Autopilot Templates
Autopilot templates are pre-built, runnable AI workers. Each one decomposes a task into roles, runs every step through the gateway governance chain, and produces a shareable HTML report. You don’t wire up agents yourself — you pick a template, give it a task, and get a report back.
There are 22 templates today. They’re registered at startup from
SKILL.md contracts (see Authoring a template) and are
available from the dashboard command center, the API, and Slack/Teams triggers.
See the output first: View a live sample report ↗
The catalog
Engineering
| Template | ID | What it does |
|---|---|---|
| Dev Team | dev_team | Full-stack dev team across multiple coding engines (Claude Code, Aider, Codex, Gemini CLI, Goose, OpenCode, Copilot). Decomposes a coding task into parallel containers, writes code, opens PRs, runs 3-persona review. |
| Code Review | code_review | Reviews a single PR end-to-end — type safety, naming, test coverage, edge cases — and posts inline comments plus an HTML summary. Read-only; never pushes commits. |
| Repo Manager | repo_manager | Multi-agent deploy pipeline: CI monitoring, HITL-gated deployment, code review, and post-deploy health checks. |
| Performance Audit | perf_audit | Profiles an endpoint or service over a recent window, finds the slowest sub-spans, costliest LLM calls, and noisiest DB queries, and ranks fixes by expected impact. |
| AutoResearch | autoresearch | The “Karpathy Loop” — an autonomous modify → test → evaluate → commit/reset loop against one measurable metric, with a results.tsv lab notebook across iterations. |
| Doc Drift | doc_drift | Keeps docs in sync with merged PRs. Fires every N merges, generates a single docs PR, and advances a last-processed-SHA pointer so no PR is doc-drifted twice. |
Security & compliance
| Template | ID | What it does |
|---|---|---|
| Security Audit | security_audit | Scans a codebase for vulnerabilities, OWASP Top 10 issues, dependency CVEs, and secret/PII leaks. Produces a severity-rated report with optional fix PRs. |
| Compliance Report | compliance_report | Builds a quarterly SOC2 / HIPAA / ISO27001 evidence pack from connectors into a board-ready PDF + evidence index. Output is always a draft for human sign-off. |
| Vendor Due Diligence | vendor_dd | Pulls a vendor’s public SOC2 / DPA / sub-processor list, compares to your requirements, flags gaps, and drafts a risk questionnaire. Never auto-approves a vendor. |
| Legal Review | legal_review | First-pass contract review — flags deviations from a standard MSA / DPA / NDA, suggests redlines, and produces a side-by-side risk matrix. Always for counsel sign-off. |
Finance, data & operations
| Template | ID | What it does |
|---|---|---|
| CFO Fleet | cfo | Parallel analyst agents covering revenue, cost, compliance, risk, reporting, and data engineering. For budget variance, financial reporting, audit readiness, and KPI investigation. |
| Data Analyst | data_analyst | Autonomous extract → clean → analyze → visualize → report over CSV / Excel / JSON / Parquet (pandas / polars / duckdb). Produces a markdown analysis report. |
| Product Research | product_research | Clusters support tickets, interviews, NPS, and feature requests into ranked themes with quotes and follow-ups. Cross-references prior reports for trend continuity. |
| Customer Support | customer_support | Ticket handling, FAQ responses, escalation triage, refund/billing routing, and outreach drafting — with PII scanning and quality review on every response. |
Sales, marketing & growth
| Template | ID | What it does |
|---|---|---|
| Sales Research | sales_research | B2B lead research — company analysis, decision-maker mapping, ICP analysis, competitive intel, and personalized cold-outreach drafting with web search. |
| SEO Research | seo_research | Plans a new docs/blog page: keyword volumes, 5+ ranking competitors, structure analysis, and a gap-targeting outline. Output is an HTML report with keyword tables. |
| PR Writer | pr_writer | Drafts launch content in parallel — blog post, tweets, LinkedIn, and changelog — from one launch brief, each polished against brand voice and length. |
| Hype Agent | hype_agent | Platform promotion via community engagement: researches trends, generates per-platform content, and posts through HITL-gated browser automation with anti-spam guards. |
| Recruiter Outreach | recruiter | Sources candidates from GitHub / LinkedIn / OSS contributor lists and drafts 3 personalized outreach variants each. Never sends autonomously. |
Research & web automation
| Template | ID | What it does |
|---|---|---|
| News Digest | news_digest | Self-improving AI news curator across Hacker News, Reddit, RSS, arXiv, Bluesky, GitHub trending, and Lobste.rs. Improves with reader feedback and automated quality judging. |
| Web Agent | web_agent | Live web research with browser streaming — visits multiple sources via Playwright over VNC (with human takeover) and produces a structured research report. |
| Webwright Agent (preview) | webwright_agent | Completes web tasks by writing rerunnable Playwright scripts with screenshots, logs, and self-verification — form-fill, filtering, shopping/search, booking lookup, pricing. |
Running a template
Every template shares the same lifecycle: list → estimate → run → poll → report.
# 1. List available templates
curl https://api.curate-me.ai/api/v1/autopilot/templates \
-H "Authorization: Bearer $CM_API_KEY"
# 2. (optional) Estimate cost before committing
curl -X POST https://api.curate-me.ai/api/v1/autopilot/estimate/news_digest \
-H "Authorization: Bearer $CM_API_KEY" -H "Content-Type: application/json" \
-d '{"topics": ["ai agents", "llm governance"]}'
# 3. Trigger a run — returns a task_id
curl -X POST https://api.curate-me.ai/api/v1/autopilot/run/news_digest \
-H "Authorization: Bearer $CM_API_KEY" -H "Content-Type: application/json" \
-d '{"topics": ["ai agents", "llm governance"]}'
# 4. Poll status (or stream events at /tasks/{task_id}/events via SSE)
curl https://api.curate-me.ai/api/v1/autopilot/tasks/$TASK_ID \
-H "Authorization: Bearer $CM_API_KEY"
# 5. Read the report — report_id is always report_{task_id}
curl https://api.curate-me.ai/api/v1/autopilot/reports/report_$TASK_IDOther lifecycle endpoints: GET /templates/{id}/readiness (check required
integrations before running), POST /tasks/{id}/pause · /resume · /cancel,
and POST /tasks/{id}/approve-plan · /reject-plan for human-in-the-loop gates.
From the dashboard, the same templates run from the Autopilot command center; from Slack/Teams, via slash command or bot mention (see Slack Integration).
Authoring a template
Templates are defined by a SKILL.md contract — frontmatter declaring name,
description, input/output schema, required integrations and permissions, cost
budget, and a sample. To add one:
- Create
services/backend/src/services/autopilot/templates/skills/<your_skill>/SKILL.mdwith the contract frontmatter. - Add
services/backend/src/services/autopilot/templates/<your_skill>.pyimplementing the entrypoint and callingregister_template(). - Import it from
services/backend/src/services/autopilot/__init__.pyso it lands in the registry at startup. - Add an execution test under
tests/services/autopilot/.
The full contract spec — every required and optional field, registry validation, and failure modes — lives in the Autopilot Template Contract .
Next steps
- Reports — the anatomy of the HTML report every template emits, and how to embed or share it
- Teammates & Roles — templates vs. SDK-managed agents, and how roles map to workers
- Gateway Governance Chain — the policy chain every template step routes through