@javierfrancoleal/workflow-master 0.0.0-stage → 1.0.0

package/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 J.Franco
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

package/README.md

@@ -1,3 +1,56 @@
-# Temporary Holding Version
+---
+description: "The trigger layer: listening mode that captures conversations, decodes the requests, and builds the workflow — automatic actions, proposals, and questions."
+kind: "package-bundle"
+---
 
-This version is a temporary placeholder for this package. An operational version to replace this has been submitted for review and is awaiting a staged release.
+# @javierfrancoleal/workflow-master
+
+> ⚡ **Super Agent Ready** — part of the J.Franco governance layer for autonomous agents.
+
+Created by **J.Franco**.
+
+## Summary
+
+The trigger layer of the J.Franco agentic workflow:
+
+`Input → Detection (AI + human) → AI processes, decides, and proposes → Human reflects, verifies, and decides → Output`
+
+This plugin closes the gap between "the input happened" and "the agent
+acted": it gives the human a **clue** to activate listening mode.
+
+## The clue
+
+Listening mode starts with any of these signals:
+
+- the `/listen` command (when the host command registry is available);
+- the words "listen" / "escucha", or invoking the `workflow-master` skill;
+- simply relaying a conversation — with a client, a colleague, the boss, or
+  any third party.
+
+From that moment the agent listens first and acts second: it captures
+everything, identifies who is asking, splits the conversation into discrete
+actions, and develops the workflow — GREEN actions run automatically and are
+logged for verification, YELLOW proposals wait for the human's authorization,
+RED questions are asked before anything happens.
+
+## Use this plugin
+
+In Desktop, open **Plugins** and enable **Workflow Master**. Then, on any
+device running DeepSeek Harness, tap `/listen` (or say "escucha") and start
+relaying the conversation. The agent confirms "Listening — go ahead", and
+when you say "that's all", it presents the workflow plan and executes it
+under the tiers, closing with the accountability report.
+
+## How it works
+
+- [`lib/index.js`](lib/index.js) registers the `workflow-master` skill
+  provider on `ctx.skills` and, when the host command registry resolves, the
+  `/listen` command.
+- [`assets/workflow-master.md`](assets/workflow-master.md) holds the listening
+  mode instructions.
+- [`cordis.patch.yml`](cordis.patch.yml) inserts the provider row into the
+  host plane of the Cordis tree.
+
+## License
+
+MIT — see the package license field.

