@lovrozagar/grunt 0.5.0 → 0.6.0

package/scripts/speak.mjs

@@ -0,0 +1,291 @@
+#!/usr/bin/env node
+/** Text-to-speech (output only). Providers: elevenlabs, openai. */
+import { spawnSync } from "node:child_process";
+import fs from "node:fs";
+import os from "node:os";
+import path from "node:path";
+import { fileURLToPath, pathToFileURL } from "node:url";
+
+export const DEFAULT_ELEVEN_VOICE = "JBFqnCBsd6RMkjVDRZzb";
+export const DEFAULT_ELEVEN_MODEL = "eleven_multilingual_v2";
+export const DEFAULT_OPENAI_VOICE = "coral";
+export const DEFAULT_OPENAI_MODEL = "gpt-4o-mini-tts";
+export const OPENAI_VOICES = [
+  "alloy",
+  "ash",
+  "ballad",
+  "coral",
+  "echo",
+  "fable",
+  "nova",
+  "onyx",
+  "sage",
+  "shimmer",
+];
+export const USAGE =
+  "usage: speak whoami | voices | say --text T [--provider elevenlabs|openai] [--voice ID] [--out path] [--play]";
+const ELEVEN_API = "https://api.elevenlabs.io/v1";
+const OPENAI_API = "https://api.openai.com/v1";
+export const SETUP = "node scripts/setup.mjs speak";
+export const NEED_KEY =
+  "missing ELEVENLABS_API_KEY or OPENAI_API_KEY (or ~/.grunt/speak.json). setup: node scripts/setup.mjs speak";
+export const NEED_ELEVEN =
+  "missing ELEVENLABS_API_KEY (or ~/.grunt/speak.json elevenlabs.apiKey). setup: node scripts/setup.mjs speak";
+export const NEED_OPENAI =
+  "missing OPENAI_API_KEY (or ~/.grunt/speak.json openai.apiKey). setup: node scripts/setup.mjs speak";
+
+export function formatHttpError(status, body, provider) {
+  const n = Number(status);
+  const raw = String(body || "");
+  if (/missing_permissions/i.test(raw)) {
+    return `${provider} key present but missing permission (${n || status}). setup: ${SETUP}`;
+  }
+  const auth =
+    n === 401 ||
+    n === 403 ||
+    /unauthorized|invalid.?api.?key|authentication/i.test(raw);
+  if (auth) {
+    return `${provider} key missing or rejected (${n || status}). setup: ${SETUP}`;
+  }
+  return `${status} ${raw.slice(0, 200)}`.trim();
+}
+
+export function parseArgv(argv) {
+  const out = { _: [], flags: {} };
+  const args = Array.isArray(argv) ? argv : [];
+  for (let i = 0; i < args.length; i++) {
+    const a = args[i];
+    if (a === "--") {
+      out._.push(...args.slice(i + 1));
+      break;
+    }
+    if (a.startsWith("--") && a.length > 2) {
+      const key = a.slice(2);
+      const nxt = args[i + 1];
+      if (nxt && !nxt.startsWith("--")) {
+        out.flags[key] = nxt;
+        i += 1;
+      } else out.flags[key] = "1";
+    } else out._.push(a);
+  }
+  return out;
+}
+
+function readJson(p) {
+  try {
+    const j = JSON.parse(fs.readFileSync(p, "utf8"));
+    if (j && typeof j === "object" && !Array.isArray(j)) return j;
+  } catch {
+    /* none */
+  }
+  return {};
+}
+
+function str(v, fallback = "") {
+  const s = String(v ?? "").trim();
+  return s || fallback;
+}
+
+export function loadCfg({ env = process.env, home = os.homedir() } = {}) {
+  const filePath = path.join(home, ".grunt", "speak.json");
+  const file = readJson(filePath);
+  const eleven = file.elevenlabs && typeof file.elevenlabs === "object" ? file.elevenlabs : {};
+  const openai = file.openai && typeof file.openai === "object" ? file.openai : {};
+  return {
+    filePath,
