Sourcing Agent
Recruiters: Build and refresh a candidate research Sheet.
View Sourcing Agent →Claude works well when a workflow needs large-context reading and deliberate drafts. Its honest limit: external actions and recurring runs require connected tools or an automation layer.
Verified free workflows come first. Each card links to a live existing product. Free copy-ready prompts follow below.
Recruiters: Build and refresh a candidate research Sheet.
View Sourcing Agent →Marketing: Maintain a recurring content calendar Sheet.
View Social Media Calendar Agent →Sales: Keep follow-ups and next steps current in a Sheet.
View Follow-Up & Pipeline Agent →Finance: Produce a recurring management report artifact.
View Management Reporting Agent →Bookkeepers: Run the reconciliation checklist and exception Sheet.
View Reconciliation Agent →Real estate: Maintain a closing checklist and review queue.
View Transaction Checklist Agent →Founders: Publish a recurring KPI update from a stable Sheet.
View KPI Update Agent →Use these prompts at no charge. They are not products. Only genuine matches inside this section carry a clearly marked free agent download link.
Trigger: weekdays at 8:30 AM. Inputs: approved inbox, VIP list and reply rules. Work: classify new mail into reply, delegate, schedule and archive queues. Output: an inbox-review Sheet with message links, urgency and next action. Boundary: no reply, deletion or archive without the approved operating rule. Result: zero unclassified messages in the checked window.
Use Claude and keep the source files in the same Project. Check my approved inbox every weekday at 8:30 AM. Sort new emails into reply, delegate, schedule or archive. Put the list in a Sheet with links and the next step. Do not reply, delete or archive anything unless I have already allowed it.
Trigger: a message enters an approved thread and meets the reply rule. Inputs: full thread, recipient, facts and voice examples. Work: draft a thread-aware response and check claims. Output: exact reply draft in a review queue. Boundary: recipient and final words need approval unless a precise standing send rule exists. Result: a ready response without losing thread context.
Use Claude and keep the source files in the same Project. Read the full message thread and write a short reply in my style. Use only facts from the thread and my notes. Show me the recipient and final message before sending. Do not send it until I approve.
Trigger: an approved sales or fundraising brief is complete. Inputs: audience, problem, proof, metrics, ask and brand assets. Work: build a conclusion-led story and source every claim. Output: Google Slides deck plus evidence log. Boundary: no invented traction or external share. Result: a review-ready deck with one decision per slide.
Use Claude and keep the source files in the same Project. Turn my brief, facts and brand files into a clear pitch deck. Keep one main idea on each slide and do not invent numbers. Add source notes for important claims. Show me the finished presentation before sharing it.
Trigger: an approved prospect batch enters the outreach Sheet. Inputs: ICP, prospect evidence, sender, offer and exclusions. Work: research fit and prepare personalized first lines and messages. Output: outreach review Sheet. Boundary: no send without exact recipient-and-copy approval. Result: every draft cites a specific fit signal.
Researches prospects and prepares evidence-backed outreach for review.
View Outreach Agent →Trigger: a reporting period closes. Inputs: approved metrics, prior report and narrative rules. Work: explain changes and build charts and conclusion-led slides. Output: presentation plus source table. Boundary: publication and distribution require approval. Result: every figure matches the source period and workbook.
Use Claude and keep the source files in the same Project. Turn my approved numbers and notes into a clear presentation. Explain what changed, use simple charts and add a source table. Do not invent missing data. Show me the finished presentation before sharing it.
Trigger: a discovery-call transcript is approved. Inputs: transcript, service scope, proof, pricing rules and proposal template. Work: extract needs, objections and commitments, then draft a tailored proposal. Output: proposal Doc or Slides with transcript evidence. Boundary: no unsupported promise, price change or client send. Result: each proposed solution maps to a stated customer need.
Use Claude and keep the source files in the same Project. Read the approved call transcript and draft a proposal for that customer. Use their needs, questions and agreed next steps, and follow my pricing rules. Mark anything that was not agreed. Show me the final proposal before sending it.
Trigger: a weekly crawl or Search Console change. Inputs: sitemap, query data, page templates and brand rules. Work: identify technical and content opportunities and prepare approved updates. Output: SEO action Sheet and page briefs. Boundary: no site publish or indexation change without approval. Result: every task has target URL, query evidence and acceptance check.
Use Claude and keep the source files in the same Project. Check my site and Search Console once a week. Find broken pages, missing details and pages that could rank better. Give me a short list with the page link and suggested fix. Show me any site change before publishing it.
Trigger: an approved site brief or issue enters the backlog. Inputs: source repository, design rules, content and deploy route. Work: build or patch the site, run tests and preview changes. Output: versioned source and deployment report. Boundary: production deploy, domain and payment changes need approval. Result: acceptance tests and visual QA pass on the intended pages.
Use Claude and keep the source files in the same Project. Use my approved site files and brief to make the requested website change. Test the links, forms and mobile layout, and give me a preview. Keep a copy of the earlier version. Do not publish the change until I approve.
