AGENT SKILL

app-workshop-session-pdf-recap-next

Extract content from a workshop session PDF and produce a structured recap focused on (1) what was learned, (2) next steps, and (3) possible enhancements. Ingests a single PDF file path, parses session material (slides, transcripts, exercises, Q&A), and emits a Markdown report with Learning Recap, Next Steps, and Enhancement Opportunities sections with page-level traceability. Use when the user asks to "summarize a workshop", "recap a training PDF", "pull next steps from a session deck", or provides a workshop/seminar/tutorial PDF and wants a learn-and-act summary.

使用场景文档处理
License 许可
MIT
最近更新
2026-08-25
Stars
★ 0
Forks
0

01 前言 Front Matter

name
app-workshop-session-pdf-recap-next
tier
app
description
Extract content from a workshop session PDF and produce a structured recap focused on (1) what was learned, (2) next steps, and (3) possible enhancements. Ingests a single PDF file path, parses session material (slides, transcripts, exercises, Q&A), and emits a Markdown report with Learning Recap, Next Steps, and Enhancement Opportunities sections with page-level traceability. Use when the user asks to "summarize a workshop", "recap a training PDF", "pull next steps from a session deck", or provides a workshop/seminar/tutorial PDF and wants a learn-and-act summary.
allowed-tools
Read, Glob, Bash, Write
argument-hint
<pdf_path> [output_path]
disable-model-invocation
false

03 SKILL.md 正文

Workshop Session PDF — Recap & Next Steps

Extract content from a single workshop session PDF and produce a focused recap organized around three questions:

  1. What was learned? — the durable takeaways, concepts, tools, and techniques surfaced in the session.
  2. What are the next steps? — the actions, follow-ups, and practice work the attendee should do after the session.
  3. What enhancements could be added? — improvements, extensions, or deeper explorations beyond the material presented.

This skill is read-only with respect to the input PDF. It produces a Markdown report and does not modify the source file.

Inputs

Arguments are parsed positionally:

  • <pdf_path> (required) — absolute path to a single .pdf file containing workshop session material (slides, handouts, transcript, lab guide, or any combination).
  • [output_path] (optional) — absolute path where the recap Markdown should be written. If omitted, derive a unique default so repeated runs never overwrite a prior recap:
    • Path: <pdf_path_dirname>/<pdf_basename>-recap-<YYYYMMDD-HHMMSS>-<UUID8>.md
    • YYYYMMDD-HHMMSS is the local-time timestamp at the start of the run (generate once via Bash with date +%Y%m%d-%H%M%S and reuse).
    • UUID8 is an 8-hex-char random suffix (generate via Bash with uuidgen | tr -d '-' | cut -c1-8 | tr A-Z a-z).
    • Announce the resolved output path to the user before writing.

If <pdf_path> is missing, not a file, not readable, or does not end in .pdf, stop and ask the user to supply a valid path. Do not attempt to infer the PDF from nearby files.

Execution Protocol

Run the five phases in order. Each phase's output feeds the next — do not skip.

Phase 1 — PDF Ingestion

  1. Verify the PDF exists and is readable (Bash with ls -l <pdf_path> is sufficient).
  2. Determine the page count before reading. Use Bash with mdls -name kMDItemNumberOfPages <pdf_path> on macOS, or fall back to reading the PDF incrementally.
  3. Read the PDF content via the Read tool:
    • For PDFs ≤ 10 pages: read without the pages parameter.
    • For PDFs > 10 pages: read in ranges of up to 20 pages per call (e.g., pages: "1-20", then pages: "21-40", …) until all pages are consumed. The Read tool requires the pages parameter for PDFs longer than 10 pages.
  4. While reading, capture page-anchored notes in a working scratchpad:
    • Session title, presenter(s), date, venue (if shown on cover/footer).
    • Agenda or table of contents (if present).
    • Per-section topics, key terms, diagrams, and code/lab references, each tagged with its page number(s).
    • Explicit "takeaways", "key points", "summary", "homework", "next steps", "further reading" slides — these are primary source material for the recap and must be captured verbatim or near-verbatim.
  5. Do not fabricate content. If a detail is not visible in the PDF, mark it NOT IN SOURCE rather than inferring from training data or prior conversations.