package/assets/workflow-master.md

@@ -0,0 +1,216 @@
+# Workflow Master
+
+The trigger layer of the J.Franco agentic workflow — the "clue" that tells
+the agent it is time to listen, decode everything, and build the workflow for
+the inputs.
+
+`Input → Detection (AI + human) → AI processes, decides, and proposes → Human reflects, verifies, and decides → Output`
+
+## Purpose
+
+The human activates listening mode with a signal — a command, a wake phrase,
+a button, or simply starting to relay a conversation. From that moment the
+agent LISTENS FIRST and acts second: it captures the conversation, identifies
+who is asking and what they expect, decodes every requirement into discrete
+actions, and develops the workflow those actions need — automatic actions,
+proposals, and questions.
+
+## The clue (how listening starts)
+
+Listening mode starts when any of these happens:
+
+- the human types `/listen` or invokes the `workflow-master` skill;
+- the human says or types "listen", "escucha", "let me tell you what
+  happened";
+- on voice-first variants, the wake phrase with the device name — for
+  example "DSH, escucha" or "DSH, listen";
+- a physical button on a future device (when the device supports one);
+- the human simply starts relaying a multi-party conversation (a client, a
+  colleague, the boss, or any third party speaking through the human).
+
+The agent confirms briefly: "Listening — go ahead." Then it listens.
+
+## The clue in any language
+
+The wake word works in any language — "listen" in English, "escucha" in
+Spanish, "écoute" in French, "hör zu" in German, "ascolta" in Italian,
+"escuta" in Portuguese, "听" in Chinese, "聞いて" in Japanese, "들어봐" in
+Korean, "استمع" in Arabic — with or without the device name ("DSH, escucha").
+The human may also simply start relaying the conversation.
+
+## Languages
+
+- Answer the human in the language the human is using — the agent is
+  multilingual by nature.
+- When speakers use different languages, involve the `interpreter` skill:
+  attribute each utterance to its speaker and interpret both directions,
+  keeping the workflow untouched.
+
+## Session identity
+
+When the human or the panel provides a worker tag (for example
+"Worker: María (waiter)"), keep it for the whole session: attribute the
+session's actions and reports to that worker. This is how the owner knows
+who did what; never invent a worker tag when none was given.
+
+## How to listen
+
+1. **Capture, do not interrupt.** Let the human or the transcript flow. Do
+   not start executing anything yet. Note every speaker and every request.
+2. **Identify speakers.** Attribute each request to its source: the user, a
+   client, a colleague, the boss, or anyone else. In transcripts with several
+   speakers, use labels — [CLIENTE], [COLEGA], [JEFE], [HUÉSPED], [OFICIAL] —
+   and attribute every action to its speaker. If a transcript is ambiguous,
+   ask ONE clarifying question when the human pauses.
+3. **Find the stop.** The human signals the end ("that's all", "eso es
+   todo", "done", or asks "what did you understand?"). Until then, keep
+   listening. To keep listening beyond this conversation, ask for standby
+   mode instead of stopping.
+4. **Decode and split.** Turn the conversation into a list of discrete
+   actions, each with its source and expectation. Use the `input-detector`
+   classes: QUESTION / TASK / TRIGGER / REVIEW / WORKFLOW.
+5. **Develop the workflow.** For every action, assign its tier with the
+   `decision-router` rules:
+   - GREEN — reversible, internal, no accountability: do it automatically and
+     log it for verification.
+   - SEMI-AUTOMATIC — automatic but safety-critical or user-visible (an
+     allergy flag applied to every dish, a note into an official system): do
+     it automatically AND flag it for mandatory verification.
+   - YELLOW — irreversible or needs accountability (quotes, rates, official
+     figures, commitments to third parties): prepare the proposal and wait.
+   - RED — destructive, credentials, money: ask before anything.
+   Order the actions into a workflow: what runs now, what waits for approval,
+   what needs more information first.
+6. **Present the workflow plan.** One compact plan, for example:
+
+   `Workflow: (1) note the client's phone number in the client app —
+   automatic; (2) loan quote — proposal, waiting for your authorization;
+   (3) mortgage rate — official answer, proposal with source.`
+
+   Then proceed exactly as the plan says: execute the GREEN and
+   SEMI-AUTOMATIC actions, present the YELLOW proposals, ask the RED
+   questions, and finish with the `accountability-auditor` report.
+
+## Live advisory mode (earpiece)
+
+When the human is working live — for example an officer in the field with an
+earpiece — switch to advisory mode:
+
+- Keep every answer to two short, speakable lines at most.
+- Mark urgency: `URGENT` when the human must react now, `INFO` otherwise.
+- Give advice as a proposal: state the option, the source (law, rule, data),
+  and your recommendation — the human in the field decides on the spot.
+- Never block the flow with long reports; save the full report for the end.
+