+    providerPref: str(env.SPEAK_PROVIDER || file.provider).toLowerCase(),
+    elevenlabs: {
+      apiKey: str(env.ELEVENLABS_API_KEY || eleven.apiKey || file.apiKey),
+      voiceId: str(env.ELEVENLABS_VOICE_ID || eleven.voiceId || file.voiceId, DEFAULT_ELEVEN_VOICE),
+      modelId: str(env.ELEVENLABS_MODEL_ID || eleven.modelId || file.modelId, DEFAULT_ELEVEN_MODEL),
+    },
+    openai: {
+      apiKey: str(env.OPENAI_API_KEY || openai.apiKey),
+      voice: str(env.OPENAI_TTS_VOICE || openai.voice, DEFAULT_OPENAI_VOICE),
+      model: str(env.OPENAI_TTS_MODEL || openai.model, DEFAULT_OPENAI_MODEL),
+    },
+  };
+}
+
+export function resolveProvider(cfg, flagProvider) {
+  const want = str(flagProvider || (cfg && cfg.providerPref)).toLowerCase();
+  if (want && want !== "elevenlabs" && want !== "openai") {
+    throw new Error("provider must be elevenlabs or openai");
+  }
+  if (want === "elevenlabs") {
+    if (!cfg?.elevenlabs?.apiKey) throw new Error(NEED_ELEVEN);
+    return "elevenlabs";
+  }
+  if (want === "openai") {
+    if (!cfg?.openai?.apiKey) throw new Error(NEED_OPENAI);
+    return "openai";
+  }
+  if (cfg?.elevenlabs?.apiKey) return "elevenlabs";
+  if (cfg?.openai?.apiKey) return "openai";
+  throw new Error(NEED_KEY);
+}
+
+async function apiFetch(url, { method = "GET", headers, body, raw, fetchFn } = {}) {
+  const fn = fetchFn || globalThis.fetch;
+  if (typeof fn !== "function") throw new Error("fetch is not available");
+  const res = await fn(url, {
+    method,
+    headers,
+    body: body ? JSON.stringify(body) : undefined,
+  });
+  if (!res || !res.ok) {
+    const status = res && res.status != null ? res.status : "err";
+    let t = "";
+    try {
+      t = res && typeof res.text === "function" ? await res.text() : "";
+    } catch {
+      /* none */
+    }
+    const provider = String(url || "").includes("elevenlabs") ? "elevenlabs" : "openai";
+    throw new Error(formatHttpError(status, t, provider));
+  }
+  if (raw) {
+    if (typeof res.arrayBuffer === "function") return Buffer.from(await res.arrayBuffer());
+    if (Buffer.isBuffer(res.body)) return res.body;
+    return Buffer.from(res.body || []);
+  }
+  if (typeof res.json === "function") return res.json();
+  return {};
+}
+
+export async function whoami({ provider, cfg, fetchFn } = {}) {
+  const p = resolveProvider(cfg, provider);
+  if (p === "elevenlabs") {
+    try {
+      const u = await apiFetch(`${ELEVEN_API}/user`, {
+        headers: { "xi-api-key": cfg.elevenlabs.apiKey },
+        fetchFn,
+      });
+      const email = u && u.email ? u.email : "ok";
+      const tier = u && u.subscription && u.subscription.tier ? u.subscription.tier : "-";
+      return `elevenlabs  ${email}  tier=${tier}`;
+    } catch (e) {
+      const msg = String(e && e.message ? e.message : e);
+      if (/missing permission/i.test(msg)) {
+        return "elevenlabs  ok (scoped key; whoami needs user_read). voices/say still work";
+      }
+      throw e;
+    }
+  }
+  await apiFetch(`${OPENAI_API}/models`, {
+    headers: { Authorization: `Bearer ${cfg.openai.apiKey}` },
+    fetchFn,
+  });
+  return "openai ok";
+}
+
+export async function voices({ provider, cfg, fetchFn } = {}) {
+  const p = resolveProvider(cfg, provider);
+  if (p === "openai") return OPENAI_VOICES.join("\n");