Trigger: a reporting Sheet refreshes or the period closes. Inputs: approved dataset, metric definitions and prior period. Work: validate data, calculate trends and explain changes. Output: analysis tabs and a report tab in Google Sheets. Boundary: no source-row overwrite without approval. Result: figures reconcile and each insight links to its range.
Builds recurring channel reports in a controlled Google Sheet.
View Performance Reporting Agent →Trigger: an approved intake process is defined. Inputs: audience, fields, validation, consent text and destination Sheet. Work: design sections, field types and response logic. Output: Google Form plus mapped response Sheet. Boundary: publishing or collecting personal data needs owner approval. Result: a fictional submission passes validation and lands in the right columns.
Use Claude and keep the source files in the same Project. Create a Google Form for my approved purpose. Use clear questions, helpful choices and the right required fields. Link the answers to a Google Sheet and test the form. Show me the final form before sharing it.
Trigger: a user gives a research question and decision criteria. Inputs: scope, preferred sources, recency and exclusions. Work: search, open primary sources and compare findings. Output: cited research Doc or decision Sheet. Boundary: no login, purchase, form submission or disclosure outside scope. Result: every action-driving claim has a live source and checked date.
Use Claude and keep the source files in the same Project. Research my question using current, trusted sources. Open the main sources, compare what they say and include the links and dates. Tell me what is certain and what is unclear. Do not sign up, contact anyone or Download anything without asking me.
Trigger: the approved ICP and territory are added to the prospecting Sheet. Inputs: ICP, exclusions, target count and evidence sources. Work: find companies and decision-makers, validate fit and deduplicate. Output: prospect Sheet with evidence, source and confidence. Boundary: no enrichment from prohibited private sources or outreach without approval. Result: every row meets the minimum fit evidence.
Researches approved prospects, records fit evidence and prepares outreach for review.
View Outreach Agent →Trigger: a repeat process has agreed stages, owners and dates. Inputs: workflow, fields, statuses, permissions and alert rules. Work: create tabs, validation, formulas and activity log. Output: working Google Sheet tracker. Boundary: no migration or overwrite of live records without approval. Result: a fictional row moves through every stage correctly.
Use Claude and keep the source files in the same Project. Build a Google Sheet to track this work. Add clear columns, owners, dates, status choices and simple checks for missing or repeated items. Include a summary view. Show me the Sheet before sharing it or adding alerts.
Trigger: weekdays at 7:30 AM or when a tracked company, topic or regulator is named in a credible source. Inputs: tracked terms, source preferences, exclusions, recency window and prior-news log. Work: search current sources, open the primary announcement, verify coverage and deduplicate by canonical URL plus event fingerprint. Output: a dated news-monitor Sheet and short briefing Doc. Boundary: no public post, investor note or customer message without approval. Result: only new verified events appear once, with source and checked time.
Use Claude and keep the source files in the same Project. Every weekday at 7:30 AM, check trusted news and official sources for my topics. Skip stories I have already seen and include the original links and dates. Give me a short summary of what changed and why it matters. Do not publish or send it anywhere without asking me.
Trigger: the approved weekly publishing clock after the news/source queue closes. Inputs: audience, editorial angle, approved source queue, prior issues, voice and length. Work: select non-duplicate stories, summarize with links, write the intro and sections, then check claims and repeated topics. Output: newsletter Doc plus issue log. Boundary: no subscriber import, email send or publication without exact audience-and-copy approval. Result: a complete issue with every item sourced and no story repeated from the prior issue.
Use Claude and keep the source files in the same Project. Once a week, turn my approved links into a short newsletter for my audience. Do not repeat stories from earlier issues. Add a clear title, short sections and links to every source. Show me the full draft before publishing or sending it.
These are informational patterns grounded in the sources below. They are not products sold by AI Agent Skills.
Trigger: counsel uploads the current agreement, proposed redline and playbook. Claude maps clause changes, risks and unanswered questions into a comparison Doc. Boundary: it does not give legal approval or send a response.
Trigger: a PM adds interview transcripts, support themes and the prior roadmap to a Claude Project. Claude groups evidence, quotes the sources and produces a decision brief. Boundary: roadmap decisions stay with the PM.
Trigger: a draft proposal and client brief enter the project. Claude checks scope, evidence, assumptions and contradictions, then returns a review Doc with exact edits. Boundary: no invented case study or commercial promise.
One-off: use a clear prompt when the question changes each time. Recurring: use a stable schedule or trigger only when the same sources and checks apply. Artifact rule: finish in a Form, Doc, presentation or Sheet, not a chat-only answer.
Use the Claude setup hub for profession guides and current setup notes. AI Agent Skills is independent and is not affiliated with or endorsed by Claude.
Community evidence also reviewed: Reddit discussion of real-world assistant uses. Community claims were treated as leads; product-specific claims above rely on official or reported sources.
Reviewed 18 September 2026 · AI Agent Skills team.