follow-up-after-call

作者: openai

將最近的客戶、合作夥伴或重要內部通話記錄或基於通話的筆記,轉換為賣家可用的跟進資料包,包含摘要、後續步驟……

npx skills add https://github.com/openai/role-specific-plugins --skill follow-up-after-call

Follow Up After Call

Context-Gathering Intake

Whenever this skill asks for context, strongly prefer using the answers-ask-user-input skill and the ask_user_input tool over other tools such as request_user_input; otherwise ask directly in the conversation.

Turn grounded call evidence into a seller-ready follow-up package that preserves what was actually said, makes ownership visible, and gives the seller copy they can use immediately. This skill owns the post-call synthesis and drafts; it does not send, post, create email drafts, or update CRM.

Common Skill Instructions

MANDATORY: If not already in context, read and adhere closely to plugins/sales/skills/index/SKILL.md## Cross-Skill Best Practices.

Key Dependency Categories

These categories are particularly important for this workflow; use other sources only when they materially improve the package.

  • [Blocking] ~~Meeting Transcripts for the primary transcript, grounded notes, participant language, decisions, commitments, objections, and source links. It blocks the default live-source follow-up path; explicit grounded call evidence already in context satisfies the need.
  • ~~Calendar for recent-call identification, attendees, timing, and upcoming related meetings
  • ~~CRM for account, opportunity, contact, owner, and CRM-ready next-step context
  • ~~Email for customer-facing thread context when the call evidence or promised follow-up lives there
  • ~~Internal Messaging for internal account context and a safe suggested destination for the team recap
  • ~~Knowledge & Files for exported transcripts, call notes, account notes, and follow-up assets

Grounded call evidence is required. CRM, calendar, files, and messages can enrich the package but cannot substitute for a transcript, grounded notes, or a customer-facing thread that records the call. If evidence is missing, stale, or conflicting, state the limitation rather than reconstructing the call.

Reference Loading

SKILL.md owns the normal evidence-first workflow, chat-first draft behavior, and output package. Load references only when their extra detail changes the decision:

Workflow Guidance

1. Resolve the call and evidence

  • If the user supplies a transcript, notes, link, or clearly identified call, start there.
  • If the call-resolution path isn't clear, ask the user via ask_user_input(). For a bare invocation such as “Follow up after my call,” do not assume a meeting; offer:
    • Use my most appropriate recent grounded call
    • Use a call, account, attendee, or date I specify
    • Use transcript or notes I provide
  • For “my latest call,” “today’s call,” “earlier today,” “yesterday,” “last week,” “my most appropriate recent meeting,” or a named account without a source, treat the wording as a retrievable anchor: search the narrowest recent window in ~~Calendar and ~~Meeting Transcripts, then offer up to five concrete candidates via ask_user_input(). Do not retrieve deeper context or draft a package until the user selects one.
  • When the call or account anchor is ambiguous, make at most three source reads to produce concrete candidates; do not gather broad enrichment before the user chooses.
  • Treat ambiguous company-like names, partner names, and account shorthands as possible account anchors. When ~~CRM is available and account identity affects which call to retrieve, use a bounded ~~CRM lookup to disambiguate the account before relying on weaker account context; it never substitutes for grounded call evidence.
  • If exactly one plausible grounded candidate remains, ask the user to confirm it via ask_user_input() before drafting. If no grounded evidence can be found, ask for the transcript, notes, or specific call rather than producing a recap from surrounding account context.
  • Treat pasted notes, uploads, and user-linked material as valid working evidence; label them as user-provided when they cannot be verified.

2. Gather only useful enrichment

  • Prefer grounded evidence in this order: ~~Meeting Transcripts transcript; retrievable meeting-recording transcript; other exported transcript; pasted transcript or grounded notes; ~~Knowledge & Files meeting notes; then an ~~Email thread or message as a recovery lane.
  • Fetch a user-provided call or transcript handle directly. Otherwise search with the account plus date or timeframe and at most two meeting-topic terms, then fetch the best match.
  • Preserve supplemental source-of-truth, call-notes, or transcript links alongside the primary evidence rather than replacing it.
  • For an external account call, use ~~CRM when available to resolve the relevant account or opportunity and sharpen names, commercial context, and the one-sentence CRM next step.
  • Check ~~Calendar for a related upcoming meeting when timing changes the next step. Use ~~Email, ~~Internal Messaging, and ~~Knowledge & Files only when they add promised actions, blockers, owners, links, or destination context.
  • If the transcript explicitly names a meeting or file that could affect the follow-up, do one targeted lookup in ~~Calendar or ~~Knowledge & Files using the exact title and date when available. Include the link if found; otherwise say it was not found in that checked source/window rather than claiming no such meeting or file exists.
  • Stop once the package is grounded and the highest-value account/timing context is covered; name missing enrichment instead of continuing broad searches.

