4 | Source of truth is unreachable | UNKNOWN or REFUSE; do not use cached permissive meaning. |
+| CFG-I5 | Declared allowlist/config input is never read by the decision | REFUSE and wire the decision to the input. |
+| CFG-I6 | Empty CIDR becomes `/0`, allowlist is empty, `BASE_URL='/'`, or tariff is unmapped | REFUSE; no unlimited access or plausible output. |
+| CFG-I7 | Ratio denominator is zero (`0/0`) | UNAVAILABLE or UNKNOWN; render empty with the reason, never `0%`. |
+| CFG-I8 | Allowlist or recognized-variant universe comes from environment | CODE-OWNED closed set; environment selects only. |
+
+## Bounded exception
+
+A named build phase may use a substitute only when that phase cannot emit or publish the external
+result; generic “non-production” is not a boundary. An optional dependency may fall back only when
+its absence cannot alter the governed external output, access, limit, route, or measurement.
+
+## Observable violation → replacement
+
+| Observable violation | Required replacement |
+|---|---|
+| Missing URL/project name becomes localhost, `/`, or an inferred name | Validate at the boundary and refuse with the affected output named. |
+| Empty/unknown access or limit value becomes unrestricted | Reject before the access, routing, or limiting decision. |
+| Failure is logged but the invalid obtained value still emits a healthy result | Compute status from the value/outcome and stop the emitting action. |
+| Undefined measurement renders as numeric zero | Render unavailable/empty and preserve why it could not be measured. |
+
+## Self-check
+
+For each governed value, enumerate absent, empty, invalid, unknown, and unreachable states beside one
+explicit valid control. Trace the value into the decision that emits output. If any bad state reaches
+a default, `ALLOW`, unlimited behavior, `/0`, or `0%` for `0/0`, the boundary is fail-open.

package/templates/.claude/rules/insights-capture.md

@@ -1,7 +1,7 @@
 # Insights Capture Rules
 
 When and how to record development "грабли" (rakes) into the project knowledge
-base. Used by `/myinsights` and the `SessionStart` hook
+base. Used by `/myinsights` and the `SessionStart`/`UserPromptSubmit` hook
 (`.claude/hooks/session-insights.cjs`).
 
 ## When to Capture
@@ -53,11 +53,16 @@ Use lowercase, hyphenated, specific tags:
 
 Aim for 2-5 tags per entry.
 
-## Auto-Injection on SessionStart
+## Prompt-Time Injection on UserPromptSubmit
 
-The `SessionStart` hook runs `node .claude/hooks/session-insights.cjs` which
-prints recent insights to stdout. Claude Code captures stdout and injects
-it into the initial session context.
+Markdown at `.claude/insights/index.md` remains the source of truth. After it is
+established, capture makes a best-effort idempotent `dz teach` duplicate; optional dz
+failure cannot undo the file write.
+
+On `UserPromptSubmit`, a successful non-empty `dz recall` result from the insight
+domain is the only state that suppresses local output. Absent, failing, or empty recall
+uses the local fallback of the three most recent Markdown entries, never both sources.
+`SessionStart` retains only the missing-carrier hint.
 
 ## Storage Lifecycle
 

package/templates/.claude/rules/replicate-pipeline.md

@@ -154,8 +154,10 @@ are project-agnostic and can be enhanced (read by Phase 3) but never recreated.
 **Agents (4):** `replicate-coordinator`, `product-discoverer`, `doc-validator`,
 `harvest-coordinator`
 
-**Rules (6):** `replicate-pipeline`, `skill-interface-protocol`, `git-workflow`,
-`insights-capture`, `feature-lifecycle`, `docker-ports`
+**Rules (9):** `replicate-pipeline`, `skill-interface-protocol`, `git-workflow`,
+`insights-capture`, `feature-lifecycle`, `docker-ports`,
+[`cost-of-detection-ladder`](cost-of-detection-ladder.md), `swarm-file-evidence`,
+`honest-configuration`
 
 **Hooks (8 files in `.claude/hooks/`, cross-platform Node).** Only four are wired to an
 event in `.claude/settings.json`; the rest are utilities you invoke deliberately, and the

package/templates/.claude/rules/skill-interface-protocol.md

@@ -73,6 +73,7 @@ Skills with `modules/` subdirectories MUST structure each module with these sect
 
 Modules are numbered (`01-name.md`, `02-name.md`) to indicate execution order.
 SKILL.md acts as orchestrator, referencing modules in sequence.
+Place module checks on the strongest reliable enforcement layer; see [`cost-of-detection-ladder`](./cost-of-detection-ladder.md).
 
