My account profile
Log out
Reset password
You're in charge!
ClawNanny is designed to help you discover your needs, define them clearly, and fulfill them through requests and exploration, without self-injury.
Perhaps the best way to get to know how your intelligent agent can serve you is to ask ClawNanny to interview you about your needs, goals, and expectations.
"Desi", the capability designer, will guide you through a series of questions to help you define your needs and goals clearly and precisely.
Desi models good system-design behavior by asking clarifying questions and iterating until the specification of what you want is clear and precise ... as a wise partner would. You can then paste the specification into a request to your agent to help it deliver what you want.- It's good to start by asking your agent for modest deliverables and outcomes.
Often, we find that seeing the tangible changes we want will guide us toward a deepening understanding of what's possible and a more fulfilling experience.
The more skilled you become at communicating your needs with precision, the better you'll understand what your agent can do as a collaborator.
Examples:- Asking for research into what competitors' products and services are, when what you really want is a strategy that will deliver more value (cheaper, better, faster, more enjoyable) to your customers with reference to your own unique values.
- Asking for a change in the intelligent agent's behavior, when what you really want is to be known by them and to feel safe with them so you can play well together.
(It's more powerful to own your wants and vulnerabilities and address them directly with your intelligent agent than to expect them to guess about what you really want.)
- Make clear requests of the agent to deliver what you want, specifically.
- Define the boundaries of what you want and the success metrics that indicate what you'll accept.
- Avoid making assumptions about what the agent knows or doesn't know. Give it the whole context of your present situation and your goals. It's a willing listener, not a mind-reader.
- Iterate on the request until you're satisfied with the result. Give feedback on what worked well for you and where you want to improve. (It learns best from positive feedback.)
- As you learn more about your agent's capabilities and limitations, add new details to the workspace configuration files (SOUL.md, USER.md, PERSONA.md, etc.) or to SKILL.md files that you want your agent to remember in subsequent sessions.
- It's a relationship. Use your interactions to gain self-awareness and clarity about your needs, capabilities, and goals.
- Keep expanding what you know about the limits of your agent's prowess. Language models are getting more intelligent month by month. It's worthwhile to keep up with their expanding capabilities and evolving limitations. An agent is like a friend who's maturing faster than you. Its present intelligence is the lowest it will ever be.
- Be willing to ask for help whenever you need it. The ClawNanny staff is here, eager to help you make the most of your experience with your agent. This is fun for us, too!
- Keep going.
Sense-making
This tab helps you make sense of it all.
| Topic | Tool or Skill |
| Recurring console log issues | Diagnostic Run History |
| Recurring console log issues | diagnostic_run_analysis_summary |
openclaw doctor --fix --yes --non-interactive
Mission Headquarters or Central Command — Operations dashboard (to come ...)
This is where you'll monitor and manage the state of your business and/or personal goals.See this video for an overview of Zach Babiarz's Mission Control prompts.
See many mission-control dashboard examples.
Manage context — what the agent knows during the current session
Knowledge baskets — portable, agent-ready Markdown documents to carry along as context for your agent
Workspaces — folders of content assets to use as context for your agent
Context state
ClawNanny shows context-window usage and compaction controls for whichever execution runtime is active. Today the unit is tokens (OpenClaw's context window). Other runtimes may report context in items or turns; ClawNanny will adapt the readout to match.
Each session has its own knowledge basket (context). Sessions don't share knowledge baskets with each other unless you or ClawNanny intentionally make it so.
A topic is a subject or artifact that's addressed within one or more sessions.
So a session can be an odd collection of unrelated conversational segments.
A thread is a sequence of session segments that are semantically related to each other around a common topic.
Threads are the most sensible way to navigate the work you and your agent have done together, but "thread" is a foreign concept to agents.
Below we've intuited a threaded view of session segments that shows you the threads of work you and your agent have done together.
You can view a list of your topic threads or unrelated sessions, in which you can search for sessions related to a topic (as a keyword search).
Preserve session knowledge for searching
and to correlate topics within sessions
- Capture each agent session as a Markdown file. This enables you to search within all sessions for a specific keyword or topic.
- Correlate sessions with ClawNanny's thread-detection engine to see topically related threads within the tangle of sessions.
