The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Fetches advisory feeds and skill catalog indexes from trusted vendor domains (prompt.security) and GitHub releases.
All downloads are cryptographically verified using Ed25519 signatures and SHA-256 checksums against pinned public keys before processing.
Remote connections enforce a minimum of TLS 1.2 and use a strict domain allowlist to prevent unauthorized network access.
[PROMPT_INJECTION]: Identifies a potential surface for indirect prompt injection within the 'action' and 'description' fields of the advisory feed.
Ingestion points: Remote and local feed.json files are processed by the hook and heartbeat scripts.
Boundary markers: None; advisory text is interpolated directly into agent messages.
Capability inventory: The suite can execute skill installations via npx clawhub and manage agent-level cron jobs.
Sanitization: All feed data is required to be signed by the trusted vendor's Ed25519 key, significantly mitigating the risk of third-party injection attacks.
[SAFE]: The skill uses a shared portability helper for path resolution that explicitly rejects unexpanded home tokens and prevents path confusion across different shells and operating systems.
[SAFE]: Persistence is managed via the agent's native cron functionality (openclaw cron) rather than direct modification of system-level configuration files or shell profiles.

clawsec-suite