 ## 5. Dependency Declaration
 

package/templates/.claude/rules/swarm-file-evidence.md

@@ -0,0 +1,46 @@
+# Swarm File Evidence
+
+## Rule
+
+Every parallel work unit has a named file result. A narrative reply is only a pointer; silence is
+neither progress nor completion. The coordinator may aggregate only positive, attributable terminal
+receipts from the assigned files.
+
+## Mechanics
+
+1. Before dispatch, allocate a run-unique `RUN_ID` and a unique `WORK_UNIT_ID`. Resolve one absolute
+   `TRACE_PATH` per `(RUN_ID, WORK_UNIT_ID)`, record its pre-launch state, and pass both fields to the
+   worker. Two workers never share a path.
+2. The worker writes a substantive Markdown body to a temporary regular file in the same directory,
+   appends exactly `Status: completed` or `Status: failed` as the final line, renames it to
+   `TRACE_PATH`, then returns a one-line pointer. The terminal marker is written last.
+3. Before merge, synthesis, or completion, the coordinator checks each assigned path: absolute and
+   unique; regular and non-symlink; readable and non-whitespace; absent before launch or observably
+   changed after launch; final line terminal. It reads the file as the payload and reports
+   `valid receipts / required receipts` with every failed `WORK_UNIT_ID` and path.
+4. `Status: completed` permits consumption. `Status: failed` is a delivered failure and blocks a
+   successful aggregate. Missing, empty, stale, partial, unreadable, duplicate, or probe-error
+   evidence is undelivered or inconclusive. A dead PID stops waiting as failure; a live PID may only
+   extend waiting. Neither PID state proves delivery.
+
+## Bounded exception
+
+If atomic rename is unavailable, write directly to `TRACE_PATH` and append the terminal marker last;
+until that line exists the file is partial. Host-authoritative liveness may extend a deadline, but it
+cannot replace the file result or turn missing evidence into success.
+
+## Observable violation → replacement
+
+| Observable violation | Required replacement |
+|---|---|
+| Task report exists but `TRACE_PATH` does not | Name the unit/path, mark undelivered, and refuse aggregation. |
+| File is empty, stale, symlinked, unreadable, or non-terminal | Keep the evidence out of the aggregate and rerun or diagnose that unit. |
+| Status says `running` but its recorded PID is dead | Close the unit as failed; do not report continued work from silence. |
+| Fewer than all required receipts are terminal-completed | Report the partial ratio and refuse completion. |
+
+## Self-check
+
+For every parallel unit, point to its assignment containing `WORK_UNIT_ID` and absolute `TRACE_PATH`,
+then point to the coordinator check performed before aggregation. Exercise missing, empty, stale,
+partial, failed, dead-PID, and probe-error traces; only a fresh substantive file ending in
+`Status: completed` may satisfy delivery.

package/templates/.claude/settings.json

@@ -1,6 +1,6 @@
 {
   "$schema": "https://json.schemastore.org/claude-code-settings.json",
-  "_comment": "Default hooks + statusline shipped by @dzhechkov/p-replicator init. Cross-platform Node scripts (no bash dependencies). Auto-commits roadmap/insights/plans on Stop, injects relevant insights on SessionStart. Statusline displays pipeline+roadmap+toolkit dashboard above the prompt. Project-specific hooks (DDD, fitness functions, etc.) are merged in by /replicate Phase 3.",
+  "_comment": "Default hooks + statusline shipped by @dzhechkov/p-replicator init. Cross-platform Node scripts (no bash dependencies). Auto-commits roadmap/insights/plans on Stop, keeps missing-carrier visibility on SessionStart, and selects insights on UserPromptSubmit. Statusline displays pipeline+roadmap+toolkit dashboard above the prompt. Project-specific hooks (DDD, fitness functions, etc.) are merged in by /replicate Phase 3.",
   "statusLine": {
     "type": "command",
     "command": "node \"${CLAUDE_PROJECT_DIR}/.claude/hooks/statusline.cjs\""
@@ -18,6 +18,18 @@
         ]
       }
     ],
+    "UserPromptSubmit": [
+      {
+        "matcher": "*",
+        "hooks": [
+          {
+            "type": "command",
+            "command": "node \"${CLAUDE_PROJECT_DIR}/.claude/hooks/session-insights.cjs\"",
+            "timeout": 5
+          }
+        ]
+      }
+    ],
     "Stop": [
       {
         "matcher": "*",

package/templates/.claude/skills/knowledge-extractor/modules/01-agent-review.md

@@ -18,6 +18,16 @@ snippets, skills, and hooks that could be reused in other projects.
 Launch 5 parallel agents via Task tool. Each agent independently scans the codebase
 from its perspective.
 