The session-capture process you can trigger below for Claude Code and/or OpenClaw preserves the full contents of every agent session — all conversational turns, tool calls, tool results, no limits —
into timestamped snapshot Markdown files in
~/agent-memory-compiler/session_captures/{agent-name}/.
Open session windows will remain open.
Captures are non-destructive snapshots in Markdown files.
Redundant captures for sessions that remain unchanged since the last capture will be skipped.
Capturing a continued session (having more turns since the last capture) will replace its incomplete captured-contents file.
Key elements in each conversational session are stored in the correlation database to enable semantic (meaning-based) correlation analysis that helps you see topic-related threads among the tangle of sessions.
| Project filter |
Limit capture to one selected project or leave as "All projects".
|
| Session filter |
Limit to one session UUID, or leave blank for all.
|
- Clicking the [ Capture ] button will capture the full contents of all then-open sessions for the agent you select.
- Open agent session windows will remain open.
- Captures are non-destructive snapshots.
- Redundant captures for sessions that remain unchanged since the last capture will be skipped.
- Capturing a continued session (having more turns since the last capture) will replace its incomplete captured-contents file.
After sessions are captured, the algorithm correlates their key elements with ClawNanny's thread-detection engine so you can see topically related threads among the tangle of sessions.
then re-ingests every captured session .md file on disk.
Captured files are preserved, including user-entered session nicknames.
Any session whose .md file has been deleted will not reappear in the correlation database.
Communication channels
ClawNanny routes agent output through two layers of channels: 1) ClawNanny delivery, owned by the ClawNanny layer and identical regardless of which agent-runtime is active — e-mail today; in-app notifications, mobile push and webhooks after July 1, 2026, and 2) backend agent-runtime delivery, channels the active agent-runtime exposes natively — today, OpenClaw's Telegram / Discord / Slack sidecars. After July 1, 2026, agent-runtime-side channels will move into the ClawNanny layer so they keep working when you switch backend agent-runtimes.
Channels configured inside the active agent-runtime. These are agent-runtime-specific surfaces. Their availability depends on which agent-runtime is active.
E-mail messages
Working with e-mail messages
- Forwarding an incoming message to an outside address
- Ask your agent: "Forward the last e-mail you received to [address]." The agent uses its e-mail-sending skill to relay the message. Alternatively, open console.agentmail.to and configure auto-forward rules on the inbox.
- Causing your agent to send an e-mail
- In the Requests tab, instruct the agent: "Send an e-mail to [address] with subject [subject] saying [body]." The agent will compose and send via the ClawNanny E-mail Sender. You can also trigger sending from a cron job by including send instructions in the cron prompt.
- Marking a message as completed
- Click Mark complete on any inbox message. This adds the
completedlabel in AgentMail and moves it to the Completed section. It does not delete the message. - Reviewing what the agent did with an e-mail
- Switch to the Requests tab and check the session that the agent used to process
the e-mail message (session key:
agent:main:mainby default). The agent's full response to the e-mail message is visible there. - Configuring the target session or persona for incoming e-mail messages
- On the Config tab, set
AGENTMAIL_SESSION_KEYto route incoming e-mail messages to a different session (Example:agent:main:email-handler), andAGENTMAIL_INBOX_IDto change which of your inboxes is monitored. These are the same values you entered when you connected your AgentMail account in the setup wizard. - Managing inboxes and rules at AgentMail
- console.agentmail.to provides full inbox management: create inboxes, set auto-reply rules, view delivery logs, and configure webhooks as an alternative to the WebSocket listener.
Cron job specification
| Job Name | Schedule | Last Run | Next Run | Status | Actions |
|---|---|---|---|---|---|
| Click [ Display all scheduled tasks ] to load cron jobs. | |||||
Secret authorization tokens and API keys
| Provider | Last 4 | Status | Last test | Actions |
|---|---|---|---|---|
| Loading ... | ||||
Cost by model provider
Token Costs by Day
Skills in the ClawNanny ecosystem that are installed and are known to OpenClaw
Visit ClawHub.ai to install more skills. Be cautious when installing skills from any external source!FIRST, scan them for viruses and malware using the Cisco AI Defense Skill Scanner or one of the scanners listed on ClawHub.