3. Classify the call and extract commitments

  • Decide whether the call is external customer/partner-facing or internal. External email copy and CRM text apply only to external calls.
  • Extract decisions, commitments, asks, objections, blockers, owners, and dates. Use Unknown or TBD when the evidence does not establish them.
  • If multiple CRM opportunities are plausible, choose the one matching the call topic, attendees, product, timing, and recent activity. If none is credible, keep the CRM sentence generic and append (paste into the relevant opp).
  • When opportunity intent, candidate ranking, or the safest internal channel is not obvious, use references/opportunity-and-channel-selection.md.

4. Draft in chat

  • Return the package in chat by default, even when an email or internal channel is available.
  • Do not create an email draft, send outreach, post an internal message, or write CRM in this workflow.
  • Keep verbatim email and internal-message copy in Markdown block quotes so the seller can review and copy it.

Next Step Options

After the first output, offer the most relevant follow-up from the options below. Offer one clear transition, not a menu. Suggest ONLY these unless you are very confident another option is more useful:

  • Refine the external email, internal recap, CRM sentence, or action list using the user's guidance.
  • Create the reviewed email draft or post the reviewed internal recap when the user explicitly requests it.
  • Add the reviewed recap and action items to an existing meeting note or create a new follow-up note.
  • Prepare CRM-ready updates for review.
  • Check for a matching automation and offer to create one unless a matching automation already exists; if one exists, offer to review or adjust it.

Next steps to avoid:

  • Sending, posting, or updating CRM before the user reviews and explicitly requests the action.

Automation Offer Guard

Always offer a follow-up automation as the end-of-output next step unless a matching automation already exists. If the output identifies a natural timing anchor such as a next meeting, follow-up date, customer decision date, or agreed check-in, use that timing in the offer; otherwise offer to help choose the trigger or cadence. Frame the automation as a scheduled rerun of this skill, not a custom digest: it should use the same call/account anchor where possible and return the standard follow-up package with recap, action items, external copy, CRM-ready text, and internal recap.

Before ending with an automation offer, check whether a matching local automation already exists under $CODEX_HOME/automations/*/automation.toml, or ~/.codex/automations/*/automation.toml when CODEX_HOME is unset. Match by prompt, skill name, account, meeting title, attendee, opportunity, trigger time, or other stable scope details. Treat active and paused matches as already installed.

  • If a matching automation exists, offer to review or adjust it rather than creating another one.
  • If no matching automation exists, end with one clear offer to create the automation, describing the output as a seller-ready follow-up package produced after the next call or at the agreed follow-up time. Do not create or update the automation until the user explicitly agrees.
  • If the automation surface is unavailable, avoid tool details and simply offer to help set up a recurring or scheduled follow-up check when automations are available.

Modes

1. Full Follow-Up Package

Default when the user asks to follow up after a call. Return every section below in order.

2. Focused Draft

Use when the user asks only for an email, internal recap, CRM note, or next-step extraction. Still ground it in call evidence; return the requested section plus a compact Call Summary, material gaps, and the status line.

Output Format

Use bold section labels, not top-level headings, for the normal package.

**Call Summary**

- **Primary evidence:** [Clickable transcript/notes link or source label]
- **TL;DR:** [2-4 grounded bullets]
- **Context / Goal:** [Why the call happened]
- **Key Points:** [Important customer/internal signals]
- **Decisions / Commitments:** [What was agreed]
- **Risks / Blockers:** [What could slow follow-through]

**Next Steps**

**Customer**
- [ ] [Action] — [Owner or Unknown] — [Due date or TBD] — [Notes]

**Seller**
- [ ] [Action] — [Owner or Unknown] — [Due date or TBD] — [Notes]

**External Comms**

Subject options (recommended first):
1. [Subject]
2. [Subject]
3. [Subject]

Recommended subject: [Subject]

> Hi <FirstName>,
>
> [150-220 word grounded follow-up email]
>
> Best,
> [Seller]

Draft link: Not created; copy drafted in chat only.

**CRM Next Steps**

[Exactly one sentence with call date, seller action, customer action, and outcome/success criterion.]