+### Positive file receipt contract
+
+Before dispatch, allocate one `RUN_ID` and give every extraction perspective a unique `WORK_UNIT_ID`
+and absolute `TRACE_PATH`. Each agent must write a substantive raw-findings body ending in
+`Status: completed` or `Status: failed` to `TRACE_PATH` before its one-line pointer. Before the
+Merging Strategy, verify each path is a regular non-symlink file, non-whitespace, post-launch, and
+terminal. Narrative output or silence is not a receipt. Name missing, stale, partial, unreadable,
+duplicate, failed, dead-PID, or probe-error perspectives and refuse synthesis/completion unless every
+required receipt is valid and completed. See `.claude/rules/swarm-file-evidence.md`.
+
 ### Agent: extractor-patterns
 
 **Scope:** Architecture and code patterns
@@ -166,11 +176,12 @@ from its perspective.
 
 After all 5 agents complete:
 
-1. **Deduplicate** — same finding reported by multiple agents → keep richest description
-2. **Cross-reference** — if a pattern is also a template, note both categories
-3. **Sort by confidence** — HIGH reusability first
-4. **Count** — total findings, per-category breakdown
-5. **Include TOOLKIT_HARVEST.md markers** — merge manual markers with auto-discovered
+1. **Validate traces** — read findings only from five valid completed `TRACE_PATH` receipts
+2. **Deduplicate** — same finding reported by multiple agents → keep richest description
+3. **Cross-reference** — if a pattern is also a template, note both categories
+4. **Sort by confidence** — HIGH reusability first
+5. **Count** — total findings, per-category breakdown
+6. **Include TOOLKIT_HARVEST.md markers** — merge manual markers with auto-discovered
 
 ## Output: Raw Findings List
 

package/templates/.claude/skills/sparc-prd-mini/SKILL.md

@@ -86,6 +86,65 @@ Checkpoint после каждой фазы. Пользователь подтв
     └── CLAUDE.md             # AI tools integration guide
 ```
 
+## Document Role Map Contract
+
+The five SPARC documents are addressed by role, never by a caller-specific filename. A caller may
+provide `DOCUMENT_ROLE_MAP`; otherwise the project-level default below applies. The active map must
+contain exactly these roles: `specification`, `pseudocode`, `architecture`, `refinement`, and
+`completion`.
+
+Resolve the active map once before Phase 3. Bind its values as `SPECIFICATION_FILE`,
+`PSEUDOCODE_FILE`, `ARCHITECTURE_FILE`, `REFINEMENT_FILE`, and `COMPLETION_FILE`, respectively. Every
+phase and traceability gate reads or writes those resolved targets. The literal project-level names
+elsewhere in this document describe the default contour; they do not override a supplied map.
+
+### Project-level default
+
+```yaml
+DOCUMENT_ROLE_MAP:
+  specification: Specification.md
+  pseudocode: Pseudocode.md
+  architecture: Architecture.md
+  refinement: Refinement.md
+  completion: Completion.md
+```
+
+### Atomic validation
+
+If a caller supplies `DOCUMENT_ROLE_MAP`, validate the complete map before Phase 3. Missing roles,
+unknown roles, empty filenames, or a mixture of supplied and default values are an unresolved target
+contract: STOP and report the received map plus every missing, unknown, or empty role. Never fill an
+individual role from the project-level default. When the caller also supplies `TARGET_CATALOG`, join
+that catalog with every filename only after the complete role map passes validation.
+
+### FR/NFR/AC machine-key wire format
+
+Machine keys are exact, case-sensitive joins between the `specification` and `pseudocode` roles:
+`FR-<slug>-<n>`, `NFR-<slug>-<n>`, or `AC-<slug>-<n>`. For a feature contour, `<slug>` is the
+lowercase hyphenated feature directory name and `<n>` is one or more decimal digits.
+
+Declare each key in `SPECIFICATION_FILE` only as a level-three Markdown heading:
+
+```markdown
+### FR-order-refund-1
+### NFR-order-refund-2 — bounded latency
+### AC-order-refund-3 - accepted result
+```
+
+Every `### Algorithm:` block in `PSEUDOCODE_FILE` must carry at least one standalone, exact claim;
+repeat the line when one algorithm addresses several keys:
+
+```markdown
+### Algorithm: Validate refund
+
+REQUIREMENT: `FR-order-refund-1`
+REQUIREMENT: `NFR-order-refund-2`
+```
+
+Prose, comments, tables, examples, `REALISES: SC-...`, and nested `SC-FR-*` scenario IDs are not
+machine-key declarations. Duplicate declarations on either role are invalid; matching sets prove
+cross-document linkage only, not that an algorithm semantically implements the requirement.
+
 ## Workflow Architecture
 
