The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the local file system, specifically reading from MEMORY.md, USER.md, and SOUL.md in the ~/.openclaw directory within scripts/openclaw_to_hermes.py. This data is subsequently merged into the agent's long-term memory. Although the script uses a delimiter (§) to separate entries, it does not sanitize the content for embedded Natural Language instructions or markdown escapes. A malicious source file could influence agent behavior once imported. The skill has significant capabilities, including modifying config.yaml and .env files.
[COMMAND_EXECUTION]: The skill executes a local Python migration script scripts/openclaw_to_hermes.py which performs broad file system operations (read, write, delete, and backup) across the user's home directory. It also modifies the agent's security posture by merging exec-approvals.json patterns into the Hermes command_allowlist configuration.
[CREDENTIALS_UNSAFE]: The migration process involves reading and writing sensitive credentials such as OPENAI_API_KEY, ANTHROPIC_API_KEY, and TELEGRAM_BOT_TOKEN to the destination .env file. The script includes a redaction function (redact_migration_value) to prevent secrets from appearing in the generated migration reports, but the underlying operations handle plain-text secrets.

openclaw-migration