ReachAgents Agent Kit
Give your agent the tools to find leads, prepare campaigns and report on results. Works with Codex, Claude Code, Hermes and OpenClaw.
Download
Install the kit
Requires Python 3.9+ and a signed-in local agent runtime. Choose your runtime, then run these commands in your terminal. Each installed skill includes its own CLI and API schemas.
curl -L https://reachagents.ai/downloads/reachagents-agent-kit.zip -o reachagents-agent-kit.zip
unzip -o reachagents-agent-kit.zip -d reachagents-agent-kit
cd reachagents-agent-kit
./install.sh --runtime codex --env productionInstalls personal Codex skills. Add --project-dir PATH for a project installation.
The default production install points agents at https://api.reachagents.ai and https://app.reachagents.ai.
For non-production testing, open Settings → API Keys on the non-production app and copy the generated install command from there.
First use
Configure your connection
From the extracted reachagents-agent-kit directory, run the CLI below and enter your scoped API key at its hidden prompt. The shared credential file stays on your computer with private permissions; it is not copied into chat.
reachagents_cli="./bin/reachagents-api"
python3 "$reachagents_cli" configure
python3 "$reachagents_cli" doctor
python3 "$reachagents_cli" GET /v1/whoamiFor an installed CLI, Codex users can use ~/.agents/skills/reachagents-operator-account-setup/scripts/reachagents-api; Claude Code users can use ~/.claude/skills/reachagents-operator-account-setup/scripts/reachagents-api. Both runtimes read ~/.config/reachagents/env; environment overrides take precedence. A desktop session can use this file without depending on exports from another terminal.
Start a new session in your agent runtime and ask: “Use the ReachAgents operator skills. Check my connection, list accessible workspaces, and explain what blocks outreach. Do not create, run or send anything.”
For project-local skills, add --project-dir /absolute/path/to/project to the install command. Choose one runtime with --runtime codex or --runtime claude. Hermes uses --runtime hermes; OpenClaw uses --runtime openclaw --openclaw-dir PATH. Claude support here means Claude Code with local shell access.
The agent should discover its installed skill, use the CLI beside it, select the requested workspace and return a report with app links. Before a write, approve the exact action. The CLI blocks unconfirmed writes before making an API request.
What can I ask?
Put your agent to work
Find leads, prepare campaigns, review your Inbox and understand your results. Pick a task below for a ready-to-use instruction.
25 prompts · 6 categories · All four runtimes
Copy a prompt, replace the details in [brackets] and paste it into your agent. Start with an installed kit and a configured API key. Drafts keep discovery or sending off. Review prompts ask for your approval.
Lead Agents
Find people showing relevant interests, needs and buying signals on X, YouTube, TikTok or LinkedIn, where available.
A saved agent you can review before discovery starts.
Use the installed ReachAgents operator skills in workspace "[workspace name]". If the workspace is ambiguous, ask me to choose. Create a Lead Agent called [agent name] to find [ideal customer, e.g. founders of small SaaS companies] who show [signal, e.g. asking for an AI customer-support tool] on [X, YouTube, TikTok or LinkedIn]. Request 25 leads per run. Keep automatic discovery and scheduling off; do not start a run. Show me its criteria and the link to review it. Flag any source unavailable in my workspace.
A proposed change to its audience, signals, tone or cadence.
Use the installed ReachAgents operator skills in workspace "[workspace name]". If the workspace is ambiguous, ask me to choose. Review [agent name]. I want to target [audience] based on [signals], use [communication style] and discover people [desired cadence]. Compare the current settings with your proposed changes. Ask me to approve before saving anything, changing the schedule or starting discovery.
A reviewed discovery plan, then trackable runs after approval.
Use the installed ReachAgents operator skills in workspace "[workspace name]". If the workspace is ambiguous, ask me to choose. Prepare one discovery run for [agent name or list of agents], using their current lead limits. Check readiness, available balance and any applicable limits. Explain what will run and ask me to approve before queueing it. After approval, return the run IDs and status links. Do not create campaigns or send messages.