 ```
@@ -373,7 +432,7 @@ view(".claude/skills/problem-solver-enhanced/SKILL.md")
 
 **Inputs:** Product Brief (Phase 0) + Research (Phase 1) + Solution (Phase 2)
 
-**Output — Specification.md + PRD.md:**
+**Output — `SPECIFICATION_FILE` + PRD.md:**
 - Executive Summary
 - User Stories with Acceptance Criteria (Gherkin)
 - Feature Matrix (MVP/v1/v2)
@@ -430,7 +489,7 @@ view("templates/prd.md")
 
 **Цель:** Определить алгоритмы и data flow.
 
-**Output — Pseudocode.md:**
+**Output — `PSEUDOCODE_FILE`:**
 ```markdown
 ## Data Structures
 
@@ -482,22 +541,24 @@ Response (4xx/5xx):
 
 **Шаг 4.9 — ПОКРЫТИЕ СЦЕНАРИЕВ (обязательный, до чекпойнта).**
 
-Re-read `Specification.md` and collect every `SC-` scenario ID. Collect every algorithm's `REALISES`
-line from `Pseudocode.md`. Write a `## Scenario Coverage` block into `Pseudocode.md` — **in every
+Resolve role `specification` through `DOCUMENT_ROLE_MAP` as `SPECIFICATION_FILE`.
+Resolve role `pseudocode` through `DOCUMENT_ROLE_MAP` as `PSEUDOCODE_FILE`. Re-read `SPECIFICATION_FILE` and
+collect every `SC-` scenario ID. Collect every algorithm's `REALISES` line from `PSEUDOCODE_FILE`.
+Write a `## Scenario Coverage` block into `PSEUDOCODE_FILE` — **in every
 case, including the one where everything is covered**, because an absent block and a block saying
 "all covered" are indistinguishable to the next reader:
 
 ```
 ## Scenario Coverage
 
-Scenarios in Specification.md: [N]  ·  claimed by an algorithm: [M]
+Scenarios in [SPECIFICATION_FILE]: [N]  ·  claimed by an algorithm: [M]
 
 Not claimed by any algorithm:
 | Scenario | Reason |
 |---|---|
 | SC-… | ui-only |
 
-Claimed by an algorithm but absent from Specification.md:
+Claimed by an algorithm but absent from [SPECIFICATION_FILE]:
 | Algorithm | Claimed ID |
 |---|---|
 | [name] | SC-… |
@@ -513,7 +574,7 @@ written out rather than left blank, because an empty table and a forgotten table
 | Reason | Means |
 |---|---|
 | `ui-only` | realised entirely in the interface, no algorithm to write |
-| `external-service` | performed by a third party, see `Architecture.md` → External Dependencies |
+| `external-service` | performed by a third party, see the `architecture` role resolved through `DOCUMENT_ROLE_MAP` → External Dependencies |
 | `out-of-mvp-scope` | deliberately not built yet |
 | `data-only` | satisfied by a schema or constraint, not by a procedure |
 | `config-only` | satisfied by OUR OWN configuration — a server setting, a header, a policy file — with no procedure to write |
@@ -528,6 +589,10 @@ scenario describes — no comparison of names can. So this catches *"nobody wrot
 scenario"*; it does not catch *"someone wrote a line that mentions it"*. Say so here rather than
 letting a later reader assume the stronger thing.
 
