@polymorphism-tech/morph-spec 9.8.0 → 10.0.0

package/CLAUDE.md

@@ -40,7 +40,7 @@ No build step — pure ESM JavaScript, runs directly via Node.js 22+.
 
 ### Module Layout
 
-- **`bin/morph-spec.js`** — Commander.js CLI entry point: install/inspect commands plus the deterministic steps of the loop (`create`, `plan-check`, `gate-check`, `approve`/`advance`, `verify`, `report-check`, `mutate`, `finish`).
+- **`bin/morph-spec.js`** — Commander.js CLI entry point: install/inspect commands plus the deterministic steps of the loop (`create`, `plan-check`, `gate-check`, `approve`/`advance`, `verify`, `report-check`, `e2e`, `finish`).
 - **`src/commands/`** — CLI command handlers organized by domain (project/, state/, validation/, scaffold/, mcp/, context/, dashboard/).
 - **`src/core/`** — Core domain logic:
   - `state/state-manager.js` — Atomic state I/O with process-ID temp files
@@ -49,14 +49,14 @@ No build step — pure ESM JavaScript, runs directly via Node.js 22+.
   - `workflows/` — Workflow type detector
 - **`src/lib/`** — Libraries and subsystems:
   - `validators/` — Validators run by the `verify` validators node (code-budget, contracts, nextjs, packages, reuse)
-  - `generators/` — Code generation (architecture context, handoff, scaffold, settings)
+  - `generators/` — Code generation (architecture context, scaffold, settings)
   - `stack-filter.js` — Stack-aware asset filtering via YAML frontmatter `stacks:` tags
   - `installers/` — Agents, skills, hooks installers
 - **`src/utils/`** — Shared utilities (logger, banner, file-copier, claude-settings-manager, agents-installer, skills-installer, hooks-installer)
 - **`framework/`** — Metadata installed into target projects (READ-ONLY at runtime):
-  - `MORPH.md` — NLH charter (the single source of behavioral truth; ≤ 20k chars, `resident-size` canary)
-  - `CLAUDE.md` — Runtime instructions injected into target's `.claude/CLAUDE.md` (≤ 4k chars)
-  - `agents.json` — 10 flat personas (see below); `maxTurns` per persona (ADR-004)
+  - `MORPH.md` — NLH charter (the single source of behavioral truth; ≤ 8k chars, `resident-size` canary)
+  - `CLAUDE.md` — Runtime instructions injected into target's `.claude/CLAUDE.md` (≤ 3k chars)
+  - `agents.json` — the 3 Gate 3 personas (see below); `maxTurns` per persona
   - `phases.json` — 4 phases (plan, uiux, implement, review)
   - `standards/STANDARDS.json` — index of available standards
   - `hooks/claude-code/core/` — fail-open guardrail hooks (+ `hooks/shared/` for what hooks and `src/` both read)
@@ -79,7 +79,7 @@ No build step — pure ESM JavaScript, runs directly via Node.js 22+.
 
 **Stack-aware filtering:** Assets declare `stacks: ['dotnet']` in YAML frontmatter or `stack: ['dotnet']` in `agents.json`. `setup-infra.js` reads `config.json → project.stack`, splits into tags, and installs only matching assets.
 
-**Personas, not tiers:** `agents.json` has 10 flat personas — 6 domain implementers (dotnet-senior, ef-modeler, maf-expert, nextjs-expert, ui-designer, infra-engineer), the 3 read-only Gate 3 roles (verifier, reviewer, security) and 1 stage persona (`scout`, read-only). `model` is sized per stage: `sonnet` for implementers, `scout`, `verifier` and `reviewer`; `security` and `ui-designer` (design quality is the deliverable, high effort) are `opus`; the dispatch routes the verifier to `opus` when `verifierModelFor` (risk-detector.js, read by `report-check --before-eval`) sees prompt/model or media (measured against haiku in `docs/specs/scout-evidence/decisions.md`). No tier system — Claude assumes the 3 NLH roles (Orchestrator/Specialist/Evaluator) per MORPH.md.