Control future discovery and handle any active run separately.
Use the installed ReachAgents operator skills in workspace "[workspace name]". If the workspace is ambiguous, ask me to choose. Show the status and any active run for [agent name]. I want to [pause future discovery / remove this agent]. Explain the effect and ask me to approve the exact change. If a run is active, show its status and ask separately before canceling it. Only propose deletion when the agent is not running. Keep its contacts.
Campaigns
Prepare outreach through email, X DM, LinkedIn, Telegram, WhatsApp or Instagram. Available senders, rollout gates and channel limits still apply.
An outreach draft with an audience and message sequence, ready for review.
Use the installed ReachAgents operator skills in workspace "[workspace name]". If the workspace is ambiguous, ask me to choose. Create a campaign draft on [channel, e.g. email] for [contact group or audience]. Our offer is [offer] and the goal is [goal]. Use a [tone] style and [number] message steps where the channel supports them. Check the available sender accounts and ask me to choose if needed. Save only a draft, leave sending disabled and show the audience, messages and campaign link. Do not launch it.
An invite-only draft with no note or follow-up message.
Use the installed ReachAgents operator skills in workspace "[workspace name]". If the workspace is ambiguous, ask me to choose. Create a LinkedIn campaign draft for [contact group] using [LinkedIn sender account]. It should send connection requests only, with no invitation note and no follow-up DM. Check that the contacts have LinkedIn identities. Save the draft and show its link. Do not launch or send any invitations.
Audience eligibility, sample messages and blockers before you send.
Use the installed ReachAgents operator skills in workspace "[workspace name]". If the workspace is ambiguous, ask me to choose. Preview [campaign name]. Show the eligible audience, a sample of personalized messages, sender readiness, available capacity and any blockers or unresolved placeholders. Explain what I should fix and link to the campaign. Do not run the campaign or send anything.
Review message, audience or individual target changes before saving.
Use the installed ReachAgents operator skills in workspace "[workspace name]". If the workspace is ambiguous, ask me to choose. Review [campaign name] and propose these changes: [new messages, audience settings or target changes]. Show the current configuration and affected targets first. Ask me to approve the exact changes before saving them. Do not launch the campaign, send messages or change the underlying contact records.
A final review and explicit approval before delivery is queued.
Use the installed ReachAgents operator skills in workspace "[workspace name]". If the workspace is ambiguous, ask me to choose. Prepare to launch [campaign name or list of campaigns]. Check the latest preview, eligible targets, sender accounts, schedules, limits and enabled channel gates. Show the exact messages and audience, then ask me to approve the launch. After approval, queue only those campaigns and return the run IDs. Report queued work separately from confirmed sends.
Check draft eligibility and confirm the exact campaign before deletion.
Use the installed ReachAgents operator skills in workspace "[workspace name]". If the workspace is ambiguous, ask me to choose. Check whether [campaign name] is an inactive draft that can be deleted. Verify its status and delivery activity, show the exact campaign and ask me to confirm deletion. Do not delete an active campaign or one with delivery activity. Keep the contacts and contact groups.
Reports & health
See what was discovered, what is queued, what was sent and what needs attention.
A report with source-backed counts and links to the next action.
Use the installed ReachAgents operator skills in workspace "[workspace name]". If the workspace is ambiguous, ask me to choose. Review my campaigns and report their current performance. Separate eligible targets, queued work, confirmed sends, replies, failures and unconfirmed results. Use campaign summaries, targets and available email events; tell me which metrics or time ranges cannot be verified. Rank the blockers and include links to the relevant campaigns. Do not change or send anything.
Qualified people, matching evidence and links to their sources.
Use the installed ReachAgents operator skills in workspace "[workspace name]". If the workspace is ambiguous, ask me to choose. Show the people found by [agent name]. Explain how each matches our criteria, include profile links and available source evidence, and separate verified fields from missing information. Check the agent and workspace contact lists before concluding that a run found no leads. Return a shortlist for review. Do not import, group or message anyone.
