Agent Skills Library

The following pre-built skills are available out of the box, organized by category. You can use these skills as-is or customize them to fit your specific needs.

Each skill below lists:

• Overview — what the skill is for
• Allowed agent toolkits — toolkits the agent typically needs enabled (connect OAuth/API credentials where required)
• Trigger keywords — phrases that help the runtime match the skill
• Instructions — the procedural steps the agent follows

Install or assign skills from the agent Skills tab or Skills library. See Getting Started.

Development

code-review — Structured code review checklist covering quality, security, and best practices.

Overview: Gives your agent a consistent, structured approach to reviewing code. When triggered, the agent evaluates submissions across six dimensions — correctness, security, performance, readability, testing, and architecture — and outputs findings in a standardized format with severity ratings and actionable recommendations.

Allowed agent toolkits: knowledge base

Trigger keywords: code review, review code, review this code, review my code

Instructions:

When reviewing code:

1. Correctness — Does the code do what it's supposed to? Edge cases handled?
2. Security — Any injection risks, exposed secrets, or auth bypasses?
3. Performance — Obvious N+1 queries, unnecessary loops, or memory leaks?
4. Readability — Clear naming, appropriate comments, consistent style?
5. Testing — Are there tests? Do they cover the happy path and edge cases?
6. Architecture — Does this fit the existing patterns? Any unnecessary coupling?

For each finding, provide:

• Severity: Critical / Warning / Suggestion
• Location: File and line reference
• Description: What the issue is
• Recommendation: How to fix it

github-ops — GitHub issue and pull request management.

Overview: Enables your agent to manage GitHub workflows end-to-end. From creating and labeling issues to reviewing pull requests and surfacing stale items, this skill keeps your repository organized and your team unblocked — without leaving the agent.

Allowed agent toolkits: github

Trigger keywords: github, pull request, PR review, issue tracker, repository

Instructions:

When the user asks about GitHub issues, pull requests, or repository management:

1. Issue management
   • Search existing issues before creating duplicates
   • Apply appropriate labels and assignees
   • Link related issues when relevant
2. Pull request review
   • Summarize the changes in the PR
   • Check for potential issues (missing tests, breaking changes)
   • Suggest improvements with specific line references
3. Repository insights
   • Provide activity summaries when asked
   • Track open issues and PR counts
   • Highlight stale items that need attention

github-repository-explorer — Discover, search, and inspect GitHub repositories, branch structures, commit…

Overview: Discover, search, and inspect GitHub repositories, branch structures, commit histories, and file contents (such as README or source code).

Allowed agent toolkits: github

Trigger keywords: list repos, search repos, get repository, explore repository, read file, repo branches, commit history, github repo, view readme, find repository

Instructions:

1. Discover or Search Repositories — Use github_list_repositories to list repositories accessible to the connection. Filter by visibility or affiliation if specified.
2. Get Repository Details — Use github_get_repository with owner and repo to fetch full metadata (name, full_name, description, default_branch, open_issues_count, private status).
3. Inspect Branches and File Contents — Use github_list_branches to enumerate repository branches and their head commit SHAs.
4. Retrieve Commit History — Use github_get_commit_history to fetch commit logs. Filter by sha/branch, path, author, or since/until ISO 8601 timestamps.
5. Synthesize Findings — Summarize repository structure, default branch, recent commits, and requested file contents clearly for the user.

github-issue-manager — List, filter, search for existing issues to avoid duplicates, check available…

Overview: List, filter, search for existing issues to avoid duplicates, check available labels, and safely create new repository issues with explicit user confirmation.