+**Single loop:** the main session plans, implements the whole plan in one window, fixes and finishes; domain know-how lives in the stack rules `framework/rules/*-standards.md`, loaded by `paths:` when the session touches a matching file and installed by `syncStackRules` from `plan-check`/`verify`/`update` (`docs/specs/single-loop/decisions.md`, ADR-010). `agents.json` has only the 3 read-only Gate 3 roles — `verifier` (executes each `Cn`), `reviewer` (plan.md + diff), `security` (sensitive paths) — dispatched in fresh context in the same response. `reviewer` is `sonnet`, `security` is `opus`; the dispatch routes the verifier to `opus` when `verifierModelFor` (risk-detector.js, read by `report-check --before-eval`) sees prompt/model or media.
 
 ### Phase Sequence
 
@@ -87,7 +87,7 @@ No build step — pure ESM JavaScript, runs directly via Node.js 22+.
 plan → [uiux] → implement → review
 ```
 
-Gates: `plan` (plan.md + `checks.json` from `plan-check`), `uiux` (optional, after the plan gate, `1-design/`), `review` (Gate 3: `verifier.md` + `reviewer.md` [+ `security.md`]). Folders keep their ids: `2-plan/`, `1-design/`, `3-implement/`, `4-review/`. `chore`/`hotfix` are born with `plan` stamped by policy. Implementation unit: the slices `F1..Fn` of plan.md; progress = `3-implement/reports/F{n}.md`.
+Gates: `plan` (plan.md + `checks.json` from `plan-check`), `uiux` (optional, after the plan gate, `1-design/`), `review` (Gate 3: `verifier.md` + `reviewer.md` [+ `security.md`]). Folders keep their ids: `2-plan/`, `1-design/`, `3-implement/`, `4-review/`. `chore`/`hotfix` are born with `plan` stamped by policy. Done-when = the scenarios `C1..Cn` of plan.md; progress = `3-implement/report.md` (one row per `Cn`: `pendente`/`feito`/`bloqueado`). The archive keeps `feature.json` + `2-plan/plan.md`.
 
 ### State Schema
 
@@ -96,7 +96,7 @@ Authoritative, committed state: `.morph/features/{f}/feature.json` (status, appr
 ## Testing Conventions
 
 - **Framework:** Node.js native `node:test` — no Jest or Mocha
-- **Concurrency:** files run in parallel (`scripts/run-tests.mjs`) — `node --test` gives each file its own process, so `process.chdir()` stays inside the file. A test never writes a tracked file of this repo: a script under test reads its root from `MORPH_REPO_ROOT`, and the test points it at `copyTrackedRepo()` (`test/helpers/repo-copy.js`, a temp git repo with the tracked files). The runner fails if a run changes the tracked tree
+- **Concurrency:** files run in parallel (`scripts/run-tests.mjs`) — `node --test` gives each file its own process, so `process.chdir()` stays inside the file. A test never writes a tracked file of this repo: a script under test reads its root from `MORPH_REPO_ROOT`, and the test points it at `copyTrackedRepo()` (`test/helpers/repo-copy.js`, a temp git repo with the tracked files). The runner fails if a run changes the tracked tree. Each file has a 10-minute `--test-timeout` (it bounds the whole file under process isolation): a hang fails by name. A test that imports a hook sets `MORPH_HOOK_CHAINED=1` first, or the hook's `runStandalone` waits on stdin, which `node --test` never closes
 - **Subprocesses:** drive the CLI through `test/helpers/cli.js` (`runCli`, no shell) and git through `test/helpers/git.js` (`git(cwd, ...args)`) — `execSync('git …')` pays a cmd.exe per call on Windows. Hooks that export `run(payload)` can be called in-process (see `test/hooks/loop-detect.test.js`); keep one spawn per hook for the standalone wiring
 - **Temp dirs:** `createTempDir()` / `cleanupTempDir()` from `test/helpers/test-utils.js`
 - **Home:** a suite that runs `setupInfra`/`updateCommand`/`initCommand` calls `isolateHome()` (test-utils) at the top — the installers write `~/.claude/settings.json` through `os.homedir()`. Canary: `test/meta/home-isolation.test.js`
@@ -125,8 +125,8 @@ A matriz "mexeu em X → verifique Y" mora em `.claude/rules/matrix-*.md`, uma r
 | `.claude/rules/matrix-hooks.md` | Hooks, telemetria e comment-guard |
 | `.claude/rules/matrix-charter.md` | Carta NLH, skills e docs instaláveis |
 | `.claude/rules/matrix-templates-ai.md` | Fases, templates, ai-kit e CI |
-| `.claude/rules/matrix-plan-review.md` | Plano, relatórios de fatia, gates e Gate 3 |
-| `.claude/rules/matrix-verify.md` | Verify, testes, mutação e CI |
+| `.claude/rules/matrix-plan-review.md` | Plano, progresso, gates e Gate 3 |
+| `.claude/rules/matrix-verify.md` | Verify, testes e CI |
 | `.claude/rules/matrix-worktree-git.md` | Worktrees, finish e git |
 | `.claude/rules/matrix-graph.md` | Grafo de codebase e validador de reuso |
 | `.claude/rules/matrix-platform.md` | MCP, installers, procs, heavy e dashboard |

package/README.md

@@ -3,7 +3,7 @@
 > Spec-driven development framework for multi-stack projects. Turns feature requests into implementation-ready code through structured, AI-orchestrated phases.