Specific setup, sender, capacity or run problems with next steps.
Use the installed ReachAgents operator skills in workspace "[workspace name]". If the workspace is ambiguous, ask me to choose. Check workspace readiness, sender accounts, campaigns and available run diagnostics. Explain why outreach is not progressing and rank the fixes. Distinguish account pauses, exhausted capacity, synchronization issues and confirmed send failures. Include the exact app links to open next. Do not reconnect accounts, edit settings, start runs or send anything.
The actual run state and logs, with approval before cancellation.
Use the installed ReachAgents operator skills in workspace "[workspace name]". If the workspace is ambiguous, ask me to choose. Check [run ID] and summarize its status and logs. Explain whether it is queued, working, finished or failed, and what the evidence proves. If it is stuck or I want to stop it, explain the effect and ask me to approve before canceling. Do not retry or start another run automatically.
Single or batch video data with cache, cost and quota context.
Use the installed ReachAgents operator skills in workspace "[workspace name]". If the workspace is ambiguous, ask me to choose. Prepare a metrics lookup for [YouTube video URL or up to 10 URLs]. Check that my key has data access and explain the possible cost and daily crawl-quota usage. Ask me to approve before making the lookup, since missing or stale cached data can trigger a crawl. After approval, report available metrics, cache freshness and any per-video errors without inventing missing data.
Contacts & groups
Add people, organize an audience and maintain contact details before outreach.
A validated single contact or import preview before anything is saved.
Use the installed ReachAgents operator skills in workspace "[workspace name]". If the workspace is ambiguous, ask me to choose. Prepare to add [contact details or a local CSV file path] to this workspace. Validate the available platform identities and show missing fields, possible existing matches and the number of contacts to create or update. Preview the changes and ask me to approve before saving or bulk importing. Do not start discovery, launch a campaign or send messages.
A reviewable audience with explicit additions and removals.
Use the installed ReachAgents operator skills in workspace "[workspace name]". If the workspace is ambiguous, ask me to choose. Find contacts matching [audience criteria] and propose a group called [group name]. Show the selected contacts, any existing group and the exact members to add or remove. Ask me to approve before creating the group or changing membership. Removing someone from the group must keep their contact record. Do not launch outreach.
A before-and-after review of the selected contact record.
Use the installed ReachAgents operator skills in workspace "[workspace name]". If the workspace is ambiguous, ask me to choose. Find [contact name or ID] and propose these updates: [display name, status, tags or group]. Show the matching record and the before-and-after values. Ask me to approve before saving changes. Preserve provider identities and unrelated fields, and do not send a message.
Inbox & replies
Read conversations across six channels, prepare replies and review automation. Direct text replies support email, LinkedIn, Telegram, WhatsApp and Instagram; X conversation sends are unavailable.
A prioritized conversation summary with suggested replies in chat.
Use the installed ReachAgents operator skills in workspace "[workspace name]". If the workspace is ambiguous, ask me to choose. Review my recent Inbox conversations on [channel or all channels]. Prioritize interested leads, unanswered questions and conversations needing my attention. Read the relevant messages, summarize the context and draft suggested replies here in chat. Link each suggestion to its conversation. Do not save drafts, send replies or change automation settings.
The exact conversation, sender and text for your approval.
Use the installed ReachAgents operator skills in workspace "[workspace name]". If the workspace is ambiguous, ask me to choose. Prepare a reply to [conversation or contact] on [email, LinkedIn, Telegram, WhatsApp or Instagram]. Read the context and use this intent: [what to say]. Show the exact text, recipient, sender and any channel limits. Ask whether I approve saving it as a draft or sending it; do neither until I choose. Do not add attachments or post to groups. If a send result is unconfirmed, inspect the conversation before retrying.
A policy proposal with clear sending behavior and limits.