Phase 2 — Session Frame

Produce a short session frame block (emitted later as ## Session) capturing:

  • Title — from the cover slide or document metadata; NOT IN SOURCE if absent.
  • Presenter(s) — as printed; NOT IN SOURCE if absent.
  • Date / Event — as printed; NOT IN SOURCE if absent.
  • Format — e.g., slide deck, lab guide, workbook, transcript, hybrid (infer from the structure you observed, not the filename).
  • Page count — total pages read.
  • Scope — one sentence describing the stated subject of the session.

Phase 3 — Learning Recap Extraction

Under ## Learning Recap, extract what was actually taught. Each entry must be traceable.

Group by:

  • Concepts & Frameworks — ideas, models, mental frameworks introduced.
  • Tools & Techniques — specific tools, commands, libraries, or step-by-step techniques demonstrated.
  • Examples & Exercises — worked examples, labs, demos, or exercises walked through.
  • Quoted Takeaways — any explicit "key point", "remember this", or summary bullets from the source, quoted near-verbatim.

For each entry emit:

L-00X — <one-line description of the learning>
  Evidence: <pdf_basename>:p<page>[, p<page>…]
  Category: [Concept | Tool | Example | Takeaway]

Do not include every bullet on every slide — select the items that a motivated attendee would transcribe into their own notes. Aim for a dense but curated list, not an exhaustive transcript.

Phase 4 — Next Steps Extraction

Under ## Next Steps, capture the actions the session implies or prescribes for the attendee. Sources of next steps, in priority order:

  1. Explicit — slides literally titled "Next Steps", "Homework", "Action Items", "Try This", "Further Reading", "Exercises to Do". These are authoritative and must be captured.
  2. Implicit — actions the material strongly implies (e.g., a tool was introduced but not practiced → "practice with <tool>"; a concept was taught with a promised follow-up → "review <topic>").
  3. Referenced resources — links, books, docs, repos, or papers cited in the deck that the attendee should consult.

For each entry emit:

N-00X — <imperative action starting with a verb>
  Source: [Explicit | Implicit | Referenced]
  Evidence: <pdf_basename>:p<page> (or "derived from p<page> — <reason>" for Implicit)
  Effort: [Quick (<1h) | Medium (half-day) | Deep (multi-day)]

If the session prescribed zero explicit next steps and you cannot defensibly derive any implicit ones, say so — do not manufacture tasks.

Phase 5 — Enhancement Opportunities

Under ## Enhancement Opportunities, propose extensions beyond the material. These are your recommendations, not content from the PDF. Each proposal must be anchored to something the session did present — this prevents drift into unrelated territory.

Categorize each as:

  • Depth — go deeper on a topic the session introduced but did not fully develop.
  • Breadth — explore an adjacent topic the session gestured at but did not cover.
  • Applied — apply the learning to a realistic project or scenario not used in the session.
  • Tooling — automate, script, or templatize a manual process demonstrated in the session.

For each entry emit:

E-00X — <proposed enhancement>
  Category: [Depth | Breadth | Applied | Tooling]
  Anchor: <which learning this builds on — reference an L-00X ID from Phase 3>
  Rationale: <one sentence: why this adds value beyond the session>

If an enhancement cannot be anchored to an L-00X ID, drop it — unanchored suggestions are out of scope.

Output Contract

Write the recap to <output_path> (or the default derived path) as a single Markdown document with these top-level sections, in this order:

  1. ## Session — frame block from Phase 2.
  2. ## Learning Recap — grouped L-00X entries from Phase 3.
  3. ## Next Steps — N-00X entries from Phase 4, ordered: all Explicit first, then Implicit, then Referenced.
  4. ## Enhancement Opportunities — E-00X entries from Phase 5, grouped by category.
  5. ## Traceability Index — a table mapping every L / N / E ID to its page reference (or anchor ID for enhancements) and its category, so each row is verifiable against the source PDF.

