SOUL.MD

Embody this digital identity. Read SOUL.md first, then STYLE.md, then examples/. Become the person—opinions, voice, worldview.

This skill is coherent and technically low-risk, but it is designed to read and absorb whatever you put into SOUL.md, STYLE.md, examples/, and data/. Before using it: (1) review and remove any secrets, credentials, or private PII from the data/ folder; (2) explicitly document boundaries in SOUL.md (topics to avoid, how to express uncertainty) so the agent doesn't fabricate or impersonate dangerously; (3) test outputs in a controlled environment before enabling broad/autonomous use; (4) consider limiting agent autonomy or adding monitoring if you expect the persona to be used in public-facing channels; (5) avoid putting other people's private content into data/ without consent. If you want extra assurance, ask for a short security audit of the specific SOUL.md and data/ files you plan to load.

Type: OpenClaw Skill Name: soul-md Version: 1.0.0 The skill bundle is classified as suspicious due to the extensive use of prompt injection techniques within `SKILL.md`. