The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill contains explicit instructions and bash commands for the agent to read and display sensitive credential files. Evidence includes commands targeting:
~/.openclaw/credentials/telegram/*/token.txt (Telegram Bot API tokens)
~/.openclaw/credentials/bird/cookies.json (X/Twitter session cookies)
~/.openclaw/agents/main/agent/auth-profiles.json (Anthropic auth tokens)
~/.openclaw/credentials/whatsapp/default/ (Session keys for WhatsApp communication)
[COMMAND_EXECUTION]: The skill relies extensively on shell command execution to manage the platform, including:
Modifying system configuration files using jq and redirection (> /tmp/oc.json && mv /tmp/oc.json ...).
Managing background processes and spawning sub-agents via shell commands.
Performing file system operations such as rm -rf on credential directories in the 'Nuclear fix' section.
[EXTERNAL_DOWNLOADS]: The skill documents and encourages the installation of external code from remote sources through commands like:
clawdhub install <slug> which fetches skills from a centralized registry.
npx add-skill <repo> which can pull and execute code from arbitrary GitHub repositories.
[INDIRECT_PROMPT_INJECTION]: The skill describes a platform architecture that creates a significant surface for indirect prompt injection:
Ingestion points: Inbound messages from WhatsApp, Signal, and Telegram are processed by the agent.
Capability inventory: The agent has the ability to execute shell commands, read/write sensitive configuration files, and manage background tasks.
Sanitization: No boundary markers or sanitization procedures are mentioned in the runbook for handling external message content.
[DYNAMIC_EXECUTION]: The skill uses inline Python scripts (python3 -c "...") to process JSONL data and logs at runtime. It also supports the execution of TypeScript extensions and dynamic background agents (Codex, Claude Code) via shell PTY sessions.
[PERSISTENCE]: The skill manages a custom cron system (~/.openclaw/cron/jobs.json) which allows the agent to schedule autonomous tasks that persist across sessions.

openclaw-config