 
 **Package:** `@polymorphism-tech/morph-spec`
-**Version:** 9.8.0  
+**Version:** 10.0.0  
 **Requires:** Node.js 22+, Claude Code
 
 ---
@@ -244,11 +244,11 @@ Guardrail hooks are installed into `.claude/settings.local.json` across 13 event
 | --------------------------- | ---------------------------------------------------------------------- | ------------------------------------------------------------------------------ |
 | `SessionStart`              | `worktree-context.js`                                                   | Injects feature identity (worktree) or active-feature summary (root)           |
 | `PreToolUse` (Write\|Edit)  | `protect-spec-files.js`, `enforce-phase-writes.js`                       | Freezes `plan.md` after the plan gate (only `plan-check` writes `checks.json`); enforces phase folders |
-| `PreToolUse` (Bash)         | `pre-bash-denylist.js`, `block-destructive-worktree-removal.js`, `block-raw-worktree-add.js`, `warn-bash-source-write.js`, `heavy-gate.js`, `worktree-env-inject.js` | Destructive-command guards, worktree governance, heavy-node semaphore, per-worktree env |
+| `PreToolUse` (Bash)         | `pre-bash-denylist.js`, `block-destructive-worktree-removal.js`, `block-raw-worktree-add.js`, `heavy-gate.js`, `worktree-env-inject.js` | Destructive-command guards, worktree governance, heavy-node semaphore, per-worktree env |
 | `WorktreeCreate` / `WorktreeRemove` | `worktree-create.js`, `worktree-remove.js`                     | Native worktrees delegate to the CLI                                            |
 | `PostToolUse` (Write\|Edit) | `comment-guard.js`                                                       | Blocks added comments the comment standard forbids                              |
 | `PostToolUse`               | `state-sync.js`, `gate-guard.js`                                         | Syncs feature state; enforces gate ordering                                     |
-| `PostToolUse`, `PostToolUseFailure`, `PostToolBatch`, `SubagentStop`, `UserPromptSubmit` | `telemetry-log.js`, `loop-detect.js`, `graph-refresh-kick.js`, `heavy-release.js` | Telemetry, repeated-action loops, graph refresh, heavy-slot release |
+| `PostToolUse`, `PostToolUseFailure`, `PostToolBatch`, `SubagentStop`, `UserPromptSubmit` | `telemetry-log.js`, `loop-detect.js`, `graph-refresh-kick.js`, `heavy-release.js`, `masked-exit.js` | Telemetry, repeated-action loops, graph refresh, heavy-slot release, masked test exit warning |
 | `Stop`                      | `validate-completion.js`                                                  | Warns about incomplete tasks, missing outputs, or pending approval gates         |
 
 
@@ -380,4 +380,4 @@ Code generated by morph-spec (contracts, templates, implementation output) belon
 
 ---
 
-*morph-spec v9.8.0 by [Polymorphism Tech](https://polymorphism.tech)*
+*morph-spec v10.0.0 by [Polymorphism Tech](https://polymorphism.tech)*

package/bin/morph-spec.js

@@ -23,16 +23,15 @@ const worktreeProvisionCommand = lazy('../src/commands/project/worktree.js', 'wo
 const worktreeLinkCommand = lazy('../src/commands/project/worktree.js', 'worktreeLinkCommand');
 const worktreeListCommand = lazy('../src/commands/project/worktree.js', 'worktreeListCommand');
 const worktreeRemoveCommand = lazy('../src/commands/project/worktree.js', 'worktreeRemoveCommand');
-const worktreeReapCommand = lazy('../src/commands/project/worktree-reap.js', 'worktreeReapCommand');
 const branchPruneCommand = lazy('../src/commands/project/branch.js', 'branchPruneCommand');
 const fleetCommand = lazy('../src/commands/project/fleet.js', 'fleetCommand');
 const envCommand = lazy('../src/commands/project/env.js', 'envCommand');
 const claimCommand = lazy('../src/commands/project/claim.js', 'claimCommand');
 const standardsCommand = lazy('../src/commands/project/standards.js', 'standardsCommand');