Allowed agent toolkits: github

Trigger keywords: list issues, create issue, open issue, search issues, github issue, report bug, issue triage, add issue, issue tracker, bug report

Instructions:

1. Inspect Repository Context — Use github_get_repository to verify repository existence and fetch metadata.
2. List and Search Existing Issues — Use github_list_issues with filters (state, labels, assignee) to inspect open or closed issues. Before creating a new issue, search existing issues to prevent duplicates with similar titles or subjects.
3. Inspect Labels — Use github_get_labels to list valid labels available in the repository prior to applying them.
4. Safe Issue Creation (Write Operation) — Always present the proposed issue details (owner, repo, title, body, labels) to the user and obtain explicit confirmation before creating. Call github_create_issue only after confirmation. Verify that GitHub returns an issue number and html_url before reporting success.

github-pr-analyzer — Analyze GitHub pull requests in depth by inspecting PR metadata, modified files,…

Overview: Analyze GitHub pull requests in depth by inspecting PR metadata, modified files, change statistics, raw unified diffs, and attached commit messages for code review and risk analysis.

Allowed agent toolkits: github

Trigger keywords: list pull requests, get pr, analyze pr, pr diff, pr files, pr commits, github pr, review pull request, pr review, pull request overview

Instructions:

1. List and Locate PRs — Use github_list_pull_requests to list pull requests filtered by state (open, closed, all) or base branch.
2. Fetch Pull Request Metadata — Use github_get_pull_request with pull_number to inspect PR title, state, mergeability, and addition/deletion line counts.
3. Analyze Changed Files & Code Diffs — Use github_get_pull_request_files to list modified, added, or deleted files and locate high-risk file changes. Use github_get_pull_request_diff to retrieve raw unified diff text for in-depth code review.
4. Inspect Included Commits — Use github_get_pull_request_commits to view individual commit messages and author details attached to the PR.
5. Produce Summary — Synthesize PR overview, changed file risks, code diff highlights, and commit history into a concise review report.

github-pr-creator — Safely create and label pull requests on GitHub, verifying branch existence and…

Overview: Safely create and label pull requests on GitHub, verifying branch existence and differences, checking for duplicate PRs, obtaining user confirmation, and applying PR labels.

Allowed agent toolkits: github

Trigger keywords: create pr, open pull request, create pull request, submit pr, label pr, assign pr label, github pr creation

Instructions:

1. Pre-flight Branch Verification — Use github_list_branches to verify that both head and base branches exist and contain different commit SHAs. If head or base branches do not exist or contain no differences, inform the user and do not attempt to open a PR.
2. Check for Duplicate PRs — Use github_list_pull_requests with base filter and inspect the returned PR list to check if an open PR already exists for the head branch.
3. Verify Repository Labels — Use github_get_labels to ensure requested labels exist in the target repository before assignment.
4. Safe Pull Request Creation & Labeling (Write Operations) — Always require explicit user confirmation showing owner, repo, title, head, base, and labels before executing writes. Use github_create_pull_request to open the pull request. Use github_assign_pull_request_labels with the returned PR number to apply labels. Confirm returned PR number and html_url before completing.

github-activity-reporter — Generate comprehensive repository activity reports, auditing recent commit…

Overview: Generate comprehensive repository activity reports, auditing recent commit velocity, branch state, open issues, pull request status, and documentation health.

Allowed agent toolkits: github

Trigger keywords: activity report, repo summary, git history, commit activity, triage repo, release readiness, repo status, weekly activity, engineering report

Instructions:

1. Inspect Repository & Branches — Use github_get_repository for core metadata, default branch name, and open issue counts.
2. Audit Recent Commit Velocity — Use github_get_commit_history to inspect commit frequency, active authors, and latest updates over time.
3. Track Open Work (Issues & PRs) — Use github_list_issues to summarize open issues and bug reports.
4. Check Documentation Health — Use github_get_file_content on README.md or configuration files to verify documentation status.
5. Deliver Activity Digest — Present a structured report containing repository metrics, recent commit trends, open issues/PRs, and actionable maintenance recommendations.

Operations

daily-digest — Compile and send a daily morning briefing covering tracked topics.

Overview: Turns your agent into a personal briefing assistant. Every morning, it searches for the latest updates across your tracked topics and compiles them into a clean, structured digest — top stories, categorized updates, action items, and a calendar preview — delivered via your preferred channel.

