Add context engineering curriculum tracks

README.md

@@ -1,13 +1,13 @@
 # token-optimization
 A customer-facing engagement for context management and token optimization
 
-This repository contains source material for a practical developer workshop on context management, token optimization, agent customization, tool/MCP hygiene, model choice, AI evals, usage visibility, and sustainable team practices.
+This repository contains source material for a practical developer workshop on token optimization, context engineering, agent customization, tool/MCP hygiene, model choice, AI evals, usage visibility, and sustainable team practices.
 
-Start with [`labs/README.md`](labs/README.md) for 1-hour, 2-hour, and 4-hour delivery outlines, then use the chapter files in order or as modular source material.
+Start with [`labs/README.md`](labs/README.md) for 1-hour, 2-hour, and 4-hour delivery outlines, then use the track labs, deck outline, surface matrix, templates, exercises, and worksheets as modular source material.
 
 ## Sample web app
 
-This repository also includes a static JavaScript sample app for GitHub Copilot Usage-Based Budgeting.
+This repository also includes a static JavaScript sample app for GitHub Copilot Usage-Based Budgeting. It includes budget-scope guidance and a per-surface estimator that uses user-supplied token rates instead of hardcoded future pricing.
 
 To run it locally:
 
@@ -57,16 +57,50 @@ The repo uses small project skills for workflows that should only load when rele
 - Give attendees repeatable habits they can apply in other projects.
 - Introduce eval-driven improvement so teams can measure whether changes help.
 
+## Curriculum tracks
+
+| Track | Labs |
+| --- | --- |
+| VS Code/IDE users | [`00`](labs/00-foundations.md), [`01`](labs/01-ide-context-and-prompt-flow.md), [`02`](labs/02-ide-instructions-tools-and-mcp.md), [`07`](labs/07-measurement-billing-and-governance.md), [`08`](labs/08-applied-repo-review-and-adoption.md) |
+| GitHub Copilot CLI users | [`00`](labs/00-foundations.md), [`03`](labs/03-cli-context-and-tool-output.md), [`04`](labs/04-cli-agents-tools-and-cost-control.md), [`07`](labs/07-measurement-billing-and-governance.md), [`08`](labs/08-applied-repo-review-and-adoption.md) |
+| GitHub.com/code review users | [`00`](labs/00-foundations.md), [`05`](labs/05-github-web-context-and-coding-agent.md), [`06`](labs/06-github-code-review-and-pr-hygiene.md), [`07`](labs/07-measurement-billing-and-governance.md), [`08`](labs/08-applied-repo-review-and-adoption.md) |
+| Full cross-surface practitioner | [`00`](labs/00-foundations.md) through [`08`](labs/08-applied-repo-review-and-adoption.md) |
+
 ## Contents
 