 const reportCheckCommand = lazy('../src/commands/state/report-check.js', 'reportCheckCommand');
-const scoutCheckCommand = lazy('../src/commands/validation/scout-check.js', 'scoutCheckCommand');
 const planCheckCommand = lazy('../src/commands/validation/plan-check.js', 'planCheckCommand');
 const stackScanCommand = lazy('../src/commands/project/stack-scan.js', 'stackScanCommand');
+const stackSetCommand = lazy('../src/commands/project/stack-set.js', 'stackSetCommand');
 const testsAuditCommand = lazy('../src/commands/validation/tests-audit.js', 'testsAuditCommand');
 
 const approveCommand = lazy('../src/commands/state/approve.js', 'approveCommand');
@@ -40,12 +39,12 @@ const approvalStatusCommand = lazy('../src/commands/state/approve.js', 'approval
 const unapproveCommand = lazy('../src/commands/state/approve.js', 'unapproveCommand');
 const advanceCommand = lazy('../src/commands/state/advance.js', 'advanceCommand');
 const archiveCommand = lazy('../src/commands/state/archive.js', 'archiveCommand');
+const recapCommand = lazy('../src/commands/state/recap.js', 'recapCommand');
 const finishCommand = lazy('../src/commands/state/finish.js', 'finishCommand');
 const gateDecisionCommand = lazy('../src/commands/state/gate-decision.js', 'gateDecisionCommand');
 const gateCheckCommand = lazy('../src/commands/state/gate-check.js', 'gateCheckCommand');
 const createCommand = lazy('../src/commands/state/create.js', 'createCommand');
 const retypeCommand = lazy('../src/commands/state/retype.js', 'retypeCommand');
-const handoffCommand = lazy('../src/commands/state/handoff.js', 'handoffCommand');
 const nextCommand = lazy('../src/commands/state/next.js', 'nextCommand');
 const listCommand = lazy('../src/commands/state/list.js', 'listCommand');
 const deleteCommand = lazy('../src/commands/state/delete.js', 'deleteCommand');
@@ -54,14 +53,15 @@ const graphStatusCommand = lazy('../src/commands/graph/status.js', 'graphStatusC
 const graphBuildCommand = lazy('../src/commands/graph/build.js', 'graphBuildCommand');
 const graphAffectedCommand = lazy('../src/commands/graph/affected.js', 'graphAffectedCommand');
 const graphExplainCommand = lazy('../src/commands/graph/explain.js', 'graphExplainCommand');
+const graphDiffCommand = lazy('../src/commands/graph/diff.js', 'graphDiffCommand');
 
 const validateCommand = lazy('./validate.js', 'validateCommand');
 const validateFeatureCommand = lazy('../src/commands/validation/validate-feature.js', 'validateFeatureCommand');
 const verifyCommand = lazy('../src/commands/validation/verify.js', 'verifyCommand');
-const mutateCommand = lazy('../src/commands/validation/mutate.js', 'mutateCommand');
 const e2eUpCommand = lazy('../src/commands/e2e/e2e.js', 'e2eUpCommand');
 const e2eStatusCommand = lazy('../src/commands/e2e/e2e.js', 'e2eStatusCommand');
 const e2eDownCommand = lazy('../src/commands/e2e/e2e.js', 'e2eDownCommand');
+const e2eResetCommand = lazy('../src/commands/e2e/e2e.js', 'e2eResetCommand');
 
 const scaffoldDotnetCommand = lazy('../src/commands/scaffold/scaffold.js', 'scaffoldDotnetCommand');
 const scaffoldNextjsCommand = lazy('../src/commands/scaffold/scaffold.js', 'scaffoldNextjsCommand');
@@ -102,6 +102,8 @@ program
   .option('-f, --force', 'Overwrite existing MORPH installation')
   .option('-p, --path <path>', 'Target path (default: current directory)')
   .option('--skip-mcp', 'Skip MCP server auto-setup')
+  .option('--stack <stack>', 'Empty project: declare the stack without asking (dotnet, nextjs or dotnet+nextjs)')
+  .option('--with <integrations>', 'With --stack: comma-separated integrations (neon, maf, docker)')
   .action(initCommand);
 
 program
@@ -181,6 +183,14 @@ program
   .option('--force', 'Archive even if Gate 3 (review) is not approved')
   .action((feature, options) => archiveCommand(feature, options));
 
+program
+  .command('recap <feature>')
