@remixmate/cli 0.9.17 → 0.9.18

package/dist/manifest.json

@@ -1,7 +1,7 @@
 {
   "schemaVersion": 1,
-  "version": "0.9.17",
-  "generatedAt": "2026-08-29T15:20:08.872Z",
+  "version": "0.9.18",
+  "generatedAt": "2026-08-30T03:02:58.749Z",
   "skills": [
     {
       "id": "export-jianying",

package/dist/registry.d.ts

@@ -20,6 +20,8 @@ export interface SkillDef {
     createsTake: boolean;
     /** Attach to an existing take, never create one — see skill-schema.ts. */
     joinsTake: boolean;
+    /** Argv keys that downgrade createsTake to joinsTake — see skill-schema.ts. */
+    createsTakeUnless: string[];
     skillDir: string;
     /** Absolute path to the python entry script (only when entry.type === 'python'). */
     scriptAbsolutePath?: string;

package/dist/registry.js

@@ -40,6 +40,7 @@ function loadOne(skillDir) {
         auth: resolveAuthMode(raw),
         createsTake: raw.createsTake === true,
         joinsTake: raw.joinsTake === true,
+        createsTakeUnless: Array.isArray(raw.createsTakeUnless) ? raw.createsTakeUnless : [],
         skillDir,
     };
     if (entry.type === 'python') {

package/dist/runner.d.ts

@@ -9,6 +9,7 @@
  *
  * Resolves to the child process's exit code so cli.ts can propagate it.
  */
+import { type SkillDef } from './registry.js';
 import type { ParsedArgs } from './argv.js';
 export interface RunOptions {
     /** Override the skills directory (defaults to the package's bundled skills/). */
@@ -19,3 +20,20 @@ export interface RunOptions {
     parsedArgs: ParsedArgs;
 }
 export declare function runSkill(skillName: string, opts: RunOptions): Promise<number>;
+/** What this invocation is allowed to do about attribution. */
+export type TakeMode = 'create' | 'join' | 'none';
+/**
+ * Decide the take mode from the skill's declaration *and* what this particular
+ * invocation was asked to do.
+ *
+ * The manifest flags alone are not enough, because they describe the skill, not
+ * the call: `render-video --help` prints usage and exits, and
+ * `render-video --resolve-only` stops after asset resolution. Both used to run
+ * the full createsTake path, so a help query was enough to mint an empty take
+ * and lazily create 「未分类」 — the exact clutter project/take.ts sets out to
+ * avoid ("an empty VideoProject is worse clutter than an unattributed image").
+ *
+ * Note that a downgraded run still *joins*: when its own pipeline already
+ * started a take under the same grouping key, assets it generates stay attached.
+ */
+export declare function resolveTakeMode(skill: Pick<SkillDef, 'createsTake' | 'joinsTake' | 'createsTakeUnless'>, args: ParsedArgs): TakeMode;

package/dist/runner.js

@@ -61,6 +61,31 @@ export async function runSkill(skillName, opts) {
         emitBillingFooter();
     }
 }
+/**
+ * Decide the take mode from the skill's declaration *and* what this particular
+ * invocation was asked to do.
+ *
+ * The manifest flags alone are not enough, because they describe the skill, not
+ * the call: `render-video --help` prints usage and exits, and
+ * `render-video --resolve-only` stops after asset resolution. Both used to run
+ * the full createsTake path, so a help query was enough to mint an empty take
+ * and lazily create 「未分类」 — the exact clutter project/take.ts sets out to
+ * avoid ("an empty VideoProject is worse clutter than an unattributed image").
+ *
+ * Note that a downgraded run still *joins*: when its own pipeline already
+ * started a take under the same grouping key, assets it generates stay attached.
+ */
+export function resolveTakeMode(skill, args) {
+    // Introspection — the skill prints its usage and exits without producing
+    // anything, so it must not touch the backend at all.
+    if (args.help === true || args.h === true)
+        return 'none';
+    if (skill.createsTake) {
+        const downgraded = (skill.createsTakeUnless ?? []).some((key) => args[key] === true);
+        return downgraded ? 'join' : 'create';
+    }
+    return skill.joinsTake ? 'join' : 'none';
+}
 /**
  * Attach this invocation to a take, publishing the result as `CONVERSATION_ID`.
  * That variable is the only thing that makes an uploaded asset reachable from
@@ -83,7 +108,8 @@ export async function runSkill(skillName, opts) {
  *     asset still uploads, just unattributed — exactly the old behaviour.
  */
 async function applyTakeContext(skill, opts, auth) {
-    if (!skill.createsTake && !skill.joinsTake)
+    const mode = resolveTakeMode(skill, opts.parsedArgs);
+    if (mode === 'none')
         return;
     if ((process.env.CONVERSATION_ID ?? '').trim())
         return;
@@ -95,7 +121,7 @@ async function applyTakeContext(skill, opts, auth) {
             preflightToken: auth.token,
         });
         let take;
-        if (skill.createsTake) {
+        if (mode === 'create') {
             const jobIdArg = opts.parsedArgs.job_id;
             const jobId = typeof jobIdArg === 'number' ? jobIdArg : Number(jobIdArg);
             const flagProject = opts.parsedArgs.project;

package/dist/skill-schema.d.ts

@@ -72,6 +72,15 @@ export interface RawSkillJson {
      * worse clutter than an unattributed asset.
      */
     joinsTake?: boolean;
+    /**
+     * Parsed-argv keys (snake_case, as produced by argv.ts) whose presence means
+     * *this* invocation won't finish a video — `render-video --resolve-only`
+     * resolves assets and stops. Such a run degrades from createsTake to
+     * joinsTake: it still attaches to a take its own pipeline already started,
+     * but never brings one — nor the lazily created 「未分类」 — into being for a
+     * pass that may end in a contract error and no video at all.
+     */
+    createsTakeUnless?: string[];
 }
 /**
  * Resolve a skill's auth mode, defaulting for records that predate the field.