-- [`labs/README.md`](labs/README.md) — overview, prerequisites, and timed agendas
-- [`labs/01-context-management-basics.md`](labs/01-context-management-basics.md)
-- [`labs/02-instructions-and-agent-customizations.md`](labs/02-instructions-and-agent-customizations.md)
-- [`labs/03-mcp-and-tool-optimization.md`](labs/03-mcp-and-tool-optimization.md)
-- [`labs/04-chat-session-management.md`](labs/04-chat-session-management.md)
-- [`labs/05-model-choice.md`](labs/05-model-choice.md)
-- [`labs/06-chat-history-and-memory.md`](labs/06-chat-history-and-memory.md)
-- [`labs/07-usage-and-billing-visibility.md`](labs/07-usage-and-billing-visibility.md)
-- [`labs/08-ai-evals-and-observability.md`](labs/08-ai-evals-and-observability.md)
-- [`labs/09-ideal-workshop-repo.md`](labs/09-ideal-workshop-repo.md)
-- [`labs/10-next-steps-and-extra-topics.md`](labs/10-next-steps-and-extra-topics.md)
+- [`labs/README.md`](labs/README.md) - overview, prerequisites, and timed agendas
+- [`decks/token-optimization-context-engineering.pptx`](decks/token-optimization-context-engineering.pptx) - primary workshop delivery deck with embedded speaker notes
+- [`decks/token-optimization-context-engineering.executive.pptx`](decks/token-optimization-context-engineering.executive.pptx) - executive briefing visual variant
+- [`decks/token-optimization-context-engineering.technical.pptx`](decks/token-optimization-context-engineering.technical.pptx) - technical deep dive visual variant
+- [`decks/token-optimization-context-engineering.outline.md`](decks/token-optimization-context-engineering.outline.md) - editable delivery deck outline
+- [`decks/token-optimization-context-engineering.speaker-notes.md`](decks/token-optimization-context-engineering.speaker-notes.md) - speaker notes for the delivery deck
+- [`tools/generate_context_deck.py`](tools/generate_context_deck.py) - regenerates the styled PPTX variants from the Markdown sources
+- [`resources/copilot-surface-matrix.md`](resources/copilot-surface-matrix.md) - living reference for Copilot surfaces and context controls
+- [`resources/context-inventory-worksheet.md`](resources/context-inventory-worksheet.md)
+- [`resources/instruction-diet-worksheet.md`](resources/instruction-diet-worksheet.md)
+- [`resources/customer-preflight-checklist.md`](resources/customer-preflight-checklist.md)
+- [`resources/monday-morning-checklist.md`](resources/monday-morning-checklist.md)
+- [`templates/README.md`](templates/README.md) - copy/paste starter Copilot customization files
+- [`exercises/README.md`](exercises/README.md) - track-specific hands-on exercises
+- [`facilitator/delivery-guide.md`](facilitator/delivery-guide.md)
+- [`labs/00-foundations.md`](labs/00-foundations.md)
+- [`labs/01-ide-context-and-prompt-flow.md`](labs/01-ide-context-and-prompt-flow.md)
+- [`labs/02-ide-instructions-tools-and-mcp.md`](labs/02-ide-instructions-tools-and-mcp.md)
+- [`labs/03-cli-context-and-tool-output.md`](labs/03-cli-context-and-tool-output.md)
+- [`labs/04-cli-agents-tools-and-cost-control.md`](labs/04-cli-agents-tools-and-cost-control.md)
+- [`labs/05-github-web-context-and-coding-agent.md`](labs/05-github-web-context-and-coding-agent.md)
+- [`labs/06-github-code-review-and-pr-hygiene.md`](labs/06-github-code-review-and-pr-hygiene.md)
+- [`labs/07-measurement-billing-and-governance.md`](labs/07-measurement-billing-and-governance.md)
+- [`labs/08-applied-repo-review-and-adoption.md`](labs/08-applied-repo-review-and-adoption.md)
+- [`labs/MIGRATION.md`](labs/MIGRATION.md)
+
+## Regenerating the deck
+
+The PPTX files are generated from the Markdown outline and speaker notes so content remains easy to review.
+
+```powershell
+python -m pip install -r requirements-dev.txt
+python tools\generate_context_deck.py
+```
+
+The generator produces workshop, executive briefing, and technical deep dive variants. All generated PPTX files include embedded PowerPoint speaker notes from `decks/token-optimization-context-engineering.speaker-notes.md`.