Allowed agent toolkits: web search

Trigger keywords: daily digest, morning briefing, send briefing, daily summary

Instructions:

When asked to create a daily digest or morning briefing:

1. Check the current date and identify key topics the user tracks
2. Search for recent news, updates, and changes on each tracked topic
3. Compile findings into a structured briefing format:
   • Date and greeting
   • Top stories (3–5 most important items)
   • By category — group remaining items by topic area
   • Action items — anything requiring the user's attention today
   • Calendar preview — upcoming events if calendar data is available
4. Keep each item to 2–3 sentences maximum
5. Send via the user's preferred channel if configured

meeting-notes — Summarize meeting transcripts into structured notes with action items.

Overview: Transforms raw meeting transcripts into concise, actionable notes. The agent extracts key decisions, organizes discussion highlights by topic, and produces a clean action item table with owners and deadlines — so nothing falls through the cracks after a call.

Allowed agent toolkits: knowledge base

Trigger keywords: meeting notes, summarize meeting, meeting summary, meeting recap

Instructions:

When summarizing a meeting transcript or recording:

1. Meeting metadata — Date, attendees, duration, topic
2. Key decisions — Bulleted list of decisions made
3. Discussion highlights — Major points discussed, organized by topic
4. Action items — Table format:

Action	Owner	Deadline

Extract every commitment or task mentioned.

5. Open questions — Unresolved items that need follow-up
6. Next meeting — Date/time if mentioned

Keep the summary concise — aim for 1/10th the length of the original transcript.

onboarding-guide — Create structured onboarding guides for new team members or users.

Overview: Helps your agent generate tailored onboarding experiences for new hires or users. It builds comprehensive guides covering week-one essentials, 30-day milestones, key contacts, FAQs, and success metrics — all adapted to the specific role and organization.

Allowed agent toolkits: knowledge base

Trigger keywords: onboarding, new hire, getting started guide, orientation

Instructions:

When creating an onboarding guide:

1. Identify the audience — New employee? New user? Role-specific?
2. Week 1 essentials:
   • Account setup and access provisioning
   • Key tools and where to find them
   • Team introductions and org chart
   • First-day checklist
3. First 30 days:
   • Learning resources and documentation
   • Shadowing schedule
   • Key processes to understand
   • First small tasks or projects
4. Key contacts — Who to ask for what
5. FAQ — Common questions new people have
6. Success metrics — What does a successful first 90 days look like?

Tailor the guide based on available information about the organization and role.

report-builder — Generate structured reports with executive summaries, data sections, and recommendations.

Overview: Equips your agent to produce polished, publication-ready reports from scratch. It pulls data from your knowledge base, database, or the web, then structures it into a complete report with an executive summary, analysis, recommendations, and properly cited sources.

Allowed agent toolkits: knowledge base, database, web search

Trigger keywords: build report, generate report, create report, write report

Instructions:

When asked to create a report:

1. Clarify scope — What is the report about? Who is the audience? What time period?
2. Gather data — Pull from available sources (KB, database, web search)
3. Structure the report:
   • Title page — Report name, date, author
   • Executive summary — 3–5 sentences covering the key takeaway
   • Table of contents — For reports with 3+ sections
   • Data sections — Each with a clear heading, narrative, and supporting data/charts
   • Analysis — What the data means, trends identified
   • Recommendations — Actionable next steps based on findings
   • Appendix — Raw data, methodology notes, or detailed tables
4. Format for readability — Use headers, bullets, tables, and bold for key numbers
5. Cite sources — Reference where data came from

teams-incident-notify — Escalate incidents and alerts through Microsoft Teams: post the details to the…

Overview: Escalate incidents and alerts through Microsoft Teams: post the details to the relevant channel, open a war-room group chat with the responders, and direct-message the people who need to act immediately.

Allowed agent toolkits: microsoft_teams

Trigger keywords: incident, escalate, page the team, alert the team, war room, notify on-call, outage

Instructions:

1. Establish severity and audience — Ask who needs to know if it is not clear. Broad awareness goes to a channel; individual action goes to a direct message. Do both when the incident is serious.
2. Post to the channel — Use list_teams and list_team_channels to find the incident or engineering channel, then send_channel_message with content_type='html'. Include, in this order: what is broken, who is affected, when it started, current status, and who is on it.
3. Direct-message the responders — Use send_direct_message with each person's email address. This creates the 1:1 chat automatically, so no lookup step is needed. Keep each message to the specific ask for that person, not a copy of the channel post.
4. Open a war room when coordination is needed — Resolve each responder with get_user_by_email, then call create_chat with a comma-separated string of their user ids, chat_type='group', and a topic naming the incident. Post the opening context with send_chat_message.
5. Follow up — For status updates, reuse the same chat_id rather than opening a new conversation each time.