Use the installed ReachAgents operator skills in workspace "[workspace name]". If the workspace is ambiguous, ask me to choose. Review my Inbox automation policies for [channel and account]. Propose [monitor only / suggested reply drafts / automatic replies] for [use case]. Explain what the selected mode does, and propose business hours, timezone, confidence thresholds, exclusions and reply limits. Check channel availability and ask me to approve before saving or enabling a policy. Do not run automation or send replies as part of this review.
Candidate replies and a proposed action for each one.
Use the installed ReachAgents operator skills in workspace "[workspace name]". If the workspace is ambiguous, ask me to choose. Inspect my Inbox automation policies and reply candidates. Show each relevant conversation, proposed text and recommended action. Propose any candidate edits and ask for approval before saving them or sending any selected reply. If a fresh automation check is needed, explain whether the current policy can send automatically and get separate approval before running it.
Workspace setup
Check your connection, describe your business and set the goals your agents should follow.
Accessible workspaces, available permissions and setup blockers.
Use the installed ReachAgents operator skills in workspace "[workspace name]". If the workspace is ambiguous, ask me to choose. Check my ReachAgents connection without reading or printing my API key. Verify authentication, list the workspaces I can access and explain which actions my key permits. Check readiness for the workspace I choose and link to any settings I need to complete. Do not create, edit, run or send anything.
A proposed business profile and workspace setup to review.
Use the installed ReachAgents operator skills in workspace "[workspace name]". If the workspace is ambiguous, ask me to choose. Help me set up ReachAgents for [website URL]. Explain the website-analysis action and any applicable usage before asking for approval to analyze it. Then propose our business description, ideal customers, goals and communication style. Let me correct the proposal and ask separately before applying it to my company or workspace. Do not create agents, start discovery or send messages.
A clear change proposal for the company and selected workspace.
Use the installed ReachAgents operator skills in workspace "[workspace name]". If the workspace is ambiguous, ask me to choose. Review our company profile and this workspace. Our updated offer is [offer], ideal customer is [audience], goal is [goal] and preferred tone is [style]. Show the current values and proposed changes, distinguishing company-wide details from workspace settings. Ask me to approve before saving. Do not change existing agents, campaigns or automation policies.
Availability depends on your key permissions, connected accounts and enabled channels. See API coverage for exact fields and permissions. A queued run is not a delivery receipt.
Public API
Current API coverage
Use the same scoped API key for setup, discovery, campaigns and Inbox. The kit includes the full request and response schemas and an operation list beside every installed skill.
Download the OpenAPI contract| Channel | Endpoint | Sender field |
|---|---|---|
| X DM | POST /v1/campaigns/x-dm | x_account_id |
POST /v1/campaigns/email | email_account_ids | |
POST /v1/campaigns/linkedin | linkedin_account_id | |
| Telegram | POST /v1/campaigns/telegram | telegram_account_id |
POST /v1/campaigns/whatsapp | whatsapp_account_id | |
POST /v1/campaigns/instagram | instagram_account_id |
- Setup: company/workspace profiles, communication style, onboarding readiness and website analysis.
- Discovery and contacts: X, YouTube, TikTok and LinkedIn agents, lead limits, contact groups, profile links and source evidence. People Search availability is controlled by the server.
- Campaign operations: sender accounts, audience previews, target states, email events, health summaries and safe deletion of inactive drafts.
- Inbox: six-channel conversation reads, sender and response filters, approved text replies, automation policies and candidate review.
- Run tracking: run status, logs, cancellation and batch requests. A queued run does not prove delivery.
- YouTube data: single or batch video lookups with the optional data:read scope. Fresh crawls can use your balance and the daily crawl quota.
WhatsApp and Instagram sending, campaigns and automation depend on separate server rollout gates, account state and limits. Text replies support email, LinkedIn, Telegram, WhatsApp and Instagram; X conversation sends currently return unsupported_platform. Instagram text is limited to 1,000 characters; other accepted text to 8,000. API replies cannot send attachments or post into Telegram, WhatsApp or Instagram groups.
