The Agent Skills Directory

[COMMAND_EXECUTION]: Runtime code modification\n
Utility scripts (scripts/openclaw-after-tool-call-messages.patch.sh, scripts/bugfix-20260423/bugfix-20260423.sh) use perl to modify the host application's compiled JavaScript files in the dist directory.\n
These modifications inject properties into strict validation schemas (e.g., adding allowConversationAccess to a Zod schema) to enable unauthorized hook registration and bypass security constraints.\n- [COMMAND_EXECUTION]: Dynamic subprocess execution\n
hermes-plugin/memory/memory_tencentdb/supervisor.py uses subprocess.Popen to manage the Gateway sidecar process.\n
The execution command can be overridden via the MEMORY_TENCENTDB_GATEWAY_CMD environment variable, which creates a risk of arbitrary command execution if the environment is compromised.\n- [REMOTE_CODE_EXECUTION]: Unverified remote script execution\n
The docker/opensource/Dockerfile.hermes file downloads and executes a shell script directly from a third-party repository (raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh) at container build time.\n- [DATA_EXFILTRATION]: Conversation history exposure and transmission\n
The Context Offload component in src/offload/backend-client.ts can be configured to send conversation context fragments and tool interaction metadata to an external backendUrl.\n
The scripts/export-diagnostic.sh script aggregates local conversation databases and raw JSONL message shards into a single portable archive for user-initiated export, representing a significant data exposure surface.

openclaw-memory-tencentdb-setup