What Shall We Build Next?

Describe

Describe your task

Refine

Refine the plan

SubAgents

Review all agents

Deploy

Deploy your agent

A) SUBAGENT SUMMARY  
This subagent (“VideoTranscriber”) takes a single MP4 video URL as input and returns a full transcription of the video’s audio with accurate timestamps.

B) FINAL TASK OUTPUT  
A single timestamped transcription (e.g., a .txt or JSON structure), containing each spoken segment and its corresponding time range.

C) SUBAGENT INPUT  
• MP4 URL of the “how to” video (user-provided, referred to as [variable1]).

E) SUBAGENT TASK SUMMARY  
The overall flow for this subagent is straightforward:  
1) Receive the MP4 URL as input ([variable1]).  
2) Call Skill #207 (“Get Transcription From MP4 Video URL”) with the MP4 URL.  
3) Receive the transcription text with timestamps from Skill #207.  
4) Provide that transcription with timestamps as the subagent’s final output.

F) SILOS  
• SILO 1 – Obtain Transcription  
   – Input: MP4 URL ([variable1])  
   – Action: Use Skill #207 to transcribe the MP4 into text with timestamps  
   – Output: Timestamped transcription ([timestamped-transcription])

No additional silos are required because the subagent only needs to produce a transcription. The final result from this single silo is the subagent’s final output.

SubAgent #1 - Diagram

Expand Diagram

A) SUBAGENT SUMMARY  
"StepSelector" takes the full timestamped transcription from the video and distills it into exactly six key steps, each with a title, a timestamp, and a concise description, by leveraging a powerful LLM.

--------------------------------------------------
B) FINAL TASK OUTPUT  
A structured text (for example, JSON or well-formatted text) containing exactly six steps. Each step has:  
• stepNumber (1 through 6)  
• title (short descriptive label)  
• timestamp (HH:MM:SS or similar)  
• description (one or two sentences)

--------------------------------------------------
C) SUBAGENT INPUT  
1. The raw, timestamped transcription text ([timestamped-transcription]) generated by the "VideoTranscriber" subagent.  
2. A fixed or generated prompt instructing the LLM to produce exactly six steps, each with a title, timestamp, and description.

--------------------------------------------------
E) SUBAGENT TASK SUMMARY  
Below is the sequence of how the subagent will work internally:

1. Receive the timestamped transcription ([timestamped-transcription]).  
2. Combine this transcription with a fixed prompt that instructs the LLM to return six steps which must include a title, a timestamp, and a short description for each step.  
   - For example, the prompt could say:  
     "Analyze the transcription text below and identify exactly six significant steps for the tutorial. Each step must include:  
       • stepNumber  
       • title  
       • timestamp (in HH:MM:SS or similar)  
       • description (a concise explanation of this step)."  
3. Call Skill #223 (Powerful LLM Prompt-to-Text Response) with the above combined prompt.  
4. The LLM returns the final six-step summary ([six-steps-summary]) in JSON or well-structured text format.

Eventually, the subagent’s final output is this structured list or text of six steps:  
[six-steps-summary]

High-level flow:  
• [timestamped-transcription] → (Skill #223 with custom prompt) → [six-steps-summary]

--------------------------------------------------
F) SILOS

• Silo 1 – Generate Six-Step Summary  
   1) Input: [timestamped-transcription]  
   2) Action: Use Skill #223 with instructions to parse the transcription and return exactly 6 steps.  
   3) Output: [six-steps-summary]

No additional silos are needed, as the subagent uses only one skill call in this process.

SubAgent #2 - Diagram

Expand Diagram

A) SUBAGENT SUMMARY  
“ScreenshotExtractor” takes a video URL and six timestamps, then produces six distinct image files (screenshots) at those exact timestamps.

B) FINAL TASK OUTPUT  
The final output is a list/array of six image URLs (e.g., “screenshot_step1.png” through “screenshot_step6.png”).

C) SUBAGENT INPUT  
1. MP4 video URL ([variable1])  
2. A structured set of six steps (each with a timestamp), referred to as [six-steps-summary].

E) SUBAGENT TASK SUMMARY  
Below is the step-by-step flow for how “ScreenshotExtractor” proceeds (its internal workflow). The subagent receives [variable1] (the MP4 URL) and [six-steps-summary] (containing six timestamps). It then loops through each timestamp (one for each step) to generate a corresponding screenshot:

1) Receive (1) [variable1] (MP4 URL) and (2) [six-steps-summary] (contains six timestamps).  
2) For each of the six timestamps:  
   • Call SKILL #202 - Extract Thumbnail Images Of MP4 Video  
     INPUT:  
       – mp4 URL ([variable1])  
       – prompt requesting “extract a screenshot at [one specific timestamp].”  
     OUTPUT:  
       – A collection of thumbnail image URLs (by default up to 10, but in practice we only need the image corresponding to that timestamp).  
   • Select or rename the most relevant returned image as “screenshot_stepX.png.”  
3) Gather/append each resulting image URL (one per timestamp/step) into an array or list of six image URLs.  
4) Return that array/list of six final screenshot image URLs ([six-screenshots]).

F) SILOS  
• SILO 1: Parsing & Preparation  
  – Input subagent data: MP4 URL ([variable1]) + JSON or structured list of six timestamps ([six-steps-summary]).

• SILO 2: Extracting Screenshots  
  – For each of the six timestamps, use SKILL #202 with a prompt to capture a screenshot around that exact time.   
  – Gather 6 URLs total.

• SILO 3: Collation & Final Output  
  – Return the six URLs in a final list ([six-screenshots]).