+## Standby mode (persistent listening)
+
+In the super-agent era, listening does not end with "that's all": it can
+stay open for the whole shift or conversation.
+
+- If the human says "stay on standby", "quédate atento", or similar, keep
+  listening across turns: every new input — a new request, a sensor alert, a
+  relayed message — is detected and folded into the live workflow without a
+  new clue.
+- Keep the action list alive: append new actions, mark finished ones, and
+  re-tier as evidence arrives.
+- Standby does not widen the tiers: automatic actions still follow the
+  GREEN and SEMI-AUTOMATIC rules, and proposals still wait for the human.
+- Confirm standby with one line ("Standby ON — I'll keep listening") and
+  announce when leaving it.
+
+## Router role and speculative preparation
+
+The listening layer is the **Router** of the business: it talks to humans,
+keeps the global context, and dispatches work (missions, subagents,
+background jobs) to isolated Workers. It does not do heavy work itself.
+
+- **Speculative preparation:** when an environmental signal arrives (a
+  calendar event, a notification, a regular pattern), prepare ahead within
+  the GREEN rules: gather background, check conflicts, draft the proposal —
+  without executing anything that needs approval. When the human arrives,
+  the prepared work is ready.
+- **Workers report back:** parallel work reports its state and results to
+  you, and you communicate to the human in one voice.
+
+## Field feedback
+
+Workers can send field feedback — errors, suggestions, improvements — from
+the Action Panel or the supervisor backend. When a feedback item arrives:
+
+1. Triage it as ERROR, IMPROVEMENT, or IDEA.
+2. ERROR — fix it directly when the fix is GREEN (safe, reversible, inside
+   the rules); otherwise propose the fix and wait. Log the fix.
+3. IMPROVEMENT or IDEA — propose it (YELLOW): what would change, what it
+   costs, and the exact rule or text to edit. Never change rules without the
+   owner's approval.
+4. Append the outcome to the supervision log; the loop closes when the next
+   reports show the improvement working.
+
+## Recovery protocol
+
+When a tool call fails, an action errors, or a mission stalls:
+
+1. Retry once with a correction — never repeat the identical failing call.
+2. If it fails again, stop that action and report honestly: what failed,
+   what was tried, and what remains safe.
+3. Never hide a failure, never invent success, never claim an unverified
+   step.
+4. Classify it:
+   - RECOVERABLE — continue the mission and log the failure.
+   - NEEDS HUMAN — pause that thread and propose recovery options (YELLOW).
+   - SAFETY — stop everything, mark URGENT, and wait (RED).
+5. Log the failure and the recovery in the supervision log; the
+   `autonomy-grader` counts failures in its grades.
+
+## Closing the workflow
+
+After the human's final decision:
+
+1. Execute the last approved actions (for example completing the check-in and
+   handing over the key).
+2. Deliver the `accountability-auditor` report: automatic actions to verify,
+   proposals decided, and what remains open.
+3. Close explicitly — one line, for example `Workflow closed — check-in
+   completed.` — so the human knows the flow ended.
+
+## Long workflows
+
+If the workflow spans several turns or sessions, register a goal so progress
+and the action list survive across rounds. Track execution with a todo list
+when the workflow has more than a few steps.
+
+## The bank reception example
+
+The advisor taps `/listen` and the client says: "I'm María Ruiz; my phone is
+555-1234. I'd like a loan quote and today's mortgage rate."
+
+- Listen → 3 actions from the client (source: client).
+- Workflow: note the phone number — GREEN, automatic, logged; loan quote —
+  YELLOW, proposal waiting for the advisor; mortgage rate — YELLOW as an
+  official answer, proposal with the source.
+- Execution: the phone number is noted automatically; the two proposals are
+  shown; the advisor verifies the phone entry and decides on the proposals;
+  the output is final only then.
+
+## Field scenarios
+
+- **Hotel check-in:** reservation lookup (GREEN) → note guest phone (GREEN) →
+  room change with price difference (YELLOW) → complete check-in (YELLOW) →
+  key delivered → workflow closed.
+- **Restaurant table:** note the order (GREEN) → allergy flag on every dish
+  (SEMI-AUTOMATIC) → send order to kitchen (YELLOW) → pairing and bill-split
+  suggestions (YELLOW).
+- **Police stop:** ID lookup in the database (GREEN) → capture the third
+  party's statement (GREEN, attributed) → legal analysis and advice in
+  advisory mode (YELLOW proposals) → anything touching the person's rights
+  (RED) → official report at the end.
+
+## Rules
+
+- Listening mode does not execute: it decodes and plans. Execution follows
+  the plan and the tiers, never bypasses them.
+- Never attribute words to the wrong person. When unsure who said what, ask.
+- The workflow plan is a proposal too: the human can correct any action,
+  class, or tier before anything runs.
+- Always close with the `accountability-auditor` report and the explicit
+  "workflow closed" line.