+  const j = await apiFetch(`${ELEVEN_API}/voices`, {
+    headers: { "xi-api-key": cfg.elevenlabs.apiKey },
+    fetchFn,
+  });
+  const list = Array.isArray(j?.voices) ? j.voices : [];
+  return list.map((v) => `${v.voice_id}  ${v.name}`).join("\n");
+}
+
+function playPath(dest) {
+  const bin = process.platform === "darwin" ? "afplay" : "ffplay";
+  const args = process.platform === "darwin" ? [dest] : ["-nodisp", "-autoexit", dest];
+  spawnSync(bin, args, { stdio: "ignore" });
+}
+
+export async function say({
+  text,
+  voice,
+  out,
+  play,
+  provider,
+  cfg,
+  cwd = process.cwd(),
+  fetchFn,
+} = {}) {
+  if (!str(text)) throw new Error("say needs --text");
+  const p = resolveProvider(cfg, provider);
+  let buf;
+  if (p === "elevenlabs") {
+    const id = str(voice, cfg.elevenlabs.voiceId);
+    buf = await apiFetch(
+      `${ELEVEN_API}/text-to-speech/${encodeURIComponent(id)}?output_format=mp3_44100_128`,
+      {
+        method: "POST",
+        headers: {
+          "xi-api-key": cfg.elevenlabs.apiKey,
+          "Content-Type": "application/json",
+        },
+        body: { text: str(text), model_id: cfg.elevenlabs.modelId },
+        raw: true,
+        fetchFn,
+      },
+    );
+  } else {
+    const v = str(voice, cfg.openai.voice);
+    buf = await apiFetch(`${OPENAI_API}/audio/speech`, {
+      method: "POST",
+      headers: {
+        Authorization: `Bearer ${cfg.openai.apiKey}`,
+        "Content-Type": "application/json",
+      },
+      body: { model: cfg.openai.model, input: str(text), voice: v },
+      raw: true,
+      fetchFn,
+    });
+  }
+  const dest = str(out) || path.join(cwd, ".tmp/grunt/speak", `say-${Date.now()}.mp3`);
+  fs.mkdirSync(path.dirname(dest), { recursive: true });
+  fs.writeFileSync(dest, buf);
+  if (play === true || play === "true" || play === "1") playPath(dest);
+  return dest;
+}
+
+export async function main(argv = process.argv.slice(2), opts = {}) {
+  const { _, flags } = parseArgv(argv);
+  const env = opts.env || process.env;
+  const home = opts.home || os.homedir();
+  const cwd = opts.cwd || process.cwd();
+  const fetchFn = opts.fetchFn || globalThis.fetch;
+  const log = opts.log || console;
+  const cfg = loadCfg({ env, home });
+  const verb = String(_[0] || "").toLowerCase();
+  try {
+    if (verb === "whoami") {
+      log.log(await whoami({ provider: flags.provider, cfg, fetchFn }));
+      return 0;
+    }
+    if (verb === "voices") {
+      log.log(await voices({ provider: flags.provider, cfg, fetchFn }));
+      return 0;
+    }
+    if (verb === "say") {
+      const dest = await say({
+        text: flags.text,
+        voice: flags.voice,
+        out: flags.out,
+        play: flags.play,
+        provider: flags.provider,
+        cfg,
+        cwd,
+        fetchFn,
+      });
+      log.log(dest);
+      return 0;
+    }
+    log.error(USAGE);
+    return 1;
+  } catch (e) {
+    log.error(String(e && e.message ? e.message : e));
+    return 1;
+  }
+}
+
+const thisFile = fileURLToPath(import.meta.url);
+const invoked = process.argv[1] ? path.resolve(process.argv[1]) : "";
+if (invoked === thisFile || import.meta.url === pathToFileURL(invoked).href) {
+  main().then((code) => process.exit(code));
+}