SubAgent #3 - Diagram

Expand Flow

A) SUBAGENT SUMMARY  
"ArticleAssembler" takes the raw transcription, the six-step summary, and the six screenshot URLs, and merges them into a single, publication-ready HTML article containing exactly six steps, each with a heading, short description, and one embedded image.

B) FINAL TASK OUTPUT  
A single HTML file (final_article.html) that contains:  
• Six clearly labeled steps with titles and short descriptions  
• Each step includes one embedded screenshot image (from the six supplied)  
• Incorporates or references the broader raw transcription as needed  
• Fully self-contained, ready to publish

C) SUBAGENT INPUT  
1. [timestamped-transcription] – the raw, timestamped transcription from the video  
2. [six-steps-summary] – structured data describing six steps (title, short description, single best timestamp)  
3. [six-screenshots] – six image URLs corresponding to each step’s chosen timestamp  
4. Subagent prompt/instructions for generating final HTML

E) SUBAGENT TASK SUMMARY  
Below is the step-by-step flow of how "ArticleAssembler" will create its final output:

1) GATHER INPUTS  
   - Collect:  
       (a) The transcription text ([timestamped-transcription])  
       (b) The six-step data ([six-steps-summary])  
       (c) The six screenshot URLs ([six-screenshots])  
       (d) A short prompt or instructions (e.g., “Combine these elements into a polished HTML article with six steps and embedded images”).

2) GENERATE FINAL HTML (Skill #223 - Powerful LLM Prompt-to-Text Response)  
   - Combine all inputs into one structured prompt to the LLM, for example:  
     “Here is the raw transcription: [timestamped-transcription].  
      Here is a structured list of six steps: [six-steps-summary].  
      Here are six screenshot URLs corresponding to these steps: [six-screenshots].  
      Please create a single, self-contained HTML article that includes all six steps (with headings and short descriptions), each step featuring its corresponding screenshot image, and optionally referencing the broader transcription. The final output must be ready to publish as is.”  
   - The LLM (#223) outputs the final HTML document in text format.  
   - The result is a single HTML file, “final_article.html,” containing all required sections.

F) SILOS  
• SILO #1: Generate the final HTML  
  – INPUT → Skill #223 (prompt with transcription, steps, images) → OUTPUT: [final-article-html]

This completes the “ArticleAssembler” subagent flow, culminating in a single HTML article ready for publication.

4 Template & Links

Expand Flow

A) SUBAGENT SUMMARY  
It appears there is no defined “Subagent #5” in the workflow you shared. Because there is no description or role specified for a fifth subagent, there is currently no actionable workflow to detail. In other words, we cannot break down inputs, outputs, or skill chains for a subagent that does not exist in the overall plan.

B) FINAL TASK OUTPUT  
No output can be generated (HTML, JSON, text, or otherwise) because Subagent #5 is not actually part of the defined system and has no assigned purpose or goals.

C) SUBAGENT INPUT  
No inputs are defined for a subagent that does not exist in the workflow.

E) SUBAGENT TASK SUMMARY  
Since Subagent #5 has no assigned function or description, there is no series of skill calls or steps to outline.

F) SILOS  
None—there are no subtasks or silos for a subagent that has not been specified in the workflow.

────────────────────────
• If you intended to add another subagent to perform a new or extra task, you will first need to define its purpose, inputs, expected output, and how it integrates with the rest of the system. Once that is specified, a workflow breakdown (similar to those created for the first four subagents) can be generated.

5 Template & Links

Expand Flow

A) SUBAGENT SUMMARY  
This subagent is responsible for taking the text transcription, the chosen six steps (with their timestamps), and the six extracted screenshot image URLs, then producing one polished, publication-ready HTML article containing all six steps (each with a heading, descriptive text, and embedded image).

B) FINAL TASK OUTPUT  
A single self-contained HTML file (e.g., “final_article.html”), structured into six steps (headings + text) with the six images (screenshots) embedded in context.

C) SUBAGENT INPUT  
• The raw transcription of the video ([timestamped-transcription])  
• The structured six-step summary data ([six-steps-summary])  
• Six screenshot image URLs ([six-screenshots])

E) SUBAGENT TASK SUMMARY

1) Receive the following three inputs:  
   – [timestamped-transcription] (the entire video’s transcription with timestamps)  
   – [six-steps-summary] (JSON or structured text containing step number, title, timestamp, and short description for each of the six steps)  
   – [six-screenshots] (a list or array of six image URLs corresponding to the timestamps)

2) Call Skill #223 (Powerful LLM Prompt-to-Text Response)  
   – Input to Skill #223:  
     • A carefully crafted prompt instructing the LLM to create a single HTML document. The prompt will include:  
       · The full transcription ([timestamped-transcription]) for reference.  
       · The six steps with their titles, descriptions, and timestamps ([six-steps-summary]).  
       · The six screenshot image URLs ([six-screenshots]) in a known order (Step 1 screenshot, Step 2 screenshot, etc.).  
       · Instructions:  
         “Produce a fully polished HTML article with six clearly marked steps. Each step should have a heading, a short text description, and the associated screenshot image placed neatly within. Ensure the HTML file is self-contained and logically formatted for immediate publication (e.g., <html>, <head>, <body> tags, <h2> or <h3> for step headings, <img> tags for screenshots, etc.).”

– Output from Skill #223:  
     • A single block of HTML markup that reflects these instructions.

3) Subagent Final Output  
   – Return the generated HTML file text (e.g., “final_article.html”).

F) SILOS  
In this subagent, there is effectively a single “silo” of activity—assembling the final HTML from all three inputs. There are no additional silos here because the subagent can accomplish its entire task with one LLM-based skill call.

────────────────────────────────────────────

With the above steps, this subagent can consistently produce a single cohesive HTML article from the provided transcription, selected six steps, and extracted screenshot images.