+  .description('Recap animado da feature (cenários provados, estados, o que mudou, gates), com legenda e narração; o finish já gera')
+  .option('--piper', 'Narra com o Piper local (áudio embutido, gravado no temp); sem ele, voz do sistema')
+  .option('-p, --path <path>', 'Target path (default: current directory)')
+  .option('--json', 'Output as JSON')
+  .action((feature, options) => recapCommand(feature, options));
+
 program
   .command('finish <feature>')
   .description('Close a feature: archive → integrate. --merge merges morph/{f} into the default (local); --pr pushes it + opens a PR. Works with and without a worktree')
@@ -259,13 +269,6 @@ program
   .option('--json', 'Output as JSON')
   .action((feature, newType, options) => retypeCommand(feature, newType, options));
 
-program
-  .command('handoff <feature>')
-  .description('Generate the compressed session-handoff document (.morph/features/{f}/handoff.md) — pointers, not copies. Resolves the tree that owns the feature\'s state (worktree, or morph/{feature} if only committed) and writes there — not necessarily cwd.')
-  .option('--json', 'Output as JSON')
-  .option('--here', 'Force building/writing in the current directory, even if another tree owns this feature')
-  .action((feature, options) => handoffCommand(feature, options));
-
 program
   .command('next <feature>')
   .description('The next step of the loop, decided from disk (phase, gates, slice reports, verify stamp, fail streak) in the tree that owns the feature. Read-only.')
@@ -313,6 +316,14 @@ graphCommand
   .option('--json', 'Output as JSON')
   .action((symbol, options) => graphExplainCommand(symbol, options));
 
+graphCommand
+  .command('diff')
+  .description('Árvore de chamadas do que mudou (+ novo, ~ alterado), do merge-base ao HEAD; recusa grafo que não é deste commit')
+  .option('-p, --path <path>', 'Target path (default: current directory)')
+  .option('--base <ref>', 'Base do diff (default: merge-base com a branch padrão)')
+  .option('--json', 'Output as JSON')
+  .action((options) => graphDiffCommand(options));
+
 program
   .command('validate [validator]')
   .description('Run project validations (all | packages | next-components | spec-tasks)')
@@ -356,18 +367,6 @@ program
     verifyCommand(feature, { ...options, noPersist: options.persist === false })
   );
 
-program
-  .command('mutate [feature]')
-  .description('Prova de mutação do Gate 3 em lote: lê o manifesto de mutantes, roda todos (um build de schemata em .NET, subconjunto primeiro, suíte inteira só para sobrevivente), restaura os bytes e imprime a seção `## Mutações`')
-  .option('--plan <file>', 'Manifesto JSON (default: .morph/features/<feature>/4-review/mutants.json)')
-  .option('--out <file>', 'Onde gravar a seção `## Mutações` (default: 4-review/mutation-section.md)')
-  .option('--allow-dirty', 'Permite mutar arquivo com alteração não commitada (o snapshot é o único caminho de volta)')
-  .option('--jobs <n>', 'Rodadas de subconjunto em paralelo (default 2); cada worker toma um slot heavy livre, e sem slot livre a rodada degrada para sequencial', (v) => Number.parseInt(v, 10))
-  .option('--timeout <ms>', 'Teto de cada comando de build/teste', (v) => Number.parseInt(v, 10))
-  .option('--restore', 'Recupera o disco de uma rodada que morreu no meio: restaura os bytes do journal, sem rodar nada')
-  .option('--json', 'Output as JSON')
-  .action((feature, options) => mutateCommand(feature, options));
-
 const scaffoldCmd = program
   .command('scaffold')
   .description('Generate feature files from templates');
@@ -443,6 +442,12 @@ e2eCommand
   .option('--json', 'Output as JSON')
   .action((feature, options) => e2eDownCommand(feature, options));
 
+e2eCommand
+  .command('reset <feature>')
+  .description('Put the running stack\'s data back in a known state between verification flows: e2e.resetCommand, or recreate the non-built services that own a named volume; then the auth seed. App services and e2e.hostServices are not restarted. Exit 1 when the stack is not up')
