Skills are new features: Building Skill-Centric Harness — Yogendra Miraje, FactSet

What was stored

cues

136

whisperx 136

chunks

30

from 136 cues

keyframes

40

kept of 52 captured

frames with text

40

702 lines read

chapters

12

from the source metadata

keyframe bytes

5.4 MB

word timings on 136 cues

Provenance

Frames, and what the machine read

• 

  0:02 #0 done2 line(s)

  shot 0·sharpness 455.2

  1. AlEngineer0.95
  2. World's Fair0.97
• 

  0:03 #1 done2 line(s)

  shot 1·sharpness 670.0

  1. AlEngineer0.96
  2. World's Fair0.99
• 

  0:11 #2 done24 line(s)

  shot 2·sharpness 2742.1

  1. LAB & PLATINUM SPONSORS0.97
  2. Amazon AGI Lab0.99
  3. ANTHROP\C1.00
  4. Google DeepMind1.00
  5. MINIMAX0.98
  6. OpenAl0.92
  7. Akamai1.00
  8. arize1.00
  9. aws1.00
  10. Braintrust bright data0.99
  11. B1.00
  12. Browserbase1.00
  13. docker1.00
  14. :neo4j0.91
  15. ORACLE1.00
  16. PayPal1.00
  17. qodo0.99
  18. reducto1.00
  19. Sonar1.00
  20. Makers of1.00
  21. together.ai0.98
  22. Unblocked1.00
  23. WorkOS1.00
  24. SonarQube1.00
• 

  0:13 #3 done2 line(s)

  shot 3·sharpness 330.6

  1. AlEngineer1.00
  2. World's Fair0.98
• 

  0:35 #4 done13 line(s)

  shot 4·sharpness 1884.8

  1. AlEngineer0.98
  2. World'sFair1.00
  3. Skills are New Features0.99
  4. PRESENTED BY1.00
  5. Microsoft1.00
  6. Building Skill-Centric Agentic Products0.99
  7. Yogi Miraje1.00
  8. AIE WF 20261.00
  9. X0.90
  10. @YogiNotTheBear1.00
  11. Engineering the future of Al0.99
  12. AlEngineer1.00
  13. World's Fair0.96
• 

  0:54 #5 done16 line(s)

  shot 5·sharpness 2475.2

  1. AlEngineer0.97
  2. World'sFair1.00
  3. How to Build Agents1.00
  4. without losing the Control0.99
  5. Yogi Miraje0.98
  6. PRESENTED BY1.00
  7. Microsoft1.00
  8. Retrieval: Summarize NVIDIA's previous earnings call.1.00
  9. Blueprint1.00
  10. Retrieval: Gather NVIDIA's latest EPS and revenue.0.99
  11. Tool: Task0.99
  12. Reasoning: Suggest questions for the next earnings call based on all the information fetched0.99
  13. Reporting: Generate a comprehensive report from fetched information0.99
  14. Engineering the future of Al1.00
  15. AlEnginee0.99
  16. World'sFair1.00
• 

  1:20 #6 done22 line(s)

  shot 6·sharpness 2243.3

  1. AlEngineer0.98
  2. Equipping agents for the real world0.99
  3. World's Fair0.99
  4. with Agent Skills0.94
  5. PRESENTED BY1.00
  6. Published Oct 16, 20250.98
  7. Claude is powerful, but real work requires procedural knowledge and organizational0.99
  8. context. Introducing Agent Skill, a new way to build specialized agents using files0.99
  9. Microsoft1.00
  10. and folders.1.00
  11. Code Summit0.98
  12. ANTHROP\C1.00
  13. Claude1.00
  14. Agent1.00
  15. an0.57
  16. DON'T BUILD AGENTS!0.98
  17. BuildSkillsInstead1.00
  18. 16:221.00
  19. TRACK 3· JULY 2,20260.96
  20. AlEnginee0.96
  21. Al in Finance0.99
  22. World's Fair0.97
• 

  2:19 #7 done15 line(s)

  shot 7·sharpness 1276.0

  1. AlEngineer0.97
  2. World'sFair0.96
  3. X0.85
  4. What it's NOT about0.99
  5. What it IS about0.99
  6. ×Coding harnesses0.98
  7. √ Skills in the context of Agentic products0.99
  8. Skills for coding agents0.99
  9. √ How to add support for skills in your own harness0.99
  10. × How to write Skills0.96
  11. √ How to scale skills across the enterprise0.99
  12. ⑤/Q1O0.52
  13. TRACK 3· JULY 2, 20260.93
  14. Alin Finance0.99
  15. World'sFair0.98
• 