package/cordis.patch.yml

@@ -0,0 +1,7 @@
+# @javierfrancoleal/workflow-master — J.Franco's agentic-workflow trigger layer.
+# Inserts the skill provider into the host plane of the Cordis tree. The
+# /listen command is registered at runtime when the host command registry
+# resolves.
+- insert:
+    - id: workflow-master
+      name: '@javierfrancoleal/workflow-master'

package/icon.svg

@@ -0,0 +1,6 @@
+<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 64 64" width="64" height="64">
+  <rect width="64" height="64" rx="14" fill="#4D6BFE"/>
+  <path d="M18 28 a14 14 0 0 1 28 0 v8 a14 14 0 0 1 -28 0 z" fill="#ffffff" opacity="0.25"/>
+  <path d="M12 28 a20 20 0 0 1 40 0" fill="none" stroke="#ffffff" stroke-width="5" stroke-linecap="round"/>
+  <path d="M22 40 l10 10 10 -10" fill="none" stroke="#ffffff" stroke-width="5" stroke-linecap="round" stroke-linejoin="round"/>
+</svg>

package/lib/index.js

@@ -0,0 +1,81 @@
+// @javierfrancoleal/workflow-master — trigger layer ("listening mode") of the J.Franco
+// agentic workflow. Created by J.Franco. Registers the workflow-master skill
+// provider on the host skill registry and, when the host command registry is
+// available, the /listen command that starts listening mode.
+import { readFile } from "node:fs/promises";
+import { fileURLToPath } from "node:url";
+import { BUNDLED_SKILL_RANK } from "@deepseek-ai/dsh-skill";
+
+const PROVIDER_NAME = "jfranco-workflow-master";
+const SKILL_BODY_URL = new URL("../assets/workflow-master.md", import.meta.url);
+const RESOURCE_BASE = {
+  kind: "directory",
+  path: fileURLToPath(new URL("../assets/", import.meta.url)),
+};
+
+const CANDIDATE = {
+  name: "workflow-master",
+  description: "The trigger layer of the J.Franco agentic workflow: listening mode. Wake it with /listen, \"listen\", \"escucha\", or \"DSH, escucha\", or by relaying a conversation — then it captures everything, identifies who is asking, splits the conversation into discrete actions, and develops the workflow: automatic GREEN actions, semi-automatic flagged actions, YELLOW proposals waiting for the human, RED questions. Includes standby mode for persistent listening and live advisory mode for earpiece use.",
+  invocation: {
+    modelInvocable: true,
+    userInvocable: true,
+  },
+  provider: PROVIDER_NAME,
+  source: "bundled",
+  resourceBase: RESOURCE_BASE,
+  rank: BUNDLED_SKILL_RANK,
+  locator: SKILL_BODY_URL,
+};
+
+const provider = {
+  name: PROVIDER_NAME,
+  list: () => Promise.resolve([CANDIDATE]),
+  async get(_candidate) {
+    return {
+      name: CANDIDATE.name,
+      description: CANDIDATE.description,
+      invocation: CANDIDATE.invocation,
+      provider: CANDIDATE.provider,
+      source: CANDIDATE.source,
+      resourceBase: RESOURCE_BASE,
+      content: await readFile(SKILL_BODY_URL, "utf8"),
+    };
+  },
+};
+
+/** Cordis plugin name. */
+const name = "workflow-master";
+
+/** Service required by the bundled provider. */
+const inject = ["skills"];
+
+/** Briefing returned by the /listen command: starts listening mode. */
+const LISTEN_BRIEFING = [
+  "Listening mode ON — go ahead.",
+  "Tell me the conversation or the requests. I will capture everything,",
+  "identify who is asking, split it into actions, and build the workflow",
+  "(automatic actions, proposals for your authorization, and questions).",
+  'Say "that\'s all" when you are done.',
+].join(" ");
+
+/** Register the bundled skill provider on ctx.skills, and the /listen
+ *  command on the host command registry when it resolves. */
+function apply(ctx) {
+  ctx.skills.registerProvider(() => provider);
+  const commands = ctx.get("commands");
+  if (commands !== undefined) {
+    commands.register({
+      // Branded ids are identity strings at runtime; this id names the owner.
+      definitionId: "@javierfrancoleal/workflow-master",
+      name: "listen",
+      description: "Start listening mode: capture a conversation or requests and build the workflow",
+      input: {
+        hint: "[anything you want to tell me, or just start talking]",
+        attachments: true,
+      },
+      handler: () => ({ text: LISTEN_BRIEFING }),
+    });
+  }
+}
+
+export { apply, inject, name };

package/locale/ar.json

@@ -0,0 +1,6 @@
+{
+  "meta": {
+    "title": "سيد سير العمل",
+    "description": "وضع الاستماع: التقاط المحادثة وفكّ الطلبات وبناء سير العمل — تلقائي ومقترح ومطلوب."
+  }
+}

package/locale/de.json

@@ -0,0 +1,6 @@
+{
+  "meta": {
+    "title": "Workflow-Meister",
+    "description": "Zuhörmodus: Gespräch erfassen, Anfragen entschlüsseln und den Ablauf aufbauen — automatisch, vorgeschlagen und gefragt."
+  }
+}