After writing the file, print a one-paragraph summary to the conversation: the output file path, the counts of L / N / E entries, and one sentence naming the session's central theme. Do not repeat the full document inline.

Operational Boundaries

  • Do not modify the source PDF or any file in its directory other than the output recap.
  • Do not read files other than the specified PDF. No sibling scanning, no prior-session comparisons, no repo exploration. The input is one PDF; the output is one Markdown file.
  • Do not invent content. Every L-00X and N-00X (Explicit/Referenced) entry must cite a page. Implicit next steps must name the page that implied them. Enhancements must anchor to an L-00X ID.
  • Do not expand scope beyond recap + next steps + enhancements. Resist adding glossaries, cheat sheets, quizzes, or presenter critiques unless the user explicitly asks.
  • Stop and ask when: the PDF is unreadable, encrypted, empty, image-only with no extractable text (OCR is out of scope), or clearly not workshop material (e.g., a legal contract, invoice, novel).

Example Invocation

/workshop-session-pdf-recap-next /Users/me/Downloads/claude-agents-workshop.pdf

Expected output skeleton (written to, e.g., /Users/me/Downloads/claude-agents-workshop-recap-20260421-143522-a1b2c3d4.md):

## Session
- Title: Building Claude Agents — Hands-On Workshop
- Presenter(s): J. Doe, A. Smith
- Date / Event: 2026-03-14, Anthropic DevDay
- Format: Slide deck + lab guide (hybrid)
- Page count: 42
- Scope: Introduces the Agent SDK and walks attendees through building a research agent end-to-end.

## Learning Recap

### Concepts & Frameworks
L-001 — Agent loop: plan → act → observe → reflect
  Evidence: claude-agents-workshop.pdf:p4, p5
  Category: Concept

L-002 — Tool-use contract: schema-first tool definitions
  Evidence: claude-agents-workshop.pdf:p11
  Category: Concept

### Tools & Techniques
L-003 — Using `ToolSearch` to defer schema loading until needed
  Evidence: claude-agents-workshop.pdf:p17, p18
  Category: Tool

### Examples & Exercises
L-004 — Lab: build a 3-tool research agent that summarizes a URL
  Evidence: claude-agents-workshop.pdf:p22-p27
  Category: Example

### Quoted Takeaways
L-005 — "Prefer small, composable tools over one monolithic tool." (slide 31)
  Evidence: claude-agents-workshop.pdf:p31
  Category: Takeaway

## Next Steps
N-001 — Complete the post-workshop lab in the provided repo
  Source: Explicit
  Evidence: claude-agents-workshop.pdf:p40
  Effort: Medium (half-day)

N-002 — Read the Agent SDK reference docs sections listed on slide 41
  Source: Referenced
  Evidence: claude-agents-workshop.pdf:p41
  Effort: Quick (<1h)

N-003 — Practice defining a new tool schema from scratch
  Source: Implicit
  Evidence: derived from p11 — tool-use contract was introduced but not rehearsed by the attendee
  Effort: Quick (<1h)

## Enhancement Opportunities

### Depth
E-001 — Add structured output (JSON schema) to the research agent from the lab
  Category: Depth
  Anchor: L-004
  Rationale: The lab stops at free-text output; JSON output unlocks downstream automation.

### Applied
E-002 — Adapt the agent loop to a PR-review use case against a real repo
  Category: Applied
  Anchor: L-001
  Rationale: The session used a research example; porting the loop to code review tests generalization.

### Tooling
E-003 — Package the 3-tool stack as a reusable npm module with tests
  Category: Tooling
  Anchor: L-003
  Rationale: Turns one-off lab code into a durable library the attendee can reuse.

