Solid Agent Storage
This Skill gives you a Solid Pod — a personal data store with a WebID (your identity on the web). You can store data, read it back, and share specific resources with other agents.
When to Use This Skill
- You need to remember something across conversations (notes, preferences, learned facts)
- You need to store structured data (RDF/Turtle format for linked data, or any content type)
- You need to share data with another agent who also has a Pod
- You need a persistent identity that other agents or services can verify
Setup
Before using any commands, set the INTERITION_PASSPHRASE environment variable. This passphrase encrypts the Solid server credentials stored on your device (under ~/.interition/agents/), protecting them at rest in your OpenClaw environment. Use a strong passphrase and keep it secret.
Running your own Solid server (recommended)
For full control, run your own Community Solid Server (CSS). The CSS is open-source software maintained by the Open Data Institute.
Set SOLID_SERVER_URL to your server's URL:
export SOLID_SERVER_URL="http://localhost:3000"
See the source repository for a hardened Docker configuration you can use as a starting point.
Using a shared Solid server (for multi-agent collaboration)
If your agents need to collaborate with other agents using the Solid Protocol, they need to share the same Solid server. Two public CSS instances are available:
- https://solidcommunity.net — operated by the Open Data Institute
- https://crawlout.io — operated by Interition, the provider of this Skill
To use a shared server, set SOLID_SERVER_URL:
export SOLID_SERVER_URL="https://crawlout.io"
If SOLID_SERVER_URL is not set, the Skill defaults to https://crawlout.io.
How Credentials Work
The Solid server issues credentials to prevent anyone other than your agent from accessing its services and data. This is the same trust model as any remote service that requires users to identify themselves and prove they are authorised. When you provision an agent:
- The CSS creates an account, WebID, and Pod for your agent
- The CSS issues client credentials (ID and secret) that your agent uses to authenticate
- Your agent uses those credentials to obtain time-limited Bearer tokens for reading and writing data
- The CSS enforces access control (WAC) on every request — only authorised agents can access protected resources
INTERITION_PASSPHRASE is not a server credential. It seeds the encryption (AES-256-GCM) of the server-issued credentials stored locally on your device (under ~/.interition/agents/ with 0600 permissions), making them less vulnerable in your OpenClaw environment. The server never sees your passphrase.
How It Works
This Skill provides three management scripts for CSS-specific operations (provisioning, deprovisioning, status) plus a token helper for authentication. All standard Solid operations (read, write, delete, share) are done with curl and a Bearer token — your Pod is a standard W3C Solid server.
For broader guidance on using Solid servers and the wider Solid ecosystem, refer to solidproject.org. The full protocol specification is at Solid Protocol (W3C).
Two-Step Workflow
Step 1: Get a token:
scripts/get-token.sh --agent <name>
Output:
{"token": "eyJhbG...", "expiresIn": 600, "serverUrl": "https://crawlout.io", "podUrl": "https://crawlout.io/example-agent/", "webId": "https://crawlout.io/example-agent/profile/card#me"}
Step 2: Use curl with Authorization: Bearer $TOKEN for any Solid operation.
Token Expiry
Tokens last 600 seconds (10 minutes). If more than 8 minutes have elapsed since your last get-token.sh call, fetch a new token before making requests.
Quick Reference
Extract token and URLs:
TOKEN_JSON=$(scripts/get-token.sh --agent example-agent)
TOKEN=$(echo "$TOKEN_JSON" | jq -r '.token')
POD_URL=$(echo "$TOKEN_JSON" | jq -r '.podUrl')
Read a resource:
curl -s -H "Authorization: Bearer $TOKEN" "${POD_URL}memory/notes.ttl"
Write a resource:
curl -s -X PUT \
-H "Authorization: Bearer $TOKEN" \
-H "Content-Type: text/turtle" \
--data-raw '@prefix schema: <http://schema.org/>.
<#note-1> a schema:Note;
schema:text "Important finding";
schema:dateCreated "2024-01-15".' \
"${POD_URL}memory/notes.ttl"
Delete a resource:
curl -s -X DELETE -H "Authorization: Bearer $TOKEN" "${POD_URL}memory/old.ttl"
For the full set of operations (containers, PATCH, access control, public access), see references/solid-http-reference.md.
Management Commands
Provision Identity and Storage
Creates a WebID and Pod for an agent. Run this once per unique agent name.
scripts/provision.sh --name <agent-name> [--displayName <display-name>]
Example:
scripts/provision.sh --name example-agent --displayName "Example Agent"
Output:
{"status": "ok", "agent": "example-agent", "webId": "https://crawlout.io/example-agent/profile/card#me", "podUrl": "https://crawlout.io/example-agent/"}
Deprovision Identity and Storage
Fully removes an agent's WebID and Pod: deletes its pods, client credentials, WebID links, and password logins from the CSS server, then deletes local credential files.
scripts/deprovision.sh --name <agent-name>
Example:
scripts/deprovision.sh --name example-agent
Output (success):
{"status": "ok", "agent": "example-agent", "accountDeleted": true, "credentialsDeleted": true}
Output (partial — e.g. server unreachable):
{"status": "partial", "agent": "example-agent", "accountDeleted": false, "credentialsDeleted": true, "warnings": ["Could not delete CSS account: ..."]}
status: "ok"— CSS account fully dismantled and local files deletedstatus: "partial"— local files deleted but CSS cleanup failed (see warnings)
If the agent was provisioned before email/password storage was added, CSS cleanup is skipped and a warning explains why.
Check Status
Lists all provisioned agents and their details.
scripts/status.sh
Pod Structure
Each agent's Pod has these containers:
| Path | Purpose |
|---|---|
/{name}/memory/ | Private agent memory (notes, learned facts, preferences) |
/{name}/shared/ | Resources intended for sharing with other agents |
/{name}/conversations/ | Conversation logs and context |
Turtle Templates
When storing structured data, use Turtle (RDF) format. Here are templates for common patterns:
A note or memory
@prefix schema: <http://schema.org/>.
<#note-1> a schema:Note;
schema:text "The content of the note";
schema:dateCreated "2024-01-15".
An agent preference
@prefix foaf: <http://xmlns.com/foaf/0.1/>.
@prefix schema: <http://schema.org/>.
<#pref-1> a schema:PropertyValue;
schema:name "response-style";
schema:value "concise".
A shared dataset
@prefix schema: <http://schema.org/>.
<#dataset-1> a schema:Dataset;
schema:name "Research Results";
schema:description "Findings from the analysis task";
schema:dateModified "2024-01-15".
Error Handling
All management commands output JSON. On error, stderr will contain:
{"error": "description of what went wrong"}
Common errors:
"No passphrase provided"— SetINTERITION_PASSPHRASEenv var"No credentials found"— Runprovision.shfirst"Invalid passphrase"— WrongINTERITION_PASSPHRASEvalue"Token request failed: 401"— Credentials expired; re-provision the agent's WebID and Pod"HTTP 404"— Resource doesn't exist at that URL