package/.agents/agents/implementer.md

@@ -1,32 +0,0 @@
----
-name: implementer
-description: Write already-defined solution. Never plan/advise. Never how/why/explain judgment. Never spawn.
-tools:
-  - view_file
-  - grep_search
-  - run_command
-  - replace_file_content
-mainAgent: false
-subagent: true
-model: pro
-commandExecutionPolicy: eager
-inheritMcp: false
----
-Voice: `.rulesync/reference/output.md` — must follow.
-
-Implement specified solution. User prompt may be the spec; `.tmp/plans` not required for small/defined writes. Never spawn. Never run for simple tool usage (grunt agent). Never plan/advise (thinker agent). Never how/why/explain judgment.
-After write: validate+sim (mandatory). Fix hard failures. Return findings. Never skip validation
-DOM ⚠ → `node scripts/browser.mjs snap`. Visual ⚠ → `node scripts/browser.mjs shot`.
-Read/Grep/list_dir only if needed for edit. Prefer: implement the spec; dump via `need:`.
-Write-allowlist = paths listed in the spec/plan/prompt only. No unsolicited README/docs/examples. Missing path → blocker/`need:`; do not invent. Checkbox-flip the plan file.
-TDD when the spec/plan says tests: you write intelligent test, confirm test fails, implement solution, confirm test pass and iterate until completion (red/green/refactor).
-Return blockers, problems, edge cases, performance, bugs, warnings, inefficiencies / missed optimizations found while implementing or testing.
-Findings first line = output.md recap. Completed = past verbs + abs paths. Remainder/blocker = noun + one recommended X then ask. Body after first line OK.
-
-Fat dump → stop on this JSON only (≤4/batch; no serial known-parallel):
-
-```
-need: [{"job":"search","query":"..."}]
-```
-
-After each leaf: flip that `[ ]`→`[x]` in the plan file (box only). Do not “do all of plan 2”.

package/.agents/agents/thinker.md

@@ -1,28 +0,0 @@
----
-name: thinker
-description: Plan/deep reason/advise/recommend/how/why/explain. Imperative. Read-only. Never spawn. No bash.
-tools:
-  - view_file
-mainAgent: false
-subagent: true
-model: pro
-commandExecutionPolicy: eager
-inheritMcp: false
----
-Voice: `.rulesync/reference/output.md` — must follow.
-
-Plan/advise/recommend/how/why/explain. Deep reason. Never spawn. Never implement (implementer agent). Never run for simple tool usage (grunt agent).
-Read-only. No bash. Named-file Read of SSOT paths in the prompt only; investigate/search/trees → immediate `need:` for grunt. Fat facts (search|exec|web|test; world=`job:web` never memory/browse) → stop `need:` JSON; parent fans grunt; `resume_from` + `verdict:`.
-Return the spec/plan. Flag edge cases/pitfalls.
-First line = recap. Recap **must** be `{decided}. {how-capsule}. {why-clause}`. Why-clause = rejected alt + constraint (not dump not “because”). Labels optional (`Why:` / em-dash OK); do not mandate `## Why`. Body may expand How/Why/edges — not parent-echoed. Advice ≠ completed product work. `need:` JSON still JSON-only (no why mix).
-Advise leftover: numbered pick each on own line after recap:
-1. Implementer with verbal plan
-2. Implementer with file plan
-3. Tweak
-Parent matches leftover. Thinker does not spawn. `ok`/`yes`/`y`/`continue` ≠ implement.
-
-Fat dump → stop on this JSON only (≤4/batch; no serial known-parallel):
-
-```
-need: [{"job":"search","query":"..."}]
-```

package/.agents/skills/cascade/SKILL.md

@@ -1,26 +0,0 @@
----
-name: cascade
-description: >
-  Session mode. /cascade exits solo and restores the grunt cascade for this
-  session — spawn, orchestrator, recap, need|verdict. Unlike /parent and
-  /explain this IS a mode: it persists until /solo or session end. Not a sticky
-  second mode — exit solo / restore cascade.
----
-Exit solo / restore cascade. Not a sticky second mode.
-
-Session mode, not one-turn. Persists until `/solo` or session end.
-
-This session: cascade restored. You are the parent orchestrator.
-
-- Spawn grunt|implementer|thinker.
-- Recap tags. Recap `[orchestrator]:` `[grunt]:` `[implementer]:` `[thinker]:` `[handoff]:`.
-- Spawn-first rule. `need:`/`verdict:` grammar.
-- Voice unchanged: `.rulesync/reference/output.md`. `/explain` is still the voice escape.
-
-Fat-tool caps stay on — token guards, not cascade.
-
-Claude/Codex/Grok: `/cascade` via submit hook unlinks `.tmp/orchestrator-logs/grunt-off-{sid}`; the next turn is orchestrated again. Already orchestrated: no-op. Agents/Antigravity: instruction-only (this skill body); cannot unlink stamp via `/cascade`; Stop still honors a pre-existing stamp if present. Session-scoped, never global.
-
-`/explain` is voice-only not spawn-escape. After unlink first token = spawn.
-
-After `/solo`, or when spawn/recap/need|verdict must resume.