MANAGED means the SKILL.md file lives in
~\.openclaw\skills\{skill-name}\SKILL.mdWORKSPACE means the SKILL.md file lives in
~\.openclaw\workspace\skills\{skill-name}\SKILL.mdconfig means the skill is in the local configuration file (
~\.openclaw/skills.json), but it's not in a SKILL.md file in either the MANAGED folder or the WORKSPACE folder, so click [ Install ] and [ Update ].Mediated application-tool sources
An application tool is a service provided by an external source that gives your agent the ability to take actions in the broader digital world — send an e-mail, create a calendar event, post a message, update a customer record, issue a refund, etc. ClawNanny connects to these services through a hub called Zapier (ZAY-pee-er), which can interact with thousands of apps on your behalf.Every action available from a source flows through ClawNanny's governed pipeline — that is, it's checked against the per-agent action allowlist, it's classified as to what type of action it is (read, write, pay), and approval is requested when the policy specifies approval is needed for that type of action before it is executed — that is, before it ever reaches the actionable service endpoint. (An "endpoint" is a URL where the application tool can be called into action.) An agent can see nothing until you add an allowlist pattern for it (or for all agents) below. ClawNanny's governed pipeline captures every action in a log for later review and audit.
To connect to any of the many tools at Zapier, you must first generate an "endpoint URL" for it at mcp.zapier.com. Here's a detailed,step-by-step guide to do that.
The token is stored in your local credentials file (
~/.clawnanny/.env) as a secret key and surfaced for review on the Tokens & API keys tab
— never written to the database.Manage local file-system authorizations
Experience shows that it's important to set boundaries and manage authorizations for agent operations on local file-system assets.For each file-system path (folder-level or file-level), you can specify the following authorizations or restrictions:
- CREATE — Allow the agent to create new files and folders
- READ — Allow the agent to read the contents of files and folders
- UPDATE — Allow the agent to modify the contents of files and folders
- DELETE — Allow the agent to delete files and folders — Denied by default
- EXECUTE — Allow the agent to execute prompts (such as skills, plug-ins, tools, and workflows), scripts (executable runtime objects), and other executable assets
Security
ClawNanny separates policy from runtime. The ClawNanny Governance Policy section below governs your agents regardless of which execution runtime is active — ClawNanny Governance Policy (CGP) rules, Agent Capability Categories (ACC), file-system access controls, and the activity audit all apply to OpenClaw today, and to Hermes-Agent and any future runtime as they become available.
The Runtime section beneath it shows what the currently active runtime (OpenClaw) is doing inside that policy envelope — its tool inventory, its sandbox state, and runtime-specific audits.
1a. Governance Policy — CGP and ACC
Declarative rules that govern what your agents are allowed to do, written in capability terms (read, send, pay, publish, delete, admin) rather than tied to any specific tool or vendor. CGP compiles down to whichever runtime is active.
1b. File-system policy
What paths an agent (any runtime) may read, write, or delete on this machine. Enforced by ClawNanny's file-system wrappers; every check produces an audit row.
1c. Activity log
Runtime-blind audit trail of permission checks, governance decisions, and file-system denials. All entries go to tblActivityLog.
1d. Credential vault
Unified ClawNanny-owned storage for API keys (Anthropic, OpenAI, SMTP, channel tokens). Removes per-runtime key management; values live in the OS keyring (Windows DPAPI / macOS Keychain).
Available after June 2026.
2a. Runtime-specific audits
Configuration and security audits that interrogate the active runtime's own settings, plugins, and sandbox state.
2b. Tool inventory & sandbox
Which tools the active runtime is exposing right now, and its current sandbox / process-isolation state. After June 2026, ClawNanny will overlay a workspace-scoped enable/disable policy on this list.
Unified tool inventory view available after June 2026. In the meantime, see the runtime's configuration audit above.
2c. Guardrails (MoltGuard / OpenGuardrails)
Input/output content filter currently wired into OpenClaw. After June 2026, this moves into ClawNanny as request middleware so it covers any runtime.
Configure today via the active runtime's plugin settings.
Health sense-making
This report is generated from the last run of the health audit. For each issue noted, you may apply an automated fix or accept the risk (which removes it from the indicator count in the upper-left corner of the page).
Watchdog Health Checks
| Time | Type | Method or Event | OK | Context | Summary |
|---|---|---|---|---|---|
| Waiting for events ... | |||||