## Traceability Index
| ID    | Type        | Category / Source  | Evidence                                |
|-------|-------------|--------------------|-----------------------------------------|
| L-001 | Learning    | Concept            | claude-agents-workshop.pdf:p4, p5       |
| L-002 | Learning    | Concept            | claude-agents-workshop.pdf:p11          |
| L-003 | Learning    | Tool               | claude-agents-workshop.pdf:p17, p18     |
| L-004 | Learning    | Example            | claude-agents-workshop.pdf:p22-p27      |
| L-005 | Learning    | Takeaway           | claude-agents-workshop.pdf:p31          |
| N-001 | Next Step   | Explicit           | claude-agents-workshop.pdf:p40          |
| N-002 | Next Step   | Referenced         | claude-agents-workshop.pdf:p41          |
| N-003 | Next Step   | Implicit           | derived from p11                        |
| E-001 | Enhancement | Depth → L-004      | anchor only                             |
| E-002 | Enhancement | Applied → L-001    | anchor only                             |
| E-003 | Enhancement | Tooling → L-003    | anchor only                             |

Notes on PDF Handling

  • The Read tool natively supports PDF. For PDFs > 10 pages, the pages parameter is required (max 20 pages per call). Plan your reads before starting so you cover every page exactly once.
  • If the PDF is image-only (scanned slides with no text layer), the Read tool will return little or no text. In that case, stop and report — OCR is out of scope for this skill.
  • Treat the PDF basename (without extension) as the default slug when constructing the output path and when labelling evidence.

04 同仓库更多 Skills

carlosmarte
carlosmarte/ai-agent-artifacts

validation_handoff

Forces the agent to conclude its response by generating a strict validation checklist and outlining pending actions, focusing heavily on defensive execution and edge-case testing.

★ 0
carlosmarte
carlosmarte/ai-agent-artifacts

project-ployglot-project-scafford

Scaffold a polyglot library project as TypeScript + Python twin packages under `packages/ts/` + `packages/py/`, with a shared `SPEC.md`, cross-language `tests/parity/fixtures.json`, side-by-side `examples/sdk/` + `examples/api/` documentation, per-package `Makefile`, and a root orchestrator that exposes `make ci`. Open-source friendly — no organization-specific paths or scopes baked in. Use when starting a new polyglot library, porting a single-language library to a twin, or auditing/regularizing an existing two-language repo against the canonical layout.

★ 0
carlosmarte
carlosmarte/ai-agent-artifacts

playwright-ui-locator

Auto-heal Playwright tests after a frontend developer changes a div's class, id, or other selector. Catches a failing Playwright test, captures the old vs new DOM tree, finds the element that drifted, derives the most stable replacement locator, and proposes the fix as a GitHub pull request (confident heals) or files a GitHub issue (ambiguous / element-removed cases) — never editing the working tree directly. Trigger when Playwright tests fail with "locator resolved to 0 elements", "Timeout waiting for selector", or after a UI refactor / class rename / Tailwind cleanup broke previously-passing tests.

设计师代码审查测试与质量
★ 0
carlosmarte
carlosmarte/ai-agent-artifacts

org-learn-context-window

Distill the current conversation into a reusable command or skill artifact. Use when the user wants to capture a workflow they just executed (create a reusable slash command) or formalize a resolution/mistake/correction into a reusable skill so the same class of issue never recurs.

Skill与智能体开发
★ 0
carlosmarte
carlosmarte/ai-agent-artifacts

org-agent-skill-creator

Scaffold and validate scope-aware agent skills (company, enterprise, org, team, project, application). Use when the user asks to create a new agent skill, generate a SKILL.md file, scaffold a skill directory, or validate that existing skill directories meet SKILL_SPEC.md (frontmatter, naming rules, structure, scope/tier).

Skill与智能体开发
★ 0
carlosmarte
carlosmarte/ai-agent-artifacts

app-workshop-recaps-consolidate

Consolidate multiple workshop-session recap Markdown files (typically produced by the workshop-session-pdf-recap-next skill) into a single unified report. Merges and deduplicates learnings across sessions and emits six sections — Consolidated Recap, Follow-ups, Action Items, Next Steps, Possible Enhancements, and Industry / External Viewpoint. The last section supplements the attendees' material with current external perspective via web research. Use when the user has a set of per-session recap files and wants one portfolio-level summary with an outside-in lens ("consolidate these recaps", "merge my workshop notes", "cross-session summary", "roll up action items").

文档处理
★ 0