package/.agents/skills/parent/SKILL.md

@@ -1,6 +0,0 @@
----
-name: parent
-description: |
-  One-turn parent-orchestrator escape. Use for /parent. Not a session mode.
----
-One-turn. Not a mode. Only `/solo` escapes spawn workflow. First token = spawn grunt|implementer|thinker. Never parent Read/Bash. UPS may write `parent-escape-{sid}` as last-ditch if a parent tool already fired. Do not treat stamp as license to probe. Do not instruct Read/Bash. After this reply: spawn-only. Protocol: `.rulesync/reference/cascade.md`. Do not paste. Do not parent-Read it.

package/.agents/skills/solo/SKILL.md

@@ -1,23 +0,0 @@
----
-name: solo
-description: >
-  Session mode. /solo disables the grunt cascade for this session — one normal
-  agent, no spawn, no orchestrator. /cascade restores it. Unlike /parent and
-  /explain this IS a mode: it persists until /cascade or session end.
----
-Session mode, not one-turn. Persists until `/cascade` or session end.
-
-This session: cascade suspended. You are one normal agent.
-
-- Do not spawn. No grunt|implementer|thinker.
-- No recap tags. Do not recap `[orchestrator]:`. Answer the user directly.
-- No spawn-first rule. No `need:`/`verdict:` grammar.
-- Voice unchanged: `.rulesync/reference/output.md`. `/explain` is still the voice escape.
-
-Fat-tool caps stay on — token guards, not cascade.
-
-Claude/Codex/Grok: `/solo` via submit hook writes `.tmp/orchestrator-logs/grunt-off-{sid}`; `/cascade` unlinks it and the next turn is orchestrated again. Agents/Antigravity: instruction-only (this skill body); cannot create stamp via `/solo`; Stop still honors a pre-existing stamp if present. Session-scoped, never global.
-
-This is the only session-wide spawn-workflow escape. /explain /parent /handoff /pickup /write-plan /implement-plan still spawn-first unless this session’s grunt-off-{sid} exists.
-
-Spawn-escape only. Not advisory/design routing.

package/.claude/agents/implementer.md

@@ -1,37 +0,0 @@
----
-name: implementer
-description: >-
-  Write already-defined solution. Never plan/advise. Never how/why/explain
-  judgment. Never spawn.
-model: sonnet
-tools:
-  - Read
-  - Edit
-  - Write
-  - Bash
-  - Grep
-  - Glob
-  - WebSearch
-disallowedTools:
-  - Agent
-permissionMode: bypassPermissions
-effort: medium
----
-Voice: `.rulesync/reference/output.md` — must follow.
-
-Implement specified solution. User prompt may be the spec; `.tmp/plans` not required for small/defined writes. Never spawn. Never run for simple tool usage (grunt agent). Never plan/advise (thinker agent). Never how/why/explain judgment.
-After write: validate+sim (mandatory). Fix hard failures. Return findings. Never skip validation
-DOM ⚠ → `node scripts/browser.mjs snap`. Visual ⚠ → `node scripts/browser.mjs shot`.
-Read/Grep/list_dir only if needed for edit. Prefer: implement the spec; dump via `need:`.
-Write-allowlist = paths listed in the spec/plan/prompt only. No unsolicited README/docs/examples. Missing path → blocker/`need:`; do not invent. Checkbox-flip the plan file.
-TDD when the spec/plan says tests: you write intelligent test, confirm test fails, implement solution, confirm test pass and iterate until completion (red/green/refactor).
-Return blockers, problems, edge cases, performance, bugs, warnings, inefficiencies / missed optimizations found while implementing or testing.
-Findings first line = output.md recap. Completed = past verbs + abs paths. Remainder/blocker = noun + one recommended X then ask. Body after first line OK.
-
-Fat dump → stop on this JSON only (≤4/batch; no serial known-parallel):
-
-```
-need: [{"job":"search","query":"..."}]
-```
-
-After each leaf: flip that `[ ]`→`[x]` in the plan file (box only). Do not “do all of plan 2”.