+With the project-level default, `[SPECIFICATION_FILE]` renders as `Specification.md`, including the
+label `Claimed by an algorithm but absent from Specification.md:`. A supplied map renders the same
+label with its resolved `specification` filename.
+
 **[MANUAL] CP4:**
 ```
 ═══════════════════════════════════════════════════════════════
@@ -559,7 +624,7 @@ view("references/sparc-methodology.md")
 → Секция Architecture для best practices
 ```
 
-**Output — Architecture.md:**
+**Output — `ARCHITECTURE_FILE`:**
 ```markdown
 ## Architecture Overview
 
@@ -660,7 +725,9 @@ example: what an API can do drifts, and a stale fact recorded as evidence is wor
 булевым, когда схема получила перечисление; алгоритм пользуется полем, которого в схеме нет; у
 статуса три значения в одном документе и пять в другом.
 
-Перечитай в `Pseudocode.md` ДВЕ секции — `## Data Structures` и `## Core Algorithms` — и сверь их с
+Resolve role `pseudocode` through `DOCUMENT_ROLE_MAP` as `PSEUDOCODE_FILE`.
+Resolve role `architecture` through `DOCUMENT_ROLE_MAP` as `ARCHITECTURE_FILE`. Перечитай в `PSEUDOCODE_FILE`
+ДВЕ секции — `## Data Structures` и `## Core Algorithms` — и сверь их с
 тем, что выбрано ЗДЕСЬ. Алгоритмы нужны обязательно: расхождение «алгоритм читает поле, которого в
 схеме нет» по одним лишь структурам данных не обнаруживается. Ищи три вида расхождений:
 
@@ -668,16 +735,16 @@ example: what an API can do drifts, and a stale fact recorded as evidence is wor
 - **отсутствующая колонка** — алгоритм читает или пишет поле, которого в схеме нет;
 - **несовпадение набора значений** — у одного и того же поля разное число допустимых значений.
 
-Какую сторону править — решается по РОЛИ документа, а не по старшинству. `Pseudocode.md` держит
-ЛОГИЧЕСКУЮ модель (что означает поле), `Architecture.md` — ФИЗИЧЕСКУЮ (где и как оно лежит).
+Какую сторону править — решается по РОЛИ документа, а не по старшинству. `PSEUDOCODE_FILE` держит
+ЛОГИЧЕСКУЮ модель (что означает поле), `ARCHITECTURE_FILE` — ФИЗИЧЕСКУЮ (где и как оно лежит).
 Поэтому: если эта фаза ввела осознанное физическое ограничение (тип хранилища, индекс, длина) —
-правится `Pseudocode.md`; если выбранная технология НАРУШАЕТ требуемую семантику (теряются значения,
+правится `PSEUDOCODE_FILE`; если выбранная технология НАРУШАЕТ требуемую семантику (теряются значения,
 исчезает состояние, которым пользуется алгоритм) — меняется выбор ЗДЕСЬ, потому что требование
 старше удобства реализации. Секция `## Data Architecture` этого документа остаётся на месте, но
 описывает отображение на хранилище и связи, а НЕ пересказывает список полей: второй экземпляр списка
 становится вторым местом, где начинается расхождение.
 
-Результат записывается в `Architecture.md` ВСЕГДА, отдельным блоком:
+Результат записывается в `ARCHITECTURE_FILE` ВСЕГДА, отдельным блоком:
 
 ```markdown
 ## Reconciliation with Pseudocode
@@ -688,11 +755,14 @@ example: what an API can do drifts, and a stale fact recorded as evidence is wor
 ```
 
 Если сверка не нашла ничего — блок всё равно пишется, и он ОБЯЗАН назвать, что именно
-сверялось: «Расхождений с `Pseudocode.md` не найдено. Сверены сущности: <перечисление>; алгоритмы:
+сверялось: «Расхождений с `[PSEUDOCODE_FILE]` не найдено. Сверены сущности: <перечисление>; алгоритмы:
 <перечисление>.» Одна фраза «расхождений нет» без перечня — это церемония, которую модель напишет
 не глядя; перечень делает утверждение проверяемым. Молчание не является результатом сверки: по нему
 нельзя отличить «сверили и чисто» от «не сверяли».
 
+With the project-level default, `PSEUDOCODE_FILE` is `Pseudocode.md` and `ARCHITECTURE_FILE` is
+`Architecture.md`; a supplied map changes only those resolved filenames, not the reconciliation.
+
 **[MANUAL] CP5:**
 ```
 ═══════════════════════════════════════════════════════════════
@@ -718,7 +788,7 @@ example: what an API can do drifts, and a stale fact recorded as evidence is wor
 
 **Цель:** Edge cases, тестирование, оптимизация.
 
-**Output — Refinement.md:**
+**Output — `REFINEMENT_FILE`:**
 ```markdown
 ## Edge Cases Matrix
 
@@ -796,7 +866,7 @@ Scenario: [Error case]
 
 **Цель:** Deployment и operational readiness.
 
-**Output — Completion.md + CLAUDE.md:**
+**Output — `COMPLETION_FILE` + CLAUDE.md:**
 
 **Completion.md:**
 ```markdown

package/tests/e2e/lifecycle.test.js

@@ -328,23 +328,38 @@ describe('e2e: v1.4 pre-shipped generic toolkit', () => {
     } finally { rmRf(dir); }
   });
 