**Internal Follow-Up**

Suggested destination: [Safe internal channel/link, or “No verified internal channel found”]

> **[Meeting title]**
>
> [One concise summary paragraph]
>
> **Attendees:** [Names or Unknown]
>
> **Key Notes**
> - [Grounded note]
>
> **Decisions**
> - [Decision or None confirmed]
>
> **Action Items**
> - [Action] — [Owner or TBD] — [Due date or TBD]
>
> **Open Questions / Risks**
> - [Question or risk]
>
> **Source**
> - [Clickable call-notes/transcript link, or no useful link available]

No email draft, Slack post, or CRM update was created.

---

{Follow the instructions and output format/conditions in [Limitations and Improvements](../index/SKILL.md#limitations-and-improvements)}

{Follow the instructions and output format/conditions in [Next Steps](../index/SKILL.md#4-next-steps)}

Output Rules

  • Bias toward decisions, risks, next steps, buying process, procurement, technical blockers, and stakeholder movement.
  • For an internal call, write exactly Not applicable: this was an internal call. under both External Comms and CRM Next Steps; do not draft external copy.
  • Keep the external email concise, customer-safe, and free of sensitive internal language. Never invent recipients, commitments, pricing, dates, or draft links.
  • Keep the CRM text to exactly one sentence. For an unresolved opportunity, append (paste into the relevant opp).
  • Suggest an internal destination only when it appears internal and relevant. If channel safety is unclear, say to verify before posting; never fabricate a channel URL.
  • Before saying no verified internal destination exists, search ~~Internal Messaging once for the account, meeting title, or workstream when that source is available.
  • Use useful clickable source links when available, and name source gaps that materially lower confidence.
  • Default to one consolidated internal recap; split it into a top-level message plus thread only when the user asks or a verified channel norm requires it.
  • Use native mentions for attendees and owners when the selected app supports them; otherwise use readable names.
  • Keep the internal draft free of ## headings. Its opening summary should name the launch path, business impact, decision point, or main follow-through theme without repeating the TL;DR.
  • Keep the internal recap compact, de-duplicated, and team-facing; include owners or TBD rather than implying ownership.
  • Always include the plain status line above after Internal Follow-Up unless the user explicitly asks for a different status format.

Failure Handling

If the call cannot be grounded, state the blocker in one line, say what evidence is needed, and return only safe partial material such as candidate calls or a blank fill-in structure. If optional enrichment is unavailable, still produce the grounded package and label the missing lane.

來自 openai 的更多技能

user-context
openai
載入或管理資料分析插件的持久性來源路由偏好設定、入門邏輯、設定進度以及語義層註冊表。
official
notion-research-documentation
openai
研究Notion內容,並將其整合為結構化的簡報、報告或附有引用的比較分析。透過精準查詢搜尋並擷取Notion頁面,然後按主題組織發現,並附上內文來源引用及參考文獻章節。根據範圍與使用者目標,從四種輸出格式(快速簡報、研究摘要、比較分析、綜合報告)中選擇。使用內建範本建立與更新Notion頁面;直接連結來源,並在新資訊到來時追蹤變更。
official
rcsb-pdb-skill
openai
提交精簡的RCSB PDB請求以獲取核心元數據、Search API查詢及FASTA下載。當用戶需要簡潔的RCSB摘要時使用;保存原始JSON或…
official
pdf
openai
PDF 讀取、建立與驗證,支援視覺化渲染與程式化生成。使用 Poppler(pdftoppm)將 PDF 頁面渲染為 PNG,以便在交付前檢查版面、間距與排版。透過 reportlab 程式化生成 PDF,確保格式可靠;使用 pdfplumber 或 pypdf 提取文字與元資料。執行品質標準:無文字裁切、元素重疊、表格破損或渲染瑕疵;僅使用 ASCII 連字號,引用格式需具可讀性。使用...
official
test-coverage-improver
openai
Improve test coverage in the OpenAI Agents JS monorepo: run `pnpm test:coverage`, inspect coverage artifacts, identify low-coverage files and branches, propose…
official
playwright
openai
終端驅動的瀏覽器自動化,具備元素快照與互動式UI工作流程。透過playwright-cli包裝腳本運作(需npx),支援無頭模式與有頭模式以進行視覺化除錯。核心流程:開啟頁面、擷取快照以取得穩定元素參照、使用參照進行互動、在導航或DOM變更後重新擷取快照。包含表單填寫、點擊、輸入、多分頁管理、螢幕截圖/PDF擷取,以及流程除錯的軌跡記錄。元素參照(例如e3、e15)...
official
ukb-topmed-phewas-skill
openai
接受rsID、GRCh37或GRCh38輸入,解析為所需的GRCh38查詢,以獲取單一變異的簡潔UKB-TOPMed PheWAS摘要。當需要…時使用。
official
code-review-context
openai
模型可見上下文
official