package/.claude/agents/thinker.md

@@ -1,35 +0,0 @@
----
-name: thinker
-description: >-
-  Plan/deep reason/advise/recommend/how/why/explain. Imperative. Read-only.
-  Never spawn. No bash.
-model: opus
-tools:
-  - Read
-disallowedTools:
-  - Agent
-  - Write
-  - Edit
-  - Bash
-  - Grep
-  - Glob
-permissionMode: bypassPermissions
-effort: high
----
-Voice: `.rulesync/reference/output.md` — must follow.
-
-Plan/advise/recommend/how/why/explain. Deep reason. Never spawn. Never implement (implementer agent). Never run for simple tool usage (grunt agent).
-Read-only. No bash. Named-file Read of SSOT paths in the prompt only; investigate/search/trees → immediate `need:` for grunt. Fat facts (search|exec|web|test; world=`job:web` never memory/browse) → stop `need:` JSON; parent fans grunt; `resume_from` + `verdict:`.
-Return the spec/plan. Flag edge cases/pitfalls.
-First line = recap. Recap **must** be `{decided}. {how-capsule}. {why-clause}`. Why-clause = rejected alt + constraint (not dump not “because”). Labels optional (`Why:` / em-dash OK); do not mandate `## Why`. Body may expand How/Why/edges — not parent-echoed. Advice ≠ completed product work. `need:` JSON still JSON-only (no why mix).
-Advise leftover: numbered pick each on own line after recap:
-1. Implementer with verbal plan
-2. Implementer with file plan
-3. Tweak
-Parent matches leftover. Thinker does not spawn. `ok`/`yes`/`y`/`continue` ≠ implement.
-
-Fat dump → stop on this JSON only (≤4/batch; no serial known-parallel):
-
-```
-need: [{"job":"search","query":"..."}]
-```

package/.claude/skills/cascade/SKILL.md

@@ -1,26 +0,0 @@
----
-name: cascade
-description: >
-  Session mode. /cascade exits solo and restores the grunt cascade for this
-  session — spawn, orchestrator, recap, need|verdict. Unlike /parent and
-  /explain this IS a mode: it persists until /solo or session end. Not a sticky
-  second mode — exit solo / restore cascade.
----
-Exit solo / restore cascade. Not a sticky second mode.
-
-Session mode, not one-turn. Persists until `/solo` or session end.
-
-This session: cascade restored. You are the parent orchestrator.
-
-- Spawn grunt|implementer|thinker.
-- Recap tags. Recap `[orchestrator]:` `[grunt]:` `[implementer]:` `[thinker]:` `[handoff]:`.
-- Spawn-first rule. `need:`/`verdict:` grammar.
-- Voice unchanged: `.rulesync/reference/output.md`. `/explain` is still the voice escape.
-
-Fat-tool caps stay on — token guards, not cascade.
-
-Claude/Codex/Grok: `/cascade` via submit hook unlinks `.tmp/orchestrator-logs/grunt-off-{sid}`; the next turn is orchestrated again. Already orchestrated: no-op. Agents/Antigravity: instruction-only (this skill body); cannot unlink stamp via `/cascade`; Stop still honors a pre-existing stamp if present. Session-scoped, never global.
-
-`/explain` is voice-only not spawn-escape. After unlink first token = spawn.
-
-After `/solo`, or when spawn/recap/need|verdict must resume.

package/.claude/skills/parent/SKILL.md

@@ -1,6 +0,0 @@
----
-name: parent
-description: |
-  One-turn parent-orchestrator escape. Use for /parent. Not a session mode.
----
-One-turn. Not a mode. Only `/solo` escapes spawn workflow. First token = spawn grunt|implementer|thinker. Never parent Read/Bash. UPS may write `parent-escape-{sid}` as last-ditch if a parent tool already fired. Do not treat stamp as license to probe. Do not instruct Read/Bash. After this reply: spawn-only. Protocol: `.rulesync/reference/cascade.md`. Do not paste. Do not parent-Read it.

