studyhall 1.1 - Agent Workflow Verbs / Skills & Plugins / Grant Matching

Link to Folder with Everything

2026.09.02.studyhall.1.1

Recording of the Video

Watch on Google Drive

Summary (by Gemini)

The study hall session explored AI agent methodology and demonstrated practical workflow integration for automated data management.

AI Agent Workflow Foundations

The session introduced principles for managing knowledge and intelligence. Standardizing workflows into 6 systematic verbs improves business outcomes by 10 to 20 percent.

Tooling and Skill Integration

Practical demonstrations utilized browser automation to gather professional dossiers. Technical solutions using industrial synchronization tools and plugin manifests enabled complex data processing and cross-device consistency.

Scaling AI Native Operations

Breaking complex processes into granular steps facilitates effective delegation and scaling. Automated matching identified grant opportunities by leveraging public digital presence and repository integrations.

Next steps

• [The group] Share AI failure stories: Submit personal narratives regarding AI implementation failures and lessons learned to the curriculum developers.
• [The group] Configure Rclone: Configure Rclone to enable data synchronization between local machines and agents using S3 storage.
• [The group] Review skill sets: Examine the newly published collection of AI skills to understand how to incorporate them into existing agent workflows.
• [Rahul Singh] Share Links: Share the public web links for the generated grant research with all meeting participants.
• [Rahul Singh] Share Skill Docs: Distribute documentation detailing the specific functions and capabilities of the created skill.
• [Elizabeth Strickler] Research Grants: Conduct a deeper investigation into the identified international teaching grant opportunities.

Details

• Study Hall 1.1 Overview: Rahul Singh opened the 11th study hall session by focusing on the use of standard, off-the-shelf desktop software like Claude and ChatGPT to optimize information gathering and lead generation, contrasting advanced power-user tools (Git, custom code) with accessible AI agents that let non-technical individuals perform various tasks effectively.
• Intelcraft Training Methodology: Rahul Singh introduced Intelcraft as a standard approach to coordinating knowledge, communication, and intelligence between teams of people and agents. The curriculum teaches lean management principles focused on agentic workflow design and operations to improve business outcomes by 10% to 20%.
• Understanding AI Capabilities and Agents: Agents are described as loops of instructions—utilizing system prompts and data—that include perception, cognition, memory, and action capabilities. While these agents currently control data and internet access, future applications will involve physical robotics.
• Alignment and Trust in AI: Rahul Singh highlighted the alignment problem — AI systems require team consensus, trust, and security. Because these are personal AIs, users entering team environments must trust their collaborators, as ownership of the AI and its output (such as qualified leads) carries responsibility.
• Systematic Workflow Verbs: The process for leveraging AI is categorized into six systematic verbs: explore, seek, gather, organize, understand, and generate. Broader criteria during initial exploration identifies trends better than starting with overly specific parameters that might miss valuable opportunities.
• Live Project Demonstration: Rahul Singh set up a project in Claude titled "Observe, gather, and report" to find AI instructors who had presented at events within the last three to six months, with Claude functioning as the interface to browse the internet and output requested in Excel and JSON formats.
• Data Gathering via Browsing: Rahul Singh demonstrated using Claude to access a logged-in browser session to gather information from LinkedIn, creating a roster and dossier of individuals — an Excel document with separate sheets for contact information, social media, and publications.
• File Management and Synchronization: Responding to Elizabeth Strickler's question about managing files across multiple machines (PCs and Macs, Obsidian, Google Drive), Rahul Singh explained the solution involves industrial-grade synchronization using S3 buckets and rclone, part of the Intelcraft toolkit, to ensure data consistency across devices.
• Data Organization and Knowledge Graphs: The Obsidian methodology uses Markdown files as a neutral format to store gathered data; linking these documents creates a knowledge graph that helps the AI understand relationships and context within the collected information.
• Codifying Workflows with Skills: Rahul Singh introduced "Skills" as plugins — collections of capabilities that codify and share workflows. A set of four skills was shared: financial domain due diligence, an opportunity finder, a sales proposal (RFP) responder, and a skill authoring tool.
• Plugin Installation and Management: Skills are effectively zip files that can be uploaded to the Claude and ChatGPT desktop applications to expand the agent's capabilities. Vicki Harrison noted the abundance of existing skills can be overwhelming, leading them to prefer creating their own purpose-built skills.
• AI-Powered Data Extraction and Formatting: JSON files serve AI processing while Excel remains the "system of truth" for consolidation and summaries. An instructor sheet was generated identifying 95 potential people — 42 conference/keynote speakers, 22 trainers, and 31 webinar hosts across East Asia, India, and Sub-Saharan Africa.
• LinkedIn Dossier and Grant Matching: Rahul Singh created a LinkedIn dossier of participant profiles and major projects to determine which grants best suit them, instructing Claude to identify two to three people per grant and visualize the results across JSON and Excel files.
• Integration of GitHub Skills Repositories: Answering Clyde Clarke, Rahul Singh demonstrated that a public Git repository containing a marketplace.json manifest can be added directly to the Claude plugin marketplace to synchronize skills — no Enterprise subscription required, as skill sharing relies on standard Git permissions.
• Establishing AI-Native Operations and Team Collaboration: Individual AI workflows can scale into "AI native" company operations that run 24/7. The core methodology involves breaking processes into smaller steps — collecting, enhancing, and organizing data — rather than attempting complex workflows in one step; defining individual human roles and a "supervisor process" creates a "harness" that automates workflows.
• The Importance of Digital Presence: Individuals should maintain active profiles on LinkedIn, GitHub, blogs, and personal websites, because AI tools constantly index public information to evaluate candidates and identify potential matches for opportunities.
• Review of AI-Generated Grant Matches: The AI matched participants with specific grant opportunities based on their dossier profiles — examples included GitHub Secure Open Source Fund grants for Eric and Emergent Ventures plus OpenAI cybersecurity programs. Clyde Clarke expressed interest in the shared web links for these opportunities.
• Meeting Conclusion and Reflection: Elizabeth Strickler shared that they built their own spreadsheet and developed a new skill during the session. Rahul Singh concluded by reiterating the methodology of breaking AI workflows into manageable steps — explore, seek, gather, organize, understand, and generate — and encouraged participants to start with one skill and cycle.