package/dist/skill-schema.js

@@ -67,6 +67,22 @@ export function validateSkillJson(raw, skillId) {
     if (raw.joinsTake != null && typeof raw.joinsTake !== 'boolean') {
         errors.push(`skill.json.joinsTake must be a boolean, got '${String(raw.joinsTake)}'`);
     }
+    if (raw.createsTakeUnless != null) {
+        if (!Array.isArray(raw.createsTakeUnless) ||
+            raw.createsTakeUnless.some((f) => typeof f !== 'string' || f === '')) {
+            errors.push("skill.json.createsTakeUnless must be an array of non-empty argv keys");
+        }
+        else if (raw.createsTake !== true) {
+            // It only ever downgrades createsTake; on any other skill it would read as
+            // working while doing nothing.
+            errors.push("skill.json.createsTakeUnless requires createsTake=true");
+        }
+        else if (raw.createsTakeUnless.some((f) => f.includes('-'))) {
+            // argv.ts hands the dispatcher snake_case keys, so `--resolve-only`
+            // arrives as `resolve_only`. A kebab entry would silently never match.
+            errors.push("skill.json.createsTakeUnless entries must be snake_case argv keys (e.g. 'resolve_only', not 'resolve-only')");
+        }
+    }
     if (raw.createsTake === true && raw.joinsTake === true) {
         errors.push("skill.json cannot set both 'createsTake' and 'joinsTake' — createsTake already implies joining");
     }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@remixmate/cli",
-  "version": "0.9.17",
+  "version": "0.9.18",
   "description": "AI media generation skills for Claude Code / Codex — 12 skills covering image, video, voice, digital human, web screenshot, web recording, script, template registry, rendering, Jianying export, and video deconstruction.",
   "type": "module",
   "main": "dist/index.js",

package/skills/render-video/SKILL.md

@@ -114,7 +114,17 @@ python3 <SkillDir>/scripts/render_video.py --render-plan <path>.render-plan.json
 python3 <SkillDir>/scripts/render_video.py --job-id 42 --upload-title "My video"
 ```
 
-After rendering, the script auto-uploads by default; `render-manifest.json` then contains an `upload.fileUrl` field. **Prefer `upload.playbackUrl`** (transcoded) over `upload.fileUrl` (raw VOD) when surfacing the URL to a user.
+After rendering, the script auto-uploads by default; `render-manifest.json` then contains an
+`upload.fileUrl` field, and `upload.playbackUrl` (transcoded) once VOD processing finishes.
+
+**Do not retype the video URL into your reply.** On success the script prints a structured
+`__render_video_asset__` line; the host reads the authoritative URL from there and renders a
+player for the user. Tell the user the video is ready — the player appears on its own.
+
+Reproducing a 32-char opaque URL from memory is unreliable: on 2026-08-30 a single character
+was dropped (`…c1c20102` → `…c1c2012`), so the user got a 404 while the file sat fine on the
+CDN. The URL in your prose is redundant with the structured asset and is the only copy that
+can be wrong — the UI now renders it as plain, non-clickable text for exactly this reason.
 
 ### Render but skip the upload
 

package/skills/render-video/scripts/render_video.py

@@ -2413,6 +2413,26 @@ Examples:
             effective_playback = playback_url or remote_url
             if effective_playback and not effective_playback.startswith("vod://"):
                 print(f"\n🎬 Video playback URL: {effective_playback}")
+
+                # 结构化资产标记 —— 宿主（ab-agent）据此把权威 URL 作为 attachment
+                # 下发前端渲染播放器，绕开模型正文。
+                #
+                # 为什么必须绕开：模型复述 32 位不透明 hex 会出错。2026-08-30 实测一次
+                # 成片链接被吞掉一个字符（…c1c20102 → …c1c2012），用户拿到 404，
+                # 而文件本身好好地在 CDN 上。约定与 web-screenshot/scripts/record.py
+                # 的 __web_record_asset__ 一致。
+                #
+                # 只在拿到**非 vod:// 的真实播放地址**时才打标记（上面的 if 已保证）：
+                # playbackUrl 来自 VOD 轮询，可能超时未就绪。宁可前端没有内联播放器
+                # （用户仍可从成片面板看），也不要再给出一条不可用的地址。
+                asset = {"url": effective_playback}
+                cover = upload_info.get("coverUrl")
+                if cover:
+                    asset["coverUrl"] = cover
+                total_duration = (render_plan.get("renderConfig") or {}).get("totalDuration")
+                if isinstance(total_duration, (int, float)) and total_duration > 0:
+                    asset["durationSec"] = total_duration
+                print("__render_video_asset__ " + json.dumps(asset, ensure_ascii=False))
         else:
             LogPrint(f"   Video: {output_path}", file=sys.stderr)
         if args.save_job:

package/skills/render-video/skill.json

@@ -7,6 +7,7 @@
   "description": "Loads a persisted RenderPlan by job_id and drives the Remotion engine to produce the final video. Assets must already be generated via prepare_video_assets — this skill never resolves or regenerates assets.",
   "auth": "required",
   "createsTake": true,
+  "createsTakeUnless": ["resolve_only"],
   "envVars": [
     "PRIV_TOKEN",
     "MM_API_BASE_URL",