deepseek-coordination 0.1.5 → 0.1.6

package/README.md

@@ -85,7 +85,7 @@ deepseek-coordination/
 ├─ tests/                          golden (Windows/POSIX/desktop injection), table conformance, encoding, plugin, permission, takeover, workflow suites
 │  ├─ conformance.ps1              command.txt and spec/verbs.tsv must describe the same surface (5 cases)
 │  ├─ takeover.test.mjs            byte-level format preservation, read windowing, eligibility (41 cases)
-│  ├─ workflow.test.mjs            phase gates, the user's choice, and the KV-cache invariant (30 cases)
+│  ├─ workflow.test.mjs            phase gates, the user's choice, and the KV-cache invariant (31 cases)
 │  ├─ smoke-desktop-sh.sh          what the POSIX desktop verbs actually inject (52 cases)
 │  └─ fake-xdotool, fake-wmctrl    recording stand-ins, so the desktop verbs can be verified with no X server
 └─ permission.json                 authorization data (an empty file in the repository; that is the initial state)
@@ -914,7 +914,7 @@ npm run test:all        # runs everything below, in order
 | Plugin | `node tests/plugin.test.mjs` | 13 | **13 passed, 0 failed** |
 | Takeover interceptions | `node tests/takeover.test.mjs` | 41 | **41 passed, 0 failed** |
 | Permission delegate | `node tests/permission.test.mjs` | 18 | **18 passed, 0 failed** |
-| Workflow | `node tests/workflow.test.mjs` | 30 | **30 passed, 0 failed** |
+| Workflow | `node tests/workflow.test.mjs` | 31 | **31 passed, 0 failed** |
 | Bundle resolution | `node tools/verify-bundle.mjs` | 16 | **16 passed, 0 failed** |
 | Package completeness | `node tools/verify-package.mjs` | 21 | **21 passed, 0 failed** |
 | Plugin load | `node tools/verify-plugin-load.mjs` | 30 | **30 passed, 0 failed** |
@@ -1068,10 +1068,11 @@ Copyright (c) 2026 Eloy Kim
 ```
 
 The licence file is pure ASCII, LF-terminated and BOM-free, consistent with the rest of the
-repository's encoding contracts. One earlier description needs correcting: it is **not** listed in
-`package.json`'s `files` whitelist (which has 20 entries; it includes `src/host/bin/dsh-os7.ps1` but
-neither `LICENSE` nor `README.zh-CN.md`), so the file stays in a checkout but is not distributed with
-the published package.
+repository's encoding contracts. An earlier description here claimed it was **not** distributed because it
+is not listed in `package.json`'s `files` whitelist. That was wrong, and the published tarball proves it:
+`npm pack deepseek-coordination@0.1.5` contains `package/LICENSE` **and** `package/README.zh-CN.md`, because
+npm always includes those two names regardless of `files`. The whitelist has 20 entries and includes
+`src/host/bin/dsh-os7.ps1`; the two files simply do not need to be in it.
 
 The verb contract and the specification documents are part of the source: a behaviour change should be
 reflected in `src/host/spec/verbs.tsv`, `src/host/command.txt` and `src/host/docs/os-spec.md`, not only

package/README.zh-CN.md

@@ -603,7 +603,7 @@ npm run test:all        # 按顺序跑下面全部
 | 插件 | `node tests/plugin.test.mjs` | 13 | **13 通过，0 失败** |
 | 接管拦截 | `node tests/takeover.test.mjs` | 41 | **41 通过，0 失败** |
 | 授权委托 | `node tests/permission.test.mjs` | 18 | **18 通过，0 失败** |
-| 工作流 | `node tests/workflow.test.mjs` | 30 | **30 通过，0 失败** |
+| 工作流 | `node tests/workflow.test.mjs` | 31 | **31 通过，0 失败** |
 | bundle 解析 | `node tools/verify-bundle.mjs` | 16 | **16 通过，0 失败** |
 | 打包完整性 | `node tools/verify-package.mjs` | 21 | **21 通过，0 失败** |
 | 插件装载 | `node tools/verify-plugin-load.mjs` | 30 | **30 通过，0 失败** |

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "deepseek-coordination",
-  "version": "0.1.5",
+  "version": "0.1.6",
   "private": false,
   "type": "module",
   "description": "Cross-platform atomic OS-interaction surface for the DeepSeek Harness: one argument-stable verb set implemented natively on each OS, with byte-exact charset, BOM and line-ending handling.",

package/src/plugin/workflow.js

@@ -1199,20 +1199,31 @@ export function workflowSectionText() {
  * The tool description.
  *
  * This is now the ONLY discovery path, because the prompt section is a frozen constant and cannot carry
- * guidance. It is in the tool catalog, which sits inside the cached prefix and is sent once — so clarity
- * here is cheaper than brevity, but it still has to be short enough not to bloat the prefix.
+ * guidance. Two consequences it has to honour, and the first version of this text failed the first one:
+ *
+ *   1. it must NAME the tool. The earlier text said "Four calls:" and then listed `start` / `plan` /
+ *      `choose` / `run`, which reads as four tool names — a model following it would call tools that do not
+ *      exist. Caught by verifying the packed tarball rather than the checkout.
+ *   2. every call in the shortest path is spelled as `workflow_plan(action: "...")`, so the model never has
+ *      to guess how the pieces fit together. The per-phase guidance is deliberately not here: it travels in
+ *      each result's `next` field, which is appended history and therefore costs the cache nothing.
+ *
+ * It lives in the tool catalog, inside the cached prefix, sent once and unchanged for the session.
  */
 const TOOL_DESCRIPTION = [
-  'Intent-first workflow: agree on what is wanted and how it will be done, before doing it. Use it when a',
-  'request is more than one obvious step, or when you would have to guess the user\'s intent, their',
-  'preferred approach, or where the work belongs. Four calls:',
-  '"start" (the prompt verbatim, plus 1-3 questions about intent OR the assumptions you are making) ->',
-  '"plan" (approach and process, at least two alternatives with pros and cons, your recommendation) ->',
-  '"choose" (ASKS THE USER; you cannot answer it for them) ->',
-  '"run" (execute the chosen option; add "outcome" to close the workflow in the same call).',
-  'Then "note" for a decision, "checkpoint" at a boundary. Long documents go to files and only a line',
-  'returns, so the conversation stays small; results list the files. A step out of order is refused, and the',
-  'refusal names the phase to record first. Every result carries "next", the one call to make. Also "status", "reset".',
+  'workflow_plan: agree on what is wanted and how it will be done, before doing it. Use it when a request is',
+  'more than one obvious step, or when you would have to guess the user\'s intent, their preferred approach,',
+  'or where the work belongs. Call it four times:',
+  '1. workflow_plan(action: "start") — the user\'s prompt verbatim, plus 1-3 questions about intent OR the',
+  'assumptions you are making.',
+  '2. workflow_plan(action: "plan") — the approach and process, at least two alternatives with pros and cons,',
+  'and which one you recommend.',
+  '3. workflow_plan(action: "choose") — ASKS THE USER; their answer is recorded and you cannot answer it for them.',
+  '4. workflow_plan(action: "run") — execute the chosen option; add "outcome" to close the workflow in this call.',
+  'Each result carries "next", the one call to make. Then workflow_plan(action: "note") for a decision, or',
+  '"checkpoint" at a boundary. Long documents go to files and only a line comes back, so the conversation stays',
+  'small; results list the files. A step out of order is refused, and the refusal names the phase to record',
+  'first. Also available: "status", "reset".',
 ].join(' ')
 
 const QUESTION_SCHEMA = {