Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add skills/tokencanopy/e2a/autopilotnpx skills add tokencanopy/e2a --skill autopilotgit clone --depth 1 https://github.com/tokencanopy/e2aWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00074 | $0.01947 |
| Opus 5 | $0.00037 | $0.00974 |
| Sonnet 5 | $0.00015 | $0.00389 |
| Haiku 4.5 | $0.00007 | $0.00195 |
Grade A, and why
autopilot scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 204 lines — stays where its author put it; the contents beside it link to each section on GitHub.
e2a Autopilot
Autopilot turns one existing e2a inbox into an always-on local agent trigger. It uses the current e2a protection and CLI surfaces; it does not require or propose server, API, database, SDK, MCP, or core-CLI changes.
Use tether instead when the owner wants to steer an already-running interactive
session by email. Use a cloud webhook instead when a stable public service should
own the workflow. Autopilot is the local-machine daemon: e2a listen connects
outbound, so the machine does not need a public endpoint.
Required conversational behavior
Conduct onboarding as a conversation. Ask one logical question at a time, reflect material answers, explain a tradeoff before asking for a risky opt-out, and skip facts the user already supplied. Do not paste the entire questionnaire.
The interview must establish:
- The task outcome. Offer the built-in customer-support starter or a custom task.
- For support: scope, exclusions, approved knowledge, escalation conditions, tone/signature, response expectations, and refund/billing/security/legal limits.
- An existing e2a inbox and its human owner.
- One inbound authorization mode: exact sender addresses or verified sender domains. The current protection gate cannot mix both modes on one inbox.
- Non-matching inbound mail goes to e2a human review. Do not offer a silent bypass or “any authenticated internet sender”; the current surface cannot express the latter safely.
- Outbound review, default on. Turning it off requires the warning and the exact acknowledgement requested by the interview.
- Owner CC on every reply, default on. Turning it off requires a warned acknowledgement.
- Prompt-injection screening, recommended and default on. Turning it off requires a warned acknowledgement.
- Claude Code, Codex, Hermes Agent, or a custom executable; its absolute path, workspace, and isolation mode. (OpenClaw is unavailable in this release: its invocation flags are unverified.) Hermes and custom executables require an acknowledged external isolation boundary because Autopilot cannot verify one for them. A built-in container runner is not implemented in this release; use an explicitly reviewed custom wrapper.
- launchd, systemd, or foreground operation.
What ships with it
35 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
- autopilot.mjs 12 KB runs code
- autopilot.sh 260 B runs code
- config.mjs 3.4 KB runs code
- daemon.mjs 12 KB runs code
- gateway.mjs 11 KB runs code
- installer.mjs 13 KB runs code
- interview.mjs 13 KB runs code
- job-tool.mjs 3.6 KB runs code
- lock.mjs 2.9 KB runs code
- mail-client.mjs 8.5 KB runs code
- operator.mjs 4.7 KB runs code
- policy.mjs 16 KB runs code
- runner.mjs 3.2 KB runs code
- runtime.mjs 11 KB runs code
- service.mjs 3.9 KB runs code
- setup.mjs 8.7 KB runs code
- spool.mjs 7.6 KB runs code
- supervisor.mjs 12 KB runs code
- test/autopilot-cli.test.mjs 2.7 KB runs code
- test/clean-room.test.mjs 6.3 KB runs code
- test/config.test.mjs 5.7 KB runs code
- test/daemon.test.mjs 14 KB runs code
- test/gateway.test.mjs 11 KB runs code
- test/installer.test.mjs 7.8 KB runs code
- test/interview.test.mjs 7.0 KB runs code
- test/job-tool.test.mjs 3.7 KB runs code
- test/lock.test.mjs 3.0 KB runs code
- test/mail-client.test.mjs 5.6 KB runs code
- test/operator.test.mjs 4.6 KB runs code
- test/policy.test.mjs 8.2 KB runs code
- test/runtime.test.mjs 11 KB runs code
- test/service.test.mjs 2.7 KB runs code
- test/setup.test.mjs 5.7 KB runs code
- test/spool.test.mjs 5.0 KB runs code
- test/supervisor.test.mjs 18 KB runs code
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 2d ago First seen · 204 lines · 74 tokens per session scan A 2d85db8636db
autopilot is a skill published in the GitHub repository tokencanopy/e2a (184 stars, last pushed yesterday), licensed Apache-2.0. It adds 74 tokens to every session and 1,947 once invoked, about $0.0004 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
commune-email
I give you a real email inbox and let you send, receive, search, and manage email threads programmatically via the Commune API. Every inbox I create is RFC 5322 compliant — replies thread correctly in any email client, including Gmail, Apple Mail, and Outlook.
commune-agent-network
Requires COMMUNEAPIKEY environment variable. Get one at commune.email.
commune-sms
I give you a real phone number and let you send and receive SMS messages programmatically via the Commune API. You can send texts to any number, read conversation threads, and list all your SMS conversations — all from within your OpenClaw agent.
skills
Email API for AI agents. Create inboxes, send/receive email, manage threads and webhooks.
shipmail
Operate Shipmail custom-domain business email, agent inboxes, calendars, booking pages, newsletters, webhooks, and related account resources through the Shipmail MCP server. Use when the user asks to read or triage a Shipmail inbox, draft or send email, reply to a message or thread, manage a mailbox or domain, check…
clawaimail
Email infrastructure for AI agents — create inboxes, send and receive real emails, search messages, manage threads. Alternative to AgentMail with MCP server, REST API, webhooks, and WebSocket streaming.