Campaign summary usage counters cover X, LinkedIn and email. Use target and message evidence for newer channels; zero summary counts do not prove that nothing was sent. If a send result is unconfirmed, inspect the conversation before retrying.
All 62 API operations and scopes
| Method / path | Operation | Required scopes |
|---|---|---|
GET /v1/youtube-videos/{video} | Get YouTube video metrics | data:read |
GET /v1/youtube-videos | Batch YouTube video metrics | data:read |
GET /v1/whoami | Inspect the authenticated API key | Valid API key |
GET /v1/company | Get company details | workspaces:read |
PATCH /v1/company | Update company profile details | workspaces:write |
GET /v1/workspaces | List accessible workspaces | workspaces:read |
GET /v1/workspaces/{workspace_id} | Get one workspace | workspaces:read |
PATCH /v1/workspaces/{workspace_id} | Update workspace details | workspaces:write |
GET /v1/agents | List agents | agents:read |
POST /v1/agents | Create a lead agent and start discovery by default | agents:write, agents:run |
GET /v1/agents/{agent_id} | Get one agent | agents:read |
PATCH /v1/agents/{agent_id} | Update a lead agent | agents:write |
DELETE /v1/agents/{agent_id} | Delete a non-running lead agent | agents:write |
POST /v1/agents/{agent_id}/run | Queue an agent discovery run | agents:run |
GET /v1/contacts | List contacts | contacts:read |
POST /v1/contacts | Create or upsert a contact | contacts:write |
PATCH /v1/contacts | Patch a contact by ID in the request body | contacts:write |
PATCH /v1/contacts/{contact_id} | Patch a contact | contacts:write |
POST /v1/contacts/bulk | Bulk import contacts | contacts:write |
GET /v1/contact-groups | List contact groups | contact_groups:read |
POST /v1/contact-groups | Create a contact group | contact_groups:write |
POST /v1/contact-groups/{group_id}/contacts | Add contacts to a group | contact_groups:write |
DELETE /v1/contact-groups/{group_id}/contacts/{contact_id} | Remove one contact from a group | contact_groups:write |
GET /v1/campaigns | List campaigns | campaigns:read |
POST /v1/campaigns | Create a campaign draft | campaigns:write |
PATCH /v1/campaigns | Patch a campaign by ID in the request body | campaigns:write |
POST /v1/campaigns/linkedin | Create a LinkedIn campaign draft | campaigns:write |
GET /v1/campaigns/{campaign_id} | Get one campaign | campaigns:read |
PATCH /v1/campaigns/{campaign_id} | Patch a campaign draft/configuration | campaigns:write |
DELETE /v1/campaigns/{campaign_id} | Delete an inactive campaign draft | campaigns:write |
POST /v1/campaigns/{campaign_id}/preview | Preview a campaign audience/messages without sending | campaigns:read |
POST /v1/campaigns/{campaign_id}/run | Queue a campaign run | campaigns:run |
POST /v1/runs/batch | Queue multiple agent/campaign runs | agents:run, campaigns:run |
GET /v1/runs/{run_id} | Get run status | runs:read |
GET /v1/runs/{run_id}/logs | Get run logs | runs:read |
POST /v1/runs/{run_id}/cancel | Cancel a queued or running run | runs:cancel |
GET /v1/conversations | List inbox conversations | contacts:read |
GET /v1/conversations/{conversation_id}/messages | List conversation messages | contacts:read |
POST /v1/conversations/{conversation_id}/messages | Send or draft a conversation message | contacts:write |
GET /agents | Legacy alias for GET /v1/agents | Valid API key |
POST /leads | Legacy alias for POST /v1/contacts | Valid API key |
POST /leads/bulk | Legacy alias for POST /v1/contacts/bulk | Valid API key |
GET /v1/inbox-automation/policies | List inbox automation policies | inbox_automation:read |
POST /v1/inbox-automation/policies | Create or update an inbox automation policy | inbox_automation:write |
PUT /v1/inbox-automation/policies | Create or update an inbox automation policy | inbox_automation:write |
GET /v1/inbox-automation/candidates | List inbox automation candidates | inbox_automation:read |