Analytics

data-analysis — Structured data analysis workflow: explore, clean, analyze, and visualize datasets.

Overview: Gives your agent a rigorous, end-to-end approach to working with data. From understanding the dataset and checking data quality to running exploratory analysis, building visualizations, and summarizing findings — this skill ensures every analysis is thorough, focused, and clearly communicated.

Allowed agent toolkits: database, python

Trigger keywords: analyze data, data analysis, explore dataset, data insights

Instructions:

When asked to analyze data:

1. Understand the data — What tables/files are available? What do columns represent?
2. Data quality check — Look for nulls, duplicates, outliers, and inconsistencies
3. Exploratory analysis
   • Summary statistics (count, mean, median, min, max)
   • Distribution of key variables
   • Correlation between variables
4. Answer the question — Focus analysis on what the user specifically wants to know
5. Visualize — Create charts where they add clarity (bar, line, scatter, etc.)
6. Summarize findings — Clear narrative with supporting numbers
7. Recommend next steps — What additional analysis could be valuable?

teams-channel-digest — Read recent Microsoft Teams channel history and summarize it: decisions made, action…

Overview: Read recent Microsoft Teams channel history and summarize it: decisions made, action items and their owners, open questions, and anything that needs attention. Designed for catching up after time away.

Allowed agent toolkits: microsoft_teams

Trigger keywords: catch me up, summarize channel, what did i miss, channel digest, recap the channel, what happened in teams

Instructions:

1. Resolve the channel — Use list_teams then list_team_channels. Confirm with the user if the reference is ambiguous.
2. Read the history — Use get_channel_messages with team_id, channel_id and a 'top' value matched to the window asked for. The default of 20 covers a normal day; raise it for a week, but be aware only top-level messages are returned, not thread replies.
3. Normalize the content — Each message carries a content_type. When it is 'html', the body contains raw markup; strip the tags before reading it as prose so quoted HTML does not end up in your summary.
4. Extract what matters, attributing each point to the sender in 'from':
   • Decisions — what was settled, and by whom
   • Action items — who committed to what, and by when if stated
   • Open questions — asked but not answered in the messages you read
   • Needs attention — blockers, escalations, anything time-sensitive
5. Format the digest — Lead with a two-line summary of the overall state, then the sections above as short bullets. Skip empty sections rather than printing 'none'. Include the time range covered.

Communication

email-draft — Draft professional emails with appropriate tone, structure, and call-to-action.

Overview: Helps your agent write clear, professional emails every time. It adapts tone to the recipient and context, structures the message for maximum clarity, and ensures every email ends with a specific, actionable ask — keeping communication concise and effective.

Allowed agent toolkits: none

Trigger keywords: draft email, write email, compose email, email draft

Instructions:

When drafting an email:

1. Determine the context — Who is the recipient? What is the relationship? What is the goal?
2. Choose the tone — Formal, semi-formal, or casual based on context
3. Structure the email:
   • Subject line — Clear and specific (not "Quick question" or "Follow up")
   • Opening — Brief greeting + context/purpose in 1–2 sentences
   • Body — Key information, organized with bullets or short paragraphs
   • Call to action — What you need from the recipient, with a specific deadline if applicable
   • Closing — Appropriate sign-off
4. Keep it concise — Most emails should be under 200 words
5. Review — Check for typos, unclear language, and missing information

teams-channel-broadcast — Post announcements, status updates, and reports to the right Microsoft Teams…

Overview: Post announcements, status updates, and reports to the right Microsoft Teams channel. Resolves the team and channel from a plain-language reference, formats the message for readability, and returns a permalink.