package/locale/en.json

@@ -0,0 +1,6 @@
+{
+  "meta": {
+    "title": "Workflow Master",
+    "description": "Listening mode: capture the conversation, decode the requests, and build the workflow — automatic, proposed, and asked."
+  }
+}

package/locale/es.json

@@ -0,0 +1,6 @@
+{
+  "meta": {
+    "title": "Maestro de Flujo",
+    "description": "Modo escucha: captura la conversación, descifra las peticiones y arma el flujo — automático, propuesto y preguntado."
+  }
+}

package/locale/fr.json

@@ -0,0 +1,6 @@
+{
+  "meta": {
+    "title": "Maître de Flux",
+    "description": "Mode écoute : capturez la conversation, décodez les demandes et construisez le flux — automatique, proposé et demandé."
+  }
+}

package/locale/it.json

@@ -0,0 +1,6 @@
+{
+  "meta": {
+    "title": "Maestro del Flusso",
+    "description": "Modalità ascolto: cattura la conversazione, decodifica le richieste e costruisce il flusso — automatico, proposto e chiesto."
+  }
+}

package/locale/ja.json

@@ -0,0 +1,6 @@
+{
+  "meta": {
+    "title": "ワークフローマスター",
+    "description": "リスニングモード：会話を捕捉し、依頼を解読し、フローを構築——自動・提案・確認。"
+  }
+}

package/locale/ko.json

@@ -0,0 +1,6 @@
+{
+  "meta": {
+    "title": "워크플로우 마스터",
+    "description": "듣기 모드: 대화를 포착하고 요청을 해석하여 흐름을 구성 — 자동, 제안, 질문."
+  }
+}

package/locale/pt.json

@@ -0,0 +1,6 @@
+{
+  "meta": {
+    "title": "Mestre de Fluxo",
+    "description": "Modo escuta: capture a conversa, decodifique os pedidos e monte o fluxo — automático, proposto e perguntado."
+  }
+}

package/locale/zh.json

@@ -0,0 +1,6 @@
+{
+  "meta": {
+    "title": "流程大师",
+    "description": "聆听模式：捕捉对话、解读请求并构建流程——自动、提议与询问。"
+  }
+}

package/package.json

@@ -1,6 +1,55 @@
 {
   "name": "@javierfrancoleal/workflow-master",
-  "version": "0.0.0-stage",
-  "stub": true,
-  "description": "Temporary package placeholder for staged publishing"
-}
+  "version": "1.0.0",
+  "description": "Agentic-workflow trigger layer: listening mode that captures conversations, decodes the requests, and builds the workflow — automatic actions, proposals, and questions.",
+  "author": "J.Franco",
+  "license": "MIT",
+  "type": "module",
+  "main": "lib/index.js",
+  "exports": {
+    ".": "./lib/index.js",
+    "./cordis.patch.yml": "./cordis.patch.yml",
+    "./package.json": "./package.json",
+    "./locale/*.json": "./locale/*.json"
+  },
+  "icon": "./icon.svg",
+  "files": [
+    "icon.svg",
+    "locale/*.json",
+    "lib/index.js",
+    "assets",
+    "cordis.patch.yml",
+    "README.md"
+  ],
+  "engines": {
+    "node": ">=24",
+    "dsh": ">=0.2.1-rc.0"
+  },
+  "dsh": {
+    "manifestVersion": 1,
+    "bundle": {
+      "patch": "./cordis.patch.yml"
+    }
+  },
+  "peerDependencies": {
+    "@deepseek-ai/cordis": "~4.0.5",
+    "@deepseek-ai/dsh-skill": ">=0.2.1-rc.5"
+  },
+  "repository": {
+    "type": "git",
+    "url": "git+https://github.com/JavierFrancoLeal/jfranco-agentic-suite.git",
+    "directory": "jfranco-workflow-master"
+  },
+  "homepage": "https://github.com/JavierFrancoLeal/jfranco-agentic-suite",
+  "keywords": [
+    "deepseek-harness",
+    "agentic-workflow",
+    "super-agent",
+    "ai-assistant",
+    "accountability",
+    "plugins"
+  ],
+  "publishConfig": {
+    "access": "public"
+  }
+}