+  .option('--json', 'Output as JSON')
+  .action((feature, options) => e2eResetCommand(feature, options));
+
 const heavyCommand = program
   .command('heavy')
   .description('Slots por usuário para os nós pesados (build, testes, e2e), compartilhados entre as sessões');
@@ -496,14 +501,6 @@ worktreeCommand
   .option('--json', 'Output as JSON (always JSON)')
   .action((feature, options) => worktreeRemoveCommand(feature, options));
 
-worktreeCommand
-  .command('reap')
-  .description('Remove the worktrees and branches of sub-agents that stopped clean (nothing uncommitted, no commit only they hold); everything else is listed with the reason and left alone')
-  .option('--dry-run', 'Report what would be removed without touching anything')
-  .option('--grace <minutes>', 'Minutes a stopped agent must stay quiet before its tree is reaped (default 30, or worktree.reap.graceMinutes)')
-  .option('--json', 'Output as a single JSON line')
-  .action((options) => worktreeReapCommand(options));
-
 worktreeCommand
   .command('list')
   .description('List active feature worktrees with their phase and gate status')
@@ -532,7 +529,7 @@ program
 
 program
   .command('fleet')
-  .description('Fleet view — every active feature across primary root and worktrees (phase, gates, tasks, ports, handoff, activity)')
+  .description('Fleet view — every active feature across primary root and worktrees (phase, gates, scenarios, ports, activity)')
   .option('--json', 'Output as JSON')
   .action((options) => fleetCommand(options));
 
@@ -556,18 +553,11 @@ program
   .action((options) => standardsCommand(options));
 
 program
-  .command('report-check <feature> [slices...]')
-  .description('Audit slice report files (3-implement/reports/{slice}.md) against the work the dispatch actually wrote (telemetry, keyed by the agent that wrote the report). Exits 0 only when every audited slice is intact; naming slices audits only those')
+  .command('report-check <feature>')
+  .description('State of the Gate 3 reports (4-review/), scenario progress of 3-implement/report.md and the tree they were written against')
   .option('--before-eval', 'Save a snapshot of the tree before dispatching the review agents and print their models; later runs report what changed since')
   .option('--json', 'Output as JSON')
-  .action((feature, slices, options) => reportCheckCommand(feature, slices, options));
-
-program
-  .command('scout-check <map>')
-  .description('Audit a scout map: every path:line citation must exist and carry the literal line it quotes (±2 lines); absence/liveness verdicts only as hypotheses with the search command and its raw count. Exits 0 only when nothing failed')
-  .option('--root <dir>', 'Tree the citations are relative to (default: top of the current worktree)')
-  .option('--json', 'Output as JSON')
-  .action((map, options) => scoutCheckCommand(map, options));
+  .action((feature, options) => reportCheckCommand(feature, options));
 
 program
   .command('plan-check <feature>')
@@ -579,10 +569,17 @@ program
 
 program
   .command('stack-scan [feature]')
-  .description('Audit the code against config.project.stack, judging only the configured stacks (runtime, ORM, database, Guid.NewGuid, generic repository, service layer, raw SQL, sealed, router, UI library) by severity. With a feature, write .morph/features/<f>/2-plan/stack-scan.md; without, print the report')
+  .description('Audit the code against config.project.stack, judging only the configured stacks (runtime, ORM, database, Guid.NewGuid, generic repository, service layer, raw SQL, sealed, router, UI library) by severity. Prints the report (markdown, or --json); never writes a file')
   .option('--json', 'Output as JSON')
   .action((feature, options) => stackScanCommand(feature, options));
 