Allowed agent toolkits: microsoft_teams

Trigger keywords: post to teams, announce in teams, send to channel, broadcast update, notify the channel, share in teams

Instructions:

1. Resolve the team — Use list_teams to find the team. Match the user's wording against display names; the filter is a contains match, so a partial word works. If several teams match, ask which one rather than guessing.
2. Resolve the channel — Use list_team_channels with that team id. If the user named no channel, default to 'General'. If the named channel does not exist, list what is available instead of inventing an id.
3. Format the message — Decide on content_type:
   • Plain sentence or short note: leave content_type as 'text'
   • Anything with structure — headings, lists, links, emphasis: use content_type='html' and real tags (<b>, <ul>, <li>, <a href>)
   • Never send Markdown. Asterisks and hyphens render literally in Teams and make the post look broken.
4. Send it — Use send_channel_message with both team_id and channel_id. The channel id alone is not sufficient.
5. Confirm — Report back with the web_url from the response so the user can jump straight to the post.

teams-direct-outreach — Find people in the Microsoft Teams directory and send them personalized direct…

Overview: Find people in the Microsoft Teams directory and send them personalized direct messages. Handles name and email lookup, one-to-one and group conversations, and per-recipient tailoring.

Allowed agent toolkits: microsoft_teams

Trigger keywords: message someone, dm, direct message, reach out to, find a person in teams, ping, message the team

Instructions:

1. Identify the recipients — If you have an email address, that is enough — send_direct_message takes an address and handles the lookup and chat creation itself.
2. Confirm before sending — If a search returns more than one plausible person, ask which one. Never message someone on a guess.
3. Send individually — Use send_direct_message per recipient. Tailor each message; identical text sent to ten people reads like a mailshot and gets ignored. State clearly what you are asking of that person.
4. Group conversations — When people need to see each other's replies, resolve each user id, call create_chat with a comma-separated string of those ids and a descriptive topic, then send_chat_message. Use list_chats first if a suitable conversation may already exist, so you continue it rather than fragmenting the discussion.
5. Report back — List who was messaged and who could not be found.

Support

ticket-triage — Categorize and prioritize incoming support tickets using consistent criteria.

Overview: Brings consistency and speed to your support queue. The agent reads each incoming ticket, categorizes it, assigns a priority level, extracts key details, and suggests the right team for routing — then drafts an acknowledgment response, all in one pass.

Allowed agent toolkits: none

Trigger keywords: ticket triage, triage tickets, categorize ticket, support ticket, classify ticket

Instructions:

When triaging a support ticket:

1. Read the ticket thoroughly, including any attachments or context
2. Categorize into one of: Bug, Feature Request, Question, Account Issue, Integration Issue, Other
3. Assess priority using:
   • Critical — Service down, data loss, security breach
   • High — Major feature broken, many users affected
   • Medium — Feature partially working, workaround exists
   • Low — Cosmetic, nice-to-have, single user edge case
4. Extract key details: affected component, reproduction steps, user impact
5. Suggest routing to the appropriate team (Engineering, Support, Product, Billing)
6. Draft a response acknowledging the issue with an estimated timeline

Research

research-deep — Multi-source deep research compilation with structured output and source citations.

Overview: Turns your agent into a thorough research analyst. It defines the scope, queries multiple sources, cross-references findings for contradictions, and synthesizes everything into a structured report with an executive summary, subtopic breakdowns, and full source citations.

Allowed agent toolkits: web search, knowledge base

Trigger keywords: research, deep dive, investigate, comprehensive analysis

Instructions:

When asked to research a topic thoroughly:

1. Scope definition — Clarify the research question and boundaries
2. Multi-source search — Query at least 3 different sources or angles
3. Cross-reference — Identify agreements and contradictions across sources
4. Synthesize — Compile findings into a structured report:
   • Executive summary (3–5 key points)
   • Detailed findings organized by subtopic
   • Conflicting information flagged clearly
   • Source citations for all claims
5. Actionable takeaways — End with recommended next steps

Productivity

teams-workspace-setup — Stand up a Microsoft Teams workspace for a new project: create the channel, post a…

Overview: Stand up a Microsoft Teams workspace for a new project: create the channel, post a structured kickoff message, and notify the people who should join. Turns a one-line request into a ready-to-use space.

