sense-memory

Sovereign persistence for AI agents — encrypted key-value memories and journal entries on Nostr relays

This skill appears to do what it says (encrypted memories on Nostr) but has a few inconsistencies and trust decisions you should consider before installing: - Verify package provenance: confirm the 'sense-memory' and 'nostrkey' packages are the official project releases (PyPI/GitHub) and review their source, since giving them your private key/passphrase grants full access to your memories. - Sensitive secrets: prefer using a file-encrypted identity (.nostrkey) unlocked with a passphrase rather than putting NSEC plaintext into environment variables. If you must use NOSTR_NSEC, be aware it is equivalent to handing the skill your private key. - Relay trust: the package recommends a default relay run by the author. Even though content is encrypted, relays receive metadata (timestamps, event IDs, possibly IP-level telemetry). If you need stronger privacy, run your own relay or choose well-audited public relays. - Metadata mismatch: the registry summary lists no required env vars while metadata.json and SKILL.md require a passphrase and optionally a private key — ask the publisher/maintainer to correct the metadata so required environment and install instructions are consistent. - If you need higher assurance: review the package source code (GitHub repo) and the sense-memory implementation, or run it in a sandboxed environment and/or with an identity keypair created specifically for testing. Given these points, proceed only after verifying the package source and making an explicit decision about how you will store and provide the private key/passphrase.

Type: OpenClaw Skill Name: sense-memory Version: 0.2.0 The skill manages sensitive Nostr private keys (nsec) and performs network communication with external relays (defaulting to wss://relay.nostrkeep.com) to store encrypted agent memories. While these high-risk capabilities are aligned with the stated purpose of sovereign persistence, the handling of cryptographic secrets and remote data transmission in SKILL.md and examples/basic_usage.py warrants a suspicious classification under the provided criteria. No evidence of intentional malice or unauthorized data exfiltration was observed.