+program
+  .command('stack-set <stack>')
+  .description('Declare the project stack (dotnet, nextjs or dotnet+nextjs) in config.json and install its domain rules into .claude/rules/; rules load when a Claude Code session starts')
+  .option('--with <integrations>', 'Comma-separated integrations (neon, maf, docker)')
+  .option('--json', 'Output as JSON')
+  .action((stack, options) => stackSetCommand(stack, options));
+
 program
   .command('tests-audit')
   .description('Read-only audit of the tracked test files: case count, largest files, tests that pin prose or a constant, [Fact] families that are [Theory] candidates, and one subject tested in several layers — candidates to delete or merge, never verdicts')

package/claude-plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "morph-spec",
-  "version": "9.8.0",
+  "version": "10.0.0",
   "displayName": "MORPH-SPEC Framework",
   "description": "Spec-driven development with 38 agents and 8-phase workflow for .NET/Blazor/Next.js/Azure",
   "publisher": "polymorphism-tech",

package/docs/CHEATSHEET.md

@@ -185,4 +185,4 @@ Never edited directly — go through CLI commands or `morph-spec update` instead
 
 ---
 
-*morph-spec v9.8.0 by Polymorphism Tech*
+*morph-spec v10.0.0 by Polymorphism Tech*

package/docs/QUICKSTART.md

@@ -249,4 +249,4 @@ morph-spec doctor
 
 ---
 
-*morph-spec v9.8.0 by Polymorphism Tech*
+*morph-spec v10.0.0 by Polymorphism Tech*

package/framework/CLAUDE.md

@@ -4,47 +4,41 @@
 
 ## Início de cada sessão
 
-1. Leia `.morph/framework/MORPH.md` uma vez (de novo após `/compact`): papéis, fases, gates, dispatch.
-2. O SessionStart injeta a feature ativa e `▶ próximo:`; sem ele, `morph-spec next <f>`. Estado autoritativo: `.morph/features/{f}/feature.json` (`state.json` é índice — nunca edite).
-3. Feature em andamento → faça o que o `next` manda, entre gates, sem esperar instrução. Nenhuma ativa → pergunte qual trabalhar.
+1. Leia `.morph/framework/MORPH.md` uma vez (de novo após `/compact`).
+2. O SessionStart injeta a feature ativa e `▶ próximo:`; sem ele, `morph-spec next <f>`. Estado autoritativo: `.morph/features/{f}/feature.json` — só pela CLI.
+3. Feature em andamento → faça o que o `next` manda, sem esperar instrução. Nenhuma → pergunte qual.
 
 ## Regras
 
-- Perguntas ao usuário só por `AskUserQuestion` (1-4 por chamada), nunca em texto solto. Na entrevista da `morph-plan`, **uma pergunta por chamada**: lote impede que uma resposta pode os ramos seguintes. Única exceção: ferramenta indisponível ou negada (headless) — texto numerado com a resposta recomendada e gate pausado (`morph-plan` §3).
-- A pergunta de gate nomeia a feature ("Gate do plano — Plano de {feature} aprovado?"): é por esse nome que o hook `state-sync` grava a aprovação no `feature.json` certo; sem nome, com mais de uma feature, ele recusa e você aprova por `morph-spec approve <feature> <gate>`.
-- Estado só pela CLI, nunca à mão: `feature.json`, `gateDecisions`, `checks.json` (escrito só por `plan-check`) e o corpo do `plan.md` depois do gate (fato errado → `## Errata` no fim; escopo mudou → `unapprove`).
-- SessionStart avisou claim vivo de outra sessão na mesma feature → confirme com o usuário antes de `finish`/`archive`/`delete`.
-- Depois de um dispatch, leia o arquivo de relatório da fatia, não a mensagem final, que pode não chegar.
+- Perguntas só por `AskUserQuestion`; na entrevista do plano, uma por chamada. Headless: texto numerado com a recomendada e gate pausado.
+- A pergunta de gate nomeia a feature ("Gate do plano — Plano de {f} aprovado?"): é por ela que o hook `state-sync` grava a aprovação.
+- "Não li" ou "não entendi" depois de uma saída longa → refaça do zero, sem supor que a anterior foi lida: até 5 passos numerados, uma frase cada, o primeiro sendo o que mudou para o usuário.
+- Plano aprovado congela: fato errado → `## Errata` no fim; escopo mudou → `unapprove`.