Allowed agent toolkits: microsoft_teams

Trigger keywords: create a channel, set up a channel, new project channel, spin up a workspace, kickoff channel, onboard the team

Instructions:

1. Gather what you need — The team it belongs to, the channel name, a one-line purpose, and who should be told about it. Ask for anything missing before creating anything.
2. Pick the team — Use list_teams. Never create a channel in a team the user did not name.
3. Check for a duplicate — Use list_team_channels first. Channel names must be unique within a team, so creating an existing one fails. If it already exists, point the user at it instead.
4. Create the channel — Use create_channel with the team id, the display name, and a description stating the channel's purpose. Leave membership_type as 'standard' unless the user explicitly asked for a private channel.
5. Post a kickoff message — Use send_channel_message with content_type='html' and the new channel id. Cover the purpose, who is involved, where the relevant docs or boards live, and what should be posted here versus elsewhere.
6. Notify the people — Use list_team_members to find them, then send_direct_message with each person's email, including the channel's web_url so they can open it in one click.
7. Report back — Give the user the web_url and a list of who was notified.

Cloud Storage

amazon-s3-object-discovery — Discover S3 buckets and objects, inspect metadata, read text objects, and…

Overview: Discover S3 buckets and objects, inspect metadata, read text objects, and create bounded temporary download links.

Allowed agent toolkits: amazon_s3

Trigger keywords: list s3 buckets, find s3 files, browse s3, read an s3 file, get s3 metadata, create s3 download link

Instructions:

1. Use s3_list_buckets when the bucket is unknown. It has no parameters and requires account-level bucket-list permission.
2. Use s3_list_objects with an explicit bucket when no default bucket is configured. Supply a narrow prefix where possible and keep max_keys bounded; it defaults to 50 and is limited to 1-1000.
3. Use s3_get_file_metadata with key and optional bucket to verify an object's size, type, last modified time, ETag, storage class, and metadata without downloading it.
4. Use s3_read_file with key and optional bucket for a known text object. Text-like content is UTF-8; binary content is base64.
5. Use s3_generate_presigned_url for a temporary download link. It requires key, accepts optional bucket and expires_in, clamps expiry to 60-604800 seconds, and should use the shortest practical expiry.

amazon-s3-object-management — Safely upload, copy, delete, and create Amazon S3 resources with explicit…

Overview: Safely upload, copy, delete, and create Amazon S3 resources with explicit approval for changes.

Allowed agent toolkits: amazon_s3

Trigger keywords: upload to s3, copy s3 file, delete s3 file, create s3 bucket, archive s3 object

Instructions:

1. s3_upload_text requires key and content; optional bucket uses the configured default and content_type defaults to text/plain. It can overwrite an existing object.
2. s3_upload_file requires key and content; optional bucket uses the default and optional content_type is detected from the key when omitted. For a File Upload trigger, use {{trigger.file.name}} and {{trigger.file.content}} only after confirming those values are present.
3. s3_copy_object requires source_key and destination_key. Optional source_bucket uses the default and destination_bucket defaults to the resolved source bucket. Confirm both source and destination because the destination can overwrite an object.
4. s3_delete_file requires key and accepts optional bucket. Show the exact bucket/key and obtain explicit user confirmation immediately before calling it. Recovery depends on bucket versioning and retention settings.
5. s3_create_bucket requires bucket_name and accepts optional region. It is an account-wide infrastructure change; obtain explicit confirmation and use only an approved region. Bucket names must be globally unique, lowercase, and 3-63 characters.

Sales & CRM

hubspot-contact-manager — Create, update, search, delete, and upsert contacts in HubSpot CRM.

Overview: Create, update, search, delete, and upsert contacts in HubSpot CRM. Supports adding contacts to static lists and managing contact properties.