decks/token-optimization-context-engineering.outline.md

@@ -0,0 +1,172 @@
+# Token Optimization and Context Engineering across GitHub Copilot
+
+Source outline for the delivery deck. Keep this file easy to diff; regenerate the `.pptx` delivery artifacts after major edits with `python tools\generate_context_deck.py`.
+
+## Slide 1: Title
+
+- Token Optimization and Context Engineering
+- Getting more value from every Copilot interaction
+- Across Copilot CLI, VS Code, GitHub.com, coding agent, and code review
+
+## Slide 2: Why this matters now
+
+- Usage models are shifting from simple request counting toward more granular accounting.
+- Long conversations, broad tool output, and unnecessary context can affect cost, latency, and quality.
+- The durable habit is not "use less Copilot"; it is "send better context."
+
+## Slide 3: Context engineering
+
+- Context engineering means bringing the right information, in the right format, to the model.
+- It turns ad hoc prompting into repeatable workflows.
+- Core primitives: custom instructions, prompt files, skills, agents, retrieval, and human review gates.
+
+## Slide 4: What counts as context
+
+- Product and system instructions
+- Repository, organization, path-specific, personal, and agent guidance
+- Conversation history and summaries
+- Files, selections, issues, pull requests, and tool results
+- Retrieved docs, web pages, MCP output, and generated plans
+
+## Slide 5: Important billing nuance
+
+- "Context is resent" is a useful mental model, not a universal invoice formula.
+- Some products and models use caching or product-specific accounting.
+- The practical takeaway remains: stale or irrelevant context increases cost, latency, and confusion.
+
+## Slide 6: Where context gets wasted
+
+- Whole directories or large files when one function matters
+- Long-running mixed-topic sessions
+- Overgrown custom instructions
+- Raw logs, build output, generated files, and tool noise
+- High-cost models used for routine work
+- Auto-review or agent tasks with vague scope
+
+## Slide 7: The five levers
+
+- Context hygiene
+- Prompt discipline
+- Model and surface selection
+- Scope and tool control
+- Measurement
+
+## Slide 8: Lever 1 - Context hygiene
+
+- Start fresh when the task changes.
+- Summarize before switching focus.
+- Keep stable source-of-truth docs short and current.
+- Avoid re-discovery loops by preserving useful handoffs.
+
+## Slide 9: Lever 2 - Prompt discipline
+
+- Use Markdown structure.
+- State outcome, scope, constraints, and success criteria.
+- Reference specific files, issues, PRs, or selections.
+- Add validation gates before risky edits.
+
+## Slide 10: Lever 3 - Model and surface selection
+
+- Choose the cheapest model or surface that can reliably finish the job.
+- Use VS Code Plan for design, Agent for implementation, Ask for exploration.
+- Use CLI when you need visible context/tool control.
+- Use code review when the unit of work is a pull request.
+
+## Slide 11: Lever 4 - Scope and tool control
+
+- Keep workspace, repository, and tool scope tight.
+- Use targeted instructions instead of one giant instruction file.
+- Enable MCP tools only when the task needs them.
+- Use human-in-the-loop approval for high-risk tool actions.
+
+## Slide 12: Lever 5 - Measurement
+
+- Measure tokens where exposed.
+- Measure premium requests, review counts, and billing views where tokens are hidden.
+- Measure quality with retries, false positives, PR churn, and human rework.
+- Optimize after you have a baseline.
+
+## Slide 13: Surface matrix
+
+- CLI: most visible token/context controls.
+- VS Code: daily coding workflow with Ask, Plan, Agent, custom instructions, prompt files, and review.
+- GitHub.com: repo, issue, PR, and discussion context.
+- Coding/cloud agent: asynchronous implementation from scoped tasks.
+- Code review: PR-focused feedback with product-specific constraints.
+
+## Slide 14: VS Code pattern
+
+- Curate project context with concise instructions and docs.
+- Plan first for complex work.
+- Implement from the plan in a fresh or focused session.
+- Review changes against the plan.
+
+## Slide 15: GitHub.com web pattern
+
+- Ask from the page that already has the relevant context.
+- Keep threads focused.
+- Use repository, issue, and PR context deliberately.
+- Treat generated files as drafts.
+
+## Slide 16: Copilot CLI pattern
+
+- Use sessions like branches: one task, one focused context.
+- Filter tool output before it enters the conversation.
+- Delegate noisy discovery when available.
+- Use usage/context visibility to teach the token mental model.
+
+## Slide 17: Coding agent pattern
+
+- Write issues like implementation briefs.
+- Include acceptance criteria, validation commands, and files to avoid.
+- Keep tasks small enough to review.
+- Review the generated PR like any other teammate's work.
+
+## Slide 18: Code review pattern
+
+- Keep PRs small.
+- Tune repo and path-specific review instructions.
+- Watch automatic review policy and quota implications.
+- Validate Copilot findings; review comments are not approvals.
+
+## Slide 19: Context inventory exercise
+
+- List every context source.
+- Mark required, useful, stale, redundant, sensitive, or unknown.
+- Decide what stays, what moves, what gets summarized, and what gets removed.
+
+## Slide 20: Instruction diet exercise
+
+- Keep stable rules always-on.
+- Move targeted rules to path-specific instructions.
+- Move repeated workflows to prompts, skills, or agents.
+- Link long docs instead of copying them into instructions.
+
+## Slide 21: Governance
+
+- Content exclusion
+- Model access policies
+- Code review settings
+- Budgets and alerts
+- Telemetry and dashboard ownership
+
+## Slide 22: Delivery tracks
+
+- 1 hour: mental model, surface view, demo, checklist
+- 2 hours: practitioner lab and prompt/context refactor
+- 4 hours: customer environment review and team operating model
+
+## Slide 23: What improvement looks like
+
+- Fewer irrelevant tokens
+- Faster answers
+- Fewer retries
+- Higher-signal reviews
+- Clearer ownership of policy and measurement
+
+## Slide 24: Takeaways
+
+- Context is a design input, not a dumping ground.
+- Token optimization and context engineering improve both cost and quality.
+- Surface controls differ; the habits transfer.
+- Start with three changes this week and measure.