+- SessionStart avisou claim de outra sessão na feature → confirme antes de `finish`/`archive`/`delete`.
 
 ## CLI do loop
 
-Prefixo `morph-spec`; flags e porquês em `.morph/framework/CLI.md` (um `###` por comando).
+Prefixo `morph-spec`; detalhe em `.morph/framework/CLI.md`.
 
 | Comando | Para quê |
 |---|---|
-| `create <f> --description "…" --type <t> [--worktree]` | nasce a feature; exit 2 = você escolhe o tipo |
-| `next <f>` | o próximo passo, decidido do disco |
-| `stack-scan [f]` | desvios do código contra `project.stack` |
-| `tests-audit` | candidatos a apagar/fundir na suíte (só lê) |
-| `plan-check <f>` | valida `plan.md`, grava `checks.json`, imprime as ondas |
-| `gate-check <f> <gate>` | decide a pausa pelos sinais do disco |
-| `approve <f> <gate> --mode auto` / `advance <f>` | aprova (advance também abre a próxima fase) |
+| `create <f> --request "…" [--worktree]` | nasce a feature |
+| `next <f>` | o próximo passo, do disco |
+| `plan-check <f>` | valida o plano, grava `checks.json` |
+| `gate-check <f> <gate>` / `approve` / `advance` | decide e aprova gates |
 | `verify <f> [--ci]` | build + testes + validadores + e2e, sem filtro |
-| `report-check <f> [F1 …]` | audita o relatório de cada fatia e diz a ação |
-| `mutate <f>` | mutação dirigida do Gate 3 (até 3 mutantes) |
-| `finish <f> --pr\|--merge` | encerra; só o `--merge` integra |
-| `standards --find "<tema>"` / `--show <id>#<âncora>` | digest do standard / uma seção |
-| `heavy status\|run -- <cmd>` | fila de build/teste; exit 3 = fila cheia |
-| `worktree setup\|provision\|link\|remove\|list\|reap` | ciclo da worktree; `remove` é a única remoção segura |
-| `graph affected\|explain <sym>` | blast radius; sem grafo, caia para grep |
+| `report-check <f> [--before-eval]` | estado dos relatórios do Gate 3 |
+| `e2e up\|status\|reset\|down <f>` | stack da feature |
+| `finish <f> --pr\|--merge` | encerra; grava o recap animado no archive |
+| `standards --find "<tema>"` | digest do standard |
+| `heavy run -- <cmd>` | build/teste pesado na fila |
+| `worktree setup\|provision\|link\|remove\|list` | ciclo da worktree; `remove` é a única remoção segura |
+| `graph affected\|explain <sym>` | blast radius; sem grafo, grep |
+| `graph diff [--base <ref>]` | árvore de chamadas do que mudou; recusa grafo velho |
 
-## Slash commands
+## Gate 3 (sub-agents)
 
-`/morph-plan {f}` (plano), `/morph-apply {f}` (execução + Gate 3), `/morph-hotfix {f}`, `/morph-status [f]`.
-
-## Personas (em `.claude/agents/`)
-
-Despache fatias com `subagent_type: "morph-{id}"` — a persona, gerada de `agents.json`, é o system prompt. Tipo não registrado: `general-purpose` com o `persona` colado, registrado no relatório da fatia. Nunca edite `.claude/agents/morph-*.md` à mão (`morph-spec doctor` acusa o drift).
+`morph-verifier`, `morph-reviewer`, `morph-security` em `.claude/agents/`, gerados de `agents.json` — nunca edite à mão. Com a crítica do plano (`Plan` nativo, `morph-plan`), são os únicos sub-agents do fluxo; o resto é feito nesta sessão. Leia o relatório que gravam, não a mensagem final.
 
 <!-- morph-agents:start -->
 _Lista derivada na instalação. Bloco vazio → `morph-spec setup-infra --reinstall`._
@@ -52,4 +46,4 @@ _Lista derivada na instalação. Bloco vazio → `morph-spec setup-infra --reins
 
 ## Precedência de skills
 
-Com o plugin superpowers ativo, `morph-*` vence o equivalente genérico (aplica os gates e os paths de `.morph/`), e `using-git-worktrees`→`morph-spec worktree setup|provision`.
+`morph-*` vence o equivalente genérico; `using-git-worktrees` → `morph-spec worktree setup|provision`.