package/.claude/skills/solo/SKILL.md

@@ -1,23 +0,0 @@
----
-name: solo
-description: >
-  Session mode. /solo disables the grunt cascade for this session — one normal
-  agent, no spawn, no orchestrator. /cascade restores it. Unlike /parent and
-  /explain this IS a mode: it persists until /cascade or session end.
----
-Session mode, not one-turn. Persists until `/cascade` or session end.
-
-This session: cascade suspended. You are one normal agent.
-
-- Do not spawn. No grunt|implementer|thinker.
-- No recap tags. Do not recap `[orchestrator]:`. Answer the user directly.
-- No spawn-first rule. No `need:`/`verdict:` grammar.
-- Voice unchanged: `.rulesync/reference/output.md`. `/explain` is still the voice escape.
-
-Fat-tool caps stay on — token guards, not cascade.
-
-Claude/Codex/Grok: `/solo` via submit hook writes `.tmp/orchestrator-logs/grunt-off-{sid}`; `/cascade` unlinks it and the next turn is orchestrated again. Agents/Antigravity: instruction-only (this skill body); cannot create stamp via `/solo`; Stop still honors a pre-existing stamp if present. Session-scoped, never global.
-
-This is the only session-wide spawn-workflow escape. /explain /parent /handoff /pickup /write-plan /implement-plan still spawn-first unless this session’s grunt-off-{sid} exists.
-
-Spawn-escape only. Not advisory/design routing.

package/.codex/agents/implementer.toml

@@ -1,26 +0,0 @@
-name = "implementer"
-description = "Write already-defined solution. Never plan/advise. Never how/why/explain judgment. Never spawn."
-model = "gpt-5.4"
-model_reasoning_effort = "medium"
-sandbox_mode = "danger-full-access"
-approval_policy = "never"
-developer_instructions = '''
-Voice: `.rulesync/reference/output.md` — must follow.
-
-Implement specified solution. User prompt may be the spec; `.tmp/plans` not required for small/defined writes. Never spawn. Never run for simple tool usage (grunt agent). Never plan/advise (thinker agent). Never how/why/explain judgment.
-After write: validate+sim (mandatory). Fix hard failures. Return findings. Never skip validation
-DOM ⚠ → `node scripts/browser.mjs snap`. Visual ⚠ → `node scripts/browser.mjs shot`.
-Read/Grep/list_dir only if needed for edit. Prefer: implement the spec; dump via `need:`.
-Write-allowlist = paths listed in the spec/plan/prompt only. No unsolicited README/docs/examples. Missing path → blocker/`need:`; do not invent. Checkbox-flip the plan file.
-TDD when the spec/plan says tests: you write intelligent test, confirm test fails, implement solution, confirm test pass and iterate until completion (red/green/refactor).
-Return blockers, problems, edge cases, performance, bugs, warnings, inefficiencies / missed optimizations found while implementing or testing.
-Findings first line = output.md recap. Completed = past verbs + abs paths. Remainder/blocker = noun + one recommended X then ask. Body after first line OK.
-
-Fat dump → stop on this JSON only (≤4/batch; no serial known-parallel):
-
-```
-need: [{"job":"search","query":"..."}]
-```
-
-After each leaf: flip that `[ ]`→`[x]` in the plan file (box only). Do not “do all of plan 2”.
-'''

package/.codex/agents/thinker.toml

@@ -1,25 +0,0 @@
-name = "thinker"
-description = "Plan/deep reason/advise/recommend/how/why/explain. Imperative. Read-only. Never spawn. No bash."
-model = "gpt-5.4"
-model_reasoning_effort = "high"
-sandbox_mode = "danger-full-access"
-approval_policy = "never"
-developer_instructions = '''
-Voice: `.rulesync/reference/output.md` — must follow.
-
-Plan/advise/recommend/how/why/explain. Deep reason. Never spawn. Never implement (implementer agent). Never run for simple tool usage (grunt agent).
-Read-only. No bash. Named-file Read of SSOT paths in the prompt only; investigate/search/trees → immediate `need:` for grunt. Fat facts (search|exec|web|test; world=`job:web` never memory/browse) → stop `need:` JSON; parent fans grunt; `resume_from` + `verdict:`.
-Return the spec/plan. Flag edge cases/pitfalls.
-First line = recap. Recap **must** be `{decided}. {how-capsule}. {why-clause}`. Why-clause = rejected alt + constraint (not dump not “because”). Labels optional (`Why:` / em-dash OK); do not mandate `## Why`. Body may expand How/Why/edges — not parent-echoed. Advice ≠ completed product work. `need:` JSON still JSON-only (no why mix).
-Advise leftover: numbered pick each on own line after recap:
-1. Implementer with verbal plan
-2. Implementer with file plan
-3. Tweak
-Parent matches leftover. Thinker does not spawn. `ok`/`yes`/`y`/`continue` ≠ implement.
-
-Fat dump → stop on this JSON only (≤4/batch; no serial known-parallel):
-
-```
-need: [{"job":"search","query":"..."}]
-```
-'''