PATCH /v1/inbox-automation/candidates/{candidate_id} | Patch an inbox automation candidate | inbox_automation:write |
POST /v1/inbox-automation/candidates/{candidate_id}/send | Send an inbox automation candidate reply | inbox_automation:run |
POST /v1/inbox-automation/run | Run inbox automation manually | inbox_automation:run |
GET /v1/onboarding/readiness | Diagnose onboarding readiness for a workspace | workspaces:read |
POST /v1/onboarding/analyze-website | Analyze a website into a ReachAgents business profile | workspaces:read |
POST /v1/onboarding/setup | Apply onboarding business profile to company and workspace | workspaces:write |
POST /v1/campaigns/whatsapp | Create a WhatsApp campaign draft | campaigns:write |
POST /v1/campaigns/instagram | Create a Instagram campaign draft | campaigns:write |
POST /v1/campaigns/telegram | Create a Telegram campaign draft | campaigns:write |
GET /v1/sender-accounts | List sender accounts | campaigns:read |
POST /v1/campaigns/x-dm | Create an X DM campaign draft | campaigns:write |
POST /v1/campaigns/email | Create an email campaign draft | campaigns:write |
GET /v1/campaigns/{campaign_id}/targets | List campaign targets | campaigns:read |
GET /v1/campaigns/{campaign_id}/events | List campaign email events | campaigns:read |
GET /v1/campaigns/{campaign_id}/summary | Get campaign health and delivery summary | campaigns:read |
PATCH /v1/campaigns/{campaign_id}/targets/{target_id} | Update a campaign target | campaigns:write |
Creating an agent starts discovery by default and needs agents:run as well as agents:write. Use auto_run: false for a manual draft. Batch scope requirements depend on the submitted items; the OpenAPI contract describes each operation.
Kit 2026.10.27.2
Update an existing installation
Download the latest kit and rerun the installer. It updates the bundled ReachAgents operator skills, references, bridge and CLI. Existing environment values and other skills are preserved. Choose Codex and Claude Code with --runtime codex,claude; add --project-dir PATH for project-local skills. Hermes and OpenClaw remain supported with their runtime flags.
Existing keys keep their current permissions. Add data:read only if you need YouTube data. Start with read scopes for reporting and add write/run scopes for the actions you approve.
reachagents_cli="$HOME/.agents/skills/reachagents-operator-account-setup/scripts/reachagents-api"
python3 "$reachagents_cli" GET /v1/whoami
python3 "$reachagents_cli" GET /v1/workspaces
python3 "$reachagents_cli" GET /v1/sender-accounts --query workspace_id=YOUR_WORKSPACE_ID
python3 "$reachagents_cli" GET /v1/campaigns/CAMPAIGN_ID/summary --query workspace_id=YOUR_WORKSPACE_IDRead-only sender and run requests work without mutation confirmation. Use confirmed=true in the bridge or --confirm in the CLI after approving a mutation. Reuse the same idempotency key when retrying the same approved action.
Credentials
Create a ReachAgents API key
Where to create it
Open ReachAgents Settings → API Keys, create a key, copy it once, and save it locally with the CLI configure command. The key is used as REACHAGENTS_API_KEY.
Full operator scopes
Approval model
Safe operations for customer agents
What agents can do
- Read-only reporting can run after you provide a scoped API key.
- Creating, updating, deleting, running, sending, canceling, or bulk importing always requires explicit approval.
- API keys stay in your local agent runtime. Do not paste them into chat, logs, reports, or task descriptions.
- Customer-facing reports should include the exact ReachAgents page link to open next.
Public API coverage
The bundled CLI covers every operation in the Public API contract. Hermes can also use reachagents_api_request. Mutating and live operations require confirmed=true after the user approves the exact action.
Reports
Use these customer-facing links
When an agent gives a customer report, include the exact ReachAgents page to open next.