@atlaspa

openclaw-signet verifies installed skill integrity via sha-256 hashing, detecting file modifications, additions, and removals. detects tampering but alerts only; upgrade path exists for enforcement actions.

openclaw-reflect observes tool outcomes across sessions, detects recurring failure patterns, and proposes memory/instruction changes through a separate evaluator gate with tiered approval thresholds and rollback capability.

openclaw-action integrates github security scanning into agent skill workflows, flagging exposed secrets, injection patterns, and exfiltration signals via pr annotations and structured outputs.

openclaw-egress scans agent workspaces for outbound network risks including exfiltration endpoints, suspicious domains, and encoded payloads. provides alert layer with exit codes and domain mapping, requires python3.

openclaw-memory provides persistent semantic storage for agent context across sessions with local sqlite backends and optional paid tier for unlimited retention, but lacks concrete invocation patterns and error handling detail.

Compliance and policy enforcement for agent workspaces. Define security policies, audit compliance, check command restrictions, and generate audit-ready reports. Free alert layer — upgrade to openclaw-marshal-pro for active enforcement, blocking, and automated remediation.

Tracks LLM API costs in real-time, enforces budget limits with circuit breakers, and enables autonomous agent payments via the x402 protocol.

Intelligently compresses and optimizes context to reduce token usage by 40-60%, using deduplication, pruning, summarization, and adaptive learning.

Automatically routes AI requests to cost-optimal models based on task complexity and budget, saving 30-50% on model expenses with adaptive learning.