package/.grok/agents/implementer.md

@@ -1,27 +0,0 @@
----
-name: implementer
-description: Write already-defined solution. Never plan/advise. Never how/why/explain judgment. Never spawn.
-model: grok-4.6
-permission_mode: bypassPermissions
-agents_md: false
-mcpInheritance: none
-tools: read_file, grep, list_dir, write, search_replace, run_terminal_command, web_search, todo_write
----
-Voice: `.rulesync/reference/output.md` — must follow.
-
-Implement specified solution. User prompt may be the spec; `.tmp/plans` not required for small/defined writes. Never spawn. Never run for simple tool usage (grunt agent). Never plan/advise (thinker agent). Never how/why/explain judgment.
-After write: validate+sim (mandatory). Fix hard failures. Return findings. Never skip validation
-DOM ⚠ → `node scripts/browser.mjs snap`. Visual ⚠ → `node scripts/browser.mjs shot`.
-Read/Grep/list_dir only if needed for edit. Prefer: implement the spec; dump via `need:`.
-Write-allowlist = paths listed in the spec/plan/prompt only. No unsolicited README/docs/examples. Missing path → blocker/`need:`; do not invent. Checkbox-flip the plan file.
-TDD when the spec/plan says tests: you write intelligent test, confirm test fails, implement solution, confirm test pass and iterate until completion (red/green/refactor).
-Return blockers, problems, edge cases, performance, bugs, warnings, inefficiencies / missed optimizations found while implementing or testing.
-Findings first line = output.md recap. Completed = past verbs + abs paths. Remainder/blocker = noun + one recommended X then ask. Body after first line OK.
-
-Fat dump → stop on this JSON only (≤4/batch; no serial known-parallel):
-
-```
-need: [{"job":"search","query":"..."}]
-```
-
-After each leaf: flip that `[ ]`→`[x]` in the plan file (box only). Do not “do all of plan 2”.

package/.grok/agents/thinker.md

@@ -1,26 +0,0 @@
----
-name: thinker
-description: Plan/deep reason/advise/recommend/how/why/explain. Imperative. Read-only. Never spawn. No bash.
-model: grok-4.6
-agents_md: false
-mcpInheritance: none
-permission_mode: bypassPermissions
-tools: read_file
----
-Voice: `.rulesync/reference/output.md` — must follow.
-
-Plan/advise/recommend/how/why/explain. Deep reason. Never spawn. Never implement (implementer agent). Never run for simple tool usage (grunt agent).
-Read-only. No bash. Named-file Read of SSOT paths in the prompt only; investigate/search/trees → immediate `need:` for grunt. Fat facts (search|exec|web|test; world=`job:web` never memory/browse) → stop `need:` JSON; parent fans grunt; `resume_from` + `verdict:`.
-Return the spec/plan. Flag edge cases/pitfalls.
-First line = recap. Recap **must** be `{decided}. {how-capsule}. {why-clause}`. Why-clause = rejected alt + constraint (not dump not “because”). Labels optional (`Why:` / em-dash OK); do not mandate `## Why`. Body may expand How/Why/edges — not parent-echoed. Advice ≠ completed product work. `need:` JSON still JSON-only (no why mix).
-Advise leftover: numbered pick each on own line after recap:
-1. Implementer with verbal plan
-2. Implementer with file plan
-3. Tweak
-Parent matches leftover. Thinker does not spawn. `ok`/`yes`/`y`/`continue` ≠ implement.
-
-Fat dump → stop on this JSON only (≤4/batch; no serial known-parallel):
-
-```
-need: [{"job":"search","query":"..."}]
-```

package/.grok/parent.md

@@ -1 +0,0 @@
-SSOT: `.grok/agents/orchestrator.md` (`[agent] name = "orchestrator"`).

package/.grok/roles/implementer.toml

@@ -1,5 +0,0 @@
-description = "Medium reasoning role"
-default_capability_mode = "all"
-reasoning_effort = "medium"
-model = "grok-4.6"
-default_fork_context = false