  2:27 #8 skipped

  shot 8·duplicate of #7

• 

  3:01 #9 done55 line(s)

  shot 9·sharpness 1470.7

  1. AlEngineer0.99
  2. Traditional Product0.99
  3. World's Fair0.97
  4. Al Platform0.98
  5. Overview1.00
  6. May 1t- May t. 20240.72
  7. riers0.80
  8. Ovenview0.88
  9. Total Predictions0.95
  10. Total insights0.90
  11. Insights1.00
  12. Analytice0.90
  13. 2.45M1.00
  14. vs May S - May 10.84
  15. 92.4%+27%0.93
  16. vs May S - May 10.93
  17. ve May 5 - May t0.82
  18. 1281.00
  19. 71.00
  20. os May 5 – May 110.84
  21. + 12.5%0.72
  22. ② tnaights0.59
  23. Predictions Over Time0.95
  24. Duly0.67
  25. Predictions by Category0.99
  26. Reporns0.86
  27. Datinets0.98
  28. 88 Models0.70
  29. 2.45M1.00
  30. Active Workflows0.97
  31. Worktows0.76
  32. C0.79
  33. Settings0.97
  34. Recent Predictions1.00
  35. PRO-23810.67
  36. Input Date0.78
  37. 0.880.73
  38. May 18.204 10:224 AM0.78
  39. Mey 18. 224 1020 AM0.74
  40. Predicted At0.83
  41. Run Prediction0.97
  42. Select Model0.90
  43. Churn Model V210.77
  44. P90-7438100.79
  45. May 18.2004 10 15 AM0.75
  46. ar tte0.54
  47. 0.70.87
  48. May 181,202 12:10 AM0.71
  49. mer Phofle - May 20240.86
  50. 0.760.98
  51. May 18. 20410 AM0.81
  52. TRACK 3• JULY 2, 20260.96
  53. Al in Finance0.97
  54. AlEngineer1.00
  55. World's Fair0.99
• 

  3:13 #10 done55 line(s)

  shot 10·sharpness 1451.6

  1. AlEngineer0.98
  2. Agentic Product1.00
  3. World's Fair0.99
  4. Q2 Market Entry Plan1.00
  5. $° Diaf updated ty Agein0.71
  6. Agent Workflow0.98
  7. Active0.97
  8. ② Saved just now0.91
  9. 1. Goal1.00
  10. Q2 Market Entry Plan0.99
  11. Create a concise market entry brief for Q20.98
  12. with recommended approach1.00
  13. Draft Brief - Last updlated by Agent just now0.92
  14. 2. Plan0.96
  15. Research market, define opportunity.0.99
  16. approach, and key considerations.0.98
  17. 1. Executive Summary1.00
  18. We recommend entering the German and Australian markets in Q2 with a phased approach,0.95
  19. focusing on segments where our solution offers clear differentiation and strong demand.0.98
  20. 3. Progress1.00
  21. 4/4 stps0.97
  22. 2. Opportunity1.00
  23. Reading notes & research0.96
  24. +:2.AM0.52
  25. High demand for ou olution in Germany n Australia driven by digitl transformmation0.81
  26. Drating brief0.89
  27. 00.52
  28. 9:07 AM0.82
  29. initiatves and a need for scalable, easy-to-integrate tools.0.99
  30. Revising outine0.96
  31. 9:51AM0.66
  32. 3. Recommended Approach1.00
  33. Preparing final draft0.93
  34. • Phase 1 (0-30 days): Validate ICP and buld local partnerships.0.97
  35. • Phese 2 30 90 days: Luch lot rorams and refie meossaging.0.73
  36. • Phase 3 (90-180 dys): Scae go-t-market ad epand chhanes.0.76
  37. 4. Evidence1.00
  38. completed this morning.0.95
  39. Based on 12 notes and market research0.99
  40. 4. Key Considerations1.00
  41. • Ensure localization for language and compliance.0.99
  42. • Align pricing with local market expectations.0.97
  43. of0.75
  44. 5. Decision1.00
  45. Pending revle0.92
  46. Waiting for your review and decision.0.99
  47. Review and approve this draft1.00
  48. ed the draft based on your goal and plar0.97
  49. Approve1.00
  50. Request Changes0.99
  51. × Deny0.93
  52. TRACK 3· JULY 2, 20260.93
  53. Al in Finance0.99
  54. AlEngineer1.00
  55. World's Fair0.98
• 

  3:32 #11 done7 line(s)

  shot 11·sharpness 995.5

  1. AlEngineer0.98
  2. World's Fair0.97
  3. If the agent is the interface,1.00
  4. where do the features live?0.98
  5. TRACK 3 · JULY 2, 20260.94
  6. Al in Finance0.97
  7. World'sFair1.00
• 