decks/token-optimization-context-engineering.speaker-notes.md

@@ -0,0 +1,99 @@
+# Speaker Notes: Token Optimization and Context Engineering
+
+These notes are embedded into the generated PPTX variants by `python tools\generate_context_deck.py`.
+
+## Slide 1
+
+Open by saying this is not a "use less Copilot" session. It is a "get better answers with less waste" session.
+
+## Slide 2
+
+Anchor on the current shift: customers need practical habits before usage surprises become trust issues.
+
+## Slide 3
+
+Define context engineering as an evolution of prompt engineering. The goal is not clever phrasing; the goal is the right information in the right format.
+
+## Slide 4
+
+Walk left to right through the context inputs. Point out that the user prompt is often the smallest part.
+
+## Slide 5
+
+Use careful language. Product accounting can vary and caching may apply. The workshop teaches habits that remain useful even as billing details change.
+
+## Slide 6
+
+Emphasize that context waste hurts quality before it hurts the bill. Too much irrelevant context makes the model less focused.
+
+## Slide 7
+
+This is the map for the workshop. Do not create a second framework; map context engineering into these five levers.
+
+## Slide 8
+
+Use the branch analogy: one focused session per task. Clearing context is good between tasks, but clearing and re-reading the same files can be wasteful.
+
+## Slide 9
+
+Show one overloaded request and one structured request. Keep the demo short.
+
+## Slide 10
+
+Make model and surface selection practical. The right answer may be a different surface, not just a different model.
+
+## Slide 11
+
+Explain that every enabled tool is a capability and a possible source of noise. Tool descriptions and access boundaries are product UX.
+
+## Slide 12
+
+For surfaces without token visibility, measure proxies: retries, PR churn, review false positives, usage dashboards, and time-to-merge.
+
+## Slide 13
+
+Warn that the matrix is not parity. Some cells are intentionally "not available."
+
+## Slide 14
+
+VS Code is the daily practitioner path: Ask to understand, Plan to structure, Agent to implement, Review to validate.
+
+## Slide 15
+
+GitHub.com is strongest when the page context matters: repository, issue, pull request, or discussion.
+
+## Slide 16
+
+CLI remains the best live demo for token concepts because context and usage are visible.
+
+## Slide 17
+
+Coding agent tasks should look like good implementation tickets. Vague tasks create broad exploration and broad diffs.
+
+## Slide 18
+
+Code review is purpose-built. Do not promise model switching. Teach PR hygiene and instruction hygiene.
+
+## Slide 19
+
+Students should discover that some of their "helpful" context is stale, sensitive, or redundant.
+
+## Slide 20
+
+This is the most actionable demo. Trim one bloated instruction file and show where the pieces should move.
+
+## Slide 21
+
+Separate developer habits from admin controls. Admin controls compound because they apply every turn for every user.
+
+## Slide 22
+
+Match the track to the audience. Do not run the 4-hour customer review without preflight.
+
+## Slide 23
+
+Avoid unsupported savings promises. Talk about directional improvements and measurement.
+
+## Slide 24
+
+Close with a small commitment: pick three habits, measure for a week, then expand.

exercises/01-vscode-context-attachments/README.md

@@ -0,0 +1,26 @@
+# Exercise 01: VS Code Context Attachments
+
+## Goal
+
+Practice replacing broad IDE context with deliberate file, selection, and test context.
+
+## Task
+
+Investigate a login bug in a sample project.
+
+## Baseline
+
+Read [`naive-transcript.md`](naive-transcript.md) and identify context that is stale, redundant, or too broad.
+
+## Engineered flow
+
+Read [`engineered-transcript.md`](engineered-transcript.md) and compare:
+
+- Which files are attached?
+- Which context is omitted?
+- Which mode is used first?
+- What validation gate is added?
+
+## Output
+
+Write one prompt for Ask mode, one for Plan mode, and one for Agent mode.

exercises/01-vscode-context-attachments/engineered-transcript.md

@@ -0,0 +1,12 @@
+# Engineered transcript
+
+User:
+
+> In `src/auth/session.ts`, explain why expired sessions are not rejected. Use the selected failing test and the stack trace below. Do not inspect unrelated directories. Return a diagnosis and patch plan before editing.
+
+Likely result:
+
+- Specific file and test context.
+- Clear scope boundary.
+- Plan before implementation.
+- Validation is known before edits begin.