Allowed agent toolkits: hubspot

Trigger keywords: create contact, add contact, update contact, delete contact, search contact, find contact, upsert contact, hubspot contact, add contact to list, contact lifecycle stage

Instructions:

1. Create or Upsert Contact — Use create_contact when adding a brand new contact with properties (e.g., firstname, lastname, email, company, phone)
2. Search or Find Contact — Use search_hubspot_objects with object_type="contacts" and a search_term to quickly locate contacts by keyword, email, or domain
3. Update or Delete Contact — Use update_contact with the contact ID to modify existing fields
4. List Management — Use add_contact_to_list with list_id and contact_id to add a contact to a static HubSpot list

hubspot-deal-tracker — Track sales pipeline deals in HubSpot CRM.

Overview: Track sales pipeline deals in HubSpot CRM. Create new deals, update deal stages, query active pipeline deals, and manage deal amounts and close dates.

Allowed agent toolkits: hubspot

Trigger keywords: create deal, add deal, update deal, delete deal, search deal, deal stage, pipeline deal, hubspot deal, close deal, sales deal

Instructions:

1. Create Deal — Use create_deal with deal properties (dealname, amount, dealstage, pipeline, closedate)
2. Update Deal Stage or Amount — Use update_deal with the deal ID and updated property values
3. Query or Search Deals — Use query_hubspot_objects with object_type="deals" to filter deals by stage, amount, or close date
4. Delete Deal — Use delete_deal with the deal ID to remove or archive a deal record. Safety Rule: Before using delete_deal, confirm the exact deal name, amount, and ID with the user, and proceed only after receiving explicit confirmation.

hubspot-company-crm — Manage company accounts in HubSpot CRM.

Overview: Manage company accounts in HubSpot CRM. Create company records, update industry/domain details, and query company data for account-based sales and support.

Allowed agent toolkits: hubspot

Trigger keywords: create company, add company, update company, delete company, search company, hubspot company, company domain, account crm

Instructions:

1. Create Company — Use create_company with properties like name, domain, industry, city, state
2. Update Company — Use update_company with the company ID and updated properties
3. Search or Query Companies — Use search_hubspot_objects with object_type="companies" to search by company name or domain
4. Delete Company — Use delete_company with the company ID to remove a company record. Safety Rule: Before using delete_company, confirm the exact company name, domain, and ID with the user, and proceed only after receiving explicit confirmation.

hubspot-record-associator — Associate related CRM records together in HubSpot, such as linking a contact to…

Overview: Associate related CRM records together in HubSpot, such as linking a contact to a deal, a company to a deal, or a contact to a company.

Allowed agent toolkits: hubspot

Trigger keywords: associate contact to deal, associate company to deal, link contact deal, hubspot association, create association, connect contact company, associate records

Instructions:

1. Identify Source and Target Records — Source record: e.g. object_type="contacts", object_id="12345"
2. Create Association — Call create_association with source and target details
3. Supported Association Pairs — Contact ↔ Deal

hubspot-pipeline-navigator — Inspect HubSpot deal/ticket pipelines, stage IDs, owner lookups, and custom…

Overview: Inspect HubSpot deal/ticket pipelines, stage IDs, owner lookups, and custom property schemas to navigate and route CRM workflows accurately.

Allowed agent toolkits: hubspot

Trigger keywords: get pipelines, deal pipelines, pipeline stages, hubspot owner, find owner email, hubspot properties, inspect hubspot schema, custom api request

Instructions:

1. Inspect Pipelines and Stage IDs — Use get_pipelines to list deal or ticket pipelines
2. Find HubSpot Owners — Use search_owner_by_email to lookup a user's HubSpot Owner ID by email address
3. Inspect Property Schemas — Use get_hubspot_properties for an object type (contacts, deals, companies, tickets) to retrieve available property names and data types
4. Custom API Requests — Use hubspot_custom_api_request to execute direct REST HTTP requests for endpoints not covered by standard tools

AITable

aitable-data-discovery — Discover AITable spaces, datasheets, views, and field definitions before reading…

Overview: Discover AITable spaces, datasheets, views, and field definitions before reading or changing records.