  3:48 #12 done22 line(s)

  shot 12·sharpness 1311.8

  1. AlEngineer0.98
  2. Who, What, How of an Agent0.99
  3. World's Fair0.99
  4. Who1.00
  5. What1.00
  6. How1.00
  7. System Prompt1.00
  8. Tools1.00
  9. Skills1.00
  10. Who the Agent is1.00
  11. What it can connect to0.99
  12. How a task gets done0.98
  13. Defines the agent's identity, purpose, role,1.00
  14. External systems, data, and APIs the agent0.99
  15. Packaged know-how the agent follows to1.00
  16. and constraints.1.00
  17. uses to get things done.0.99
  18. do a specific task well.1.00
  19. TRACK 3· JULY 2, 20260.96
  20. Al in Finance0.99
  21. AlEngineer1.00
  22. World's Fair0.96
• 

  4:08 #13 done25 line(s)

  shot 13·sharpness 1466.5

  1. AlEngineer0.98
  2. Skills are New Features1.00
  3. World's Fair0.98
  4. Equity Research1.00
  5. Wealth Management1.00
  6. catalyst-calendar1.00
  7. model-update1.00
  8. client-report0.99
  9. tax-loss-harvesting1.00
  10. earnings-analysis1.00
  11. morning-note1.00
  12. client-review1.00
  13. earnings-preview1.00
  14. sector-overview1.00
  15. financial-plan1.00
  16. idea-generation1.00
  17. thesis-tracker1.00
  18. investment-proposal1.00
  19. initiating-coverage1.00
  20. portfolio-rebalance1.00
  21. Source: Anthropic Financial Services Plugins1.00
  22. TRACK 3· JULY 2, 20260.96
  23. Al in Finance1.00
  24. AlEngineer1.00
  25. World's Fair0.96
• 

  4:44 #14 done9 line(s)

  shot 14·sharpness 1001.7

  1. AlEngineer0.98
  2. World'sFair1.00
  3. Role of Product Engineers0.99
  4. Shipping Features1.00
  5. Shipping Harness1.00
  6. ⑤/Q10.59
  7. TRACK 3• JULY 2, 20260.97
  8. Al in Finance0.96
  9. World'sFair1.00
• 

  5:36 #15 done26 line(s)

  shot 15·sharpness 1895.8

  1. AlEngineer0.99
  2. Definition of a Skill1.00
  3. skill1.00
  4. World's Fair0.99
  5. /skdl0.68
  6. noun1.00
  7. the abilit to do something well.0.98
  8. A standardized way to teach AI agents how to do specific tasks0.99
  9. PRESENTED BY1.00
  10. well.1.00
  11. Microsoft1.00
  12. A skill is a directory containing, at minimum, a skILL.md file:0.96
  13. skill-name/1.00
  14. SKILL.md0.97
  15. # Required: metadata + instructions0.99
  16. scripts/0.96
  17. # Optional: executable code0.99
  18. references/1.00
  19. # Optional: documentation0.98
  20. assets/1.00
  21. # Optional: templates, resources0.98
  22. # Any additional files or directories1.00
  23. TRACK 3· JULY 2,20260.96
  24. Al in Finance1.00
  25. AlEnginee1.00
  26. World's Fair0.96
• 

  5:55 #16 done54 line(s)

  shot 16·sharpness 1441.1

  1. AlEngineer0.99
  2. Anatomy of a Skill1.00
  3. World's Fair0.99
  4. pdf/SKILL.md1.00
  5. pdf/reference.md1.00
  6. YAML Frontmatter0.97
  7. # PDF Processing Advanced Reference0.98
  8. name: pdf0.98
  9. description: Comprehensive PDF toolkit for extracting text and0.98
  10. tabtes, merging/splitting documents, and filling-out forms.0.99
  11. This document contains advanced PDF processing features,0.99
  12. detailed examples, and additional libraries not covered0.98
  13. in the main skill instructions.0.98
  14. # pypdfium2 Library (Apache/BSD License)0.99
  15. PRESENTED BY1.00
  16. Microsoft1.00
  17. Markdown1.00
  18. This guide covers essential PDF processing operations using Python0.99
  19. a8 Overview0.85
  20. pypdfium2 is a Python binding for POFium (Chromium's PDF0.94
  21. Library). It's excellent for fast PDF rendering, image0.97
  22. generation, and serves as ...0.96
  23. ### Overview0.88
  24. JavaScript libraries, and detailed examples, see /reference.nd.0.98
  25. libraries and command-line tools. For advanced features,0.99
  26. pdf/forms.md1.00
  27. If you need to fill out a PDF fors, read /forms.nd and follow its0.97
  28. instructions.1.00
  29. Quick Start0.97
  30. If you need to fill out a PoF forn, first check to see0.98
  31. if the POF has fillable fors fields. Run this script0.92
  32. from this file's directory:0.99
  33. python scripts/check_fillable_fields <file.pdf>,0.97
  34. from pypdf import PdfReader, Pafriter0.95
  35. # Read a POF0.99
  36. and depending on the result go to either the "Fillable1.00
  37. fields" or "Non-fillable fields" and follow those0.99
  38. instructions.1.00
  39. print(fPages: {len(reader.pages)}")0.81
  40. for page in reader.pages:0.99
  41. reader = PdfReader("document.pdf")0.97
  42. # Extract text0.97
  43. text =0.94
  44. text += page.extract_text()1.00
  45. # Fillable fields0.95
  46. If the POF has fillable fors fields:0.98
  47. <fields.json>.0.98
  48. - Run this script from this file's directory:0.99
  49. python scripts/extract_form_field_info.py <input.pdf>1.00
  50. Source: Anthropic Blog1.00
  51. TRACK 3· JULY 2, 20260.96
  52. Al in Finance0.99
  53. AlEngineer1.00
  54. World'sFair1.00
• 