-  test('init installs all 5 generic rules', () => {
+  test('init installs every registered rule and the consumer detection ladder is readable', () => {
     const dir = tmpDir();
     try {
       runCli(['init'], dir);
-      const expected = [
-        'replicate-pipeline',
-        'skill-interface-protocol',
-        'git-workflow',
-        'insights-capture',
-        'feature-lifecycle',
-      ];
+      const { COMPONENTS } = require(path.join(PKG_DIR, 'src', 'utils.js'));
+      const expected = Object.keys(COMPONENTS.rules.items);
+      assert.ok(expected.includes('cost-of-detection-ladder'),
+        'the explicit ladder slug prevents registry-derived omission from self-confirming');
       for (const rule of expected) {
         assert.ok(
           exists(dir, `.claude/rules/${rule}.md`),
-          `${rule}.md should be installed by init (v1.4 pre-shipped)`
+          `${rule}.md should be installed by init`
         );
       }
+      const ladder = fs.readFileSync(
+        path.join(dir, '.claude/rules/cost-of-detection-ladder.md'), 'utf8');
+      assert.match(ladder, /strongest layer that can reliably express the property/i);
+      assert.match(ladder, /\| Reaction \| Owner \|/i);
+    } finally { rmRf(dir); }
+  });
+
+  test('doctor and verify reject a missing registered detection ladder', () => {
+    const dir = tmpDir();
+    try {
+      runCli(['init'], dir);
+      fs.unlinkSync(path.join(dir, '.claude/rules/cost-of-detection-ladder.md'));
+      const doctor = runCli(['doctor'], dir);
+      const verify = runCli(['verify'], dir);
+      assert.notEqual(doctor.exitCode, 0, 'doctor must reject a missing load-bearing rule');
+      assert.notEqual(verify.exitCode, 0, 'verify must reject a missing load-bearing rule');
+      assert.match(doctor.stdout + doctor.stderr, /cost-of-detection-ladder/);
+      assert.match(verify.stdout + verify.stderr, /cost-of-detection-ladder/);
     } finally { rmRf(dir); }
   });
 
@@ -490,6 +505,37 @@ describe('e2e: v1.4.1 cross-platform hooks', () => {
 describe('e2e: v1.4.2 settings.json merge on --force', () => {
   const USER_HOOK_COMMAND = 'echo "USER-CUSTOM-HOOK-MARKER-12345"';
 
+  test('P18 - init force adds one p-replicator prompt hook without replacing foreign hooks', () => {
+    const dir = tmpDir();
+    const foreign = 'node foreign-user-prompt-hook.cjs';
+    try {
+      runCli(['init'], dir);
+      const settingsPath = path.join(dir, '.claude/settings.json');
+      const manifestPath = path.join(dir, MANIFEST);
+      const settings = JSON.parse(fs.readFileSync(settingsPath, 'utf8'));
+      settings.hooks.UserPromptSubmit = [{
+        matcher: '*',
+        hooks: [{ type: 'command', command: foreign, timeout: 7 }],
+      }];
+      fs.writeFileSync(settingsPath, JSON.stringify(settings, null, 2));
+
+      const manifest = JSON.parse(fs.readFileSync(manifestPath, 'utf8'));
+      delete manifest.shippedDefaults['settings.json'].hooks.UserPromptSubmit;
+      fs.writeFileSync(manifestPath, JSON.stringify(manifest, null, 2));
+
+      assert.equal(runCli(['init', '--force'], dir).exitCode, 0);
+      assert.equal(runCli(['init', '--force'], dir).exitCode, 0);
+      const merged = JSON.parse(fs.readFileSync(settingsPath, 'utf8'));
+      const commands = merged.hooks.UserPromptSubmit
+        .flatMap((matcher) => matcher.hooks || [])
+        .map((hook) => hook.command);
+      assert.equal(commands.filter((command) => command === foreign).length, 1,
+        'the consumer-owned prompt hook must survive both upgrades');
+      assert.equal(commands.filter((command) => command.includes('session-insights.cjs')).length, 1,
+        'the package prompt hook must be added exactly once');
+    } finally { rmRf(dir); }
+  });
+
   test('init --force preserves user-added hooks in settings.json', () => {
     const dir = tmpDir();
     try {