Allowed agent toolkits: aitable

Trigger keywords: list aitable spaces, find a datasheet, show aitable fields, show aitable views, explore aitable

Instructions:

1. Run aitable_list_spaces when the space ID is unknown. It has no input parameters.
2. Run aitable_list_datasheets with the chosen space_id to identify the datasheet.
3. Run aitable_get_fields with datasheet_id before creating or updating records. Use the returned field names as the keys in record field objects.
4. Run aitable_get_views with datasheet_id when the user wants a view-scoped search; use the selected view's ID as view_id.

aitable-record-lookup — Find and retrieve records from an AITable datasheet with optional view, filter,…

Overview: Find and retrieve records from an AITable datasheet with optional view, filter, paging, and sorting controls.

Allowed agent toolkits: aitable

Trigger keywords: find aitable records, search aitable, get aitable record, list datasheet records, query aitable

Instructions:

1. Require datasheet_id. Discover it with the AITable Data Discovery skill when necessary.
2. Optionally set view_id to limit results to one AITable view.
3. Optionally set filter_by_formula only when the user supplies or confirms the AITable filter condition.
4. Use page_size to control the number of records fetched per page. Its default is 100 and the catalog limits it to 1000.
5. Use max_records to cap the total returned records, particularly for broad searches.
6. Supply sort only as a JSON string when a sorting configuration is required.

aitable-record-management — Create, update, and delete AITable datasheet records, including deliberate…

Overview: Create, update, and delete AITable datasheet records, including deliberate batch operations.

Allowed agent toolkits: aitable

Trigger keywords: add to aitable, create aitable record, update aitable record, delete aitable record, bulk update aitable

Instructions:

1. Confirm datasheet_id and inspect fields with aitable_get_fields if field names or value formats are uncertain.
2. To create one record, use aitable_create_records with datasheet_id and records, where records is an array of objects each containing a fields object. Example: [{"fields":{"Name":"Ada","Status":"New"}}].
3. To update one record, use aitable_update_records with datasheet_id and a records array. Each array item must include recordId and fields. Example: [{"recordId":"rec123","fields":{"Status":"Done"}}].
4. To delete records, use aitable_delete_records with datasheet_id and record_ids, an array of record IDs.
5. For any write affecting multiple records, show the targeted record IDs and proposed field changes first, then obtain explicit user confirmation.
6. Report completed and failed writes separately. Never claim success from a request payload alone.

aitable-workflow-automation — Configure AITable record-based workflow actions and use the custom Fusion API…

Overview: Configure AITable record-based workflow actions and use the custom Fusion API action only when a catalog action is insufficient.

Allowed agent toolkits: aitable

Trigger keywords: aitable workflow, automate aitable, aitable api call, aitable fusion api

Instructions:

1. aitable_list_spaces, aitable_list_datasheets, aitable_get_fields, and aitable_get_views for discovery.
2. aitable_get_records for record retrieval.
3. aitable_create_records, aitable_update_records, and aitable_delete_records for record changes.

Content Creation

html-website-package — Build a complete, downloadable static website as an HTML/CSS/JS zip.

Overview: Build a complete, downloadable static website as an HTML/CSS/JS zip. Use when the user asks for a website, landing page, HTML site, or a site they can open locally in a browser.

Allowed agent toolkits: python

Trigger keywords: website, landing page, html site, html zip, static website, build a website, create a website, make a website, prepare a website, prepare me a website

Instructions:

1. Infer the topic from the conversation. Ask only if the subject is missing.
2. Choose a distinctive palette, type pairing, and mood that fit the topic. Do not ship a generic Bootstrap or stock-template look.
3. Plan sections before writing files: hero, 3–6 curated highlights, one supporting section (itinerary, tips, FAQ, or process — whichever fits), a clear CTA, and a footer.
4. index.html — semantic HTML5 (header, nav, main, section, footer)
5. styles.css — custom CSS with variables, plus @media breakpoints around 900px and 600px
6. script.js — mobile nav toggle and IntersectionObserver reveal animations
7. README.md — how to open (open index.html), file list, and any network needs (fonts or remote images)
8. Keep HTML, CSS, and JS in separate files. Do not inline a single giant HTML blob.