  6:03 #17 done10 line(s)

  shot 17·sharpness 1133.8

  1. AlEngineer0.97
  2. World'sFair1.00
  3. PRESENTED BY1.00
  4. Microsoft1.00
  5. Adding Skill Support in Harness1.00
  6. ⑤/Q0.57
  7. TRACK 3• JULY 2, 20260.96
  8. Alin Finance0.99
  9. AlEngineer0.99
  10. World'sFair0.96
• 

  6:21 #18 done20 line(s)

  shot 18·sharpness 1396.0

  1. AlEngineer0.98
  2. Basic components1.00
  3. World's Fair0.99
  4. PRESENTED BY0.97
  5. REQUIRED1.00
  6. REQUIRED1.00
  7. REQUIRED1.00
  8. Microsoft1.00
  9. Skill Registry1.00
  10. System Prompt1.00
  11. File Read Tool1.00
  12. Where skills are loaded from0.99
  13. Controllable — where a skill gets0.99
  14. Loads skill bodies & resources on0.99
  15. injected1.00
  16. demand1.00
  17. TRACK 3· JULY 2, 20260.95
  18. Al in Finance0.99
  19. AlEnginee0.99
  20. World's Fair0.97
• 

  6:38 #19 done25 line(s)

  shot 19·sharpness 1538.2

  1. AlEngineer0.98
  2. A simple Skill Registry1.00
  3. World's Fair0.96
  4. python1.00
  5. registry = [0.96
  6. frontmatter(open(f, encoding="utf-8").read()) I {"path": f}0.98
  7. for f in sorted(glob("skills/*/SKILL.md"))1.00
  8. Skill0.99
  9. Description1.00
  10. Path to SKILL.md1.00
  11. Creates a beginner-friendly research brief about a public1.00
  12. company-research1.00
  13. company.1.00
  14. skills/company-research/SKILL.md1.00
  15. Turns Markdown into a polished, standalone HTML report while1.00
  16. report-html1.00
  17. preserving all content and structure.0.98
  18. skills/build-report/SKILL.md1.00
  19. report-pdf1.00
  20. Fills, merges, and extracts data from PDF files.0.99
  21. skills/pdf/SKILL.md1.00
  22. TRACK 3· JULY 2,20260.94
  23. AlEngineer0.99
  24. Al in Finance0.99
  25. World's Fair0.98
• 

  6:55 #20 done18 line(s)

  shot 20·sharpness 1778.2

  1. AlEngineer0.97
  2. World'sFair1.00
  3. name: company-research1.00
  4. description: Creates a beginner-friendly research brief0.99
  5. about a public company.0.99
  6. tools: web_search0.96
  7. 1. Make exactly three parallel web_search calls:0.99
  8. one query combining the business model and latest1.00
  9. financial results, one for a recent positive0.99
  10. development, and one for an important risk. Do not1.00
  11. request more searches afterward.1.00
  12. 2. Combine the results into a plain-language1.00
  13. brief. Separate facts from interpretation, cite0.98
  14. source links, and do not give buy or sell advice.1.00
  15. TRACK 3· JULY 2,20260.95
  16. Alin Finance0.98
  17. AlEngineer0.99
  18. World'sFair1.00
• 

  7:07 #21 skipped

  shot 21·duplicate of #11

• 

  7:39 #22 skipped

  shot 22·duplicate of #20

• 

  8:13 #23 skipped

  shot 23·duplicate of #20

Transcript

136 cues· 2,124 words· 11,614 chars

Chapters

1. 0:00 From blueprints to skills
2. 1:46 Building skill centric agents
3. 2:52 Skills in context at enterprise scale
4. 3:46 Skills as the new features
5. 5:04 Inside a skill: the skill.md
6. 6:37 A minimal skill registry
7. 7:42 Progressive disclosure
8. 9:13 Trigger words and routing
9. 12:17 Skills without evals drift
10. 12:56 Search and embeddings past ten skills
11. 13:46 Governance past a hundred skills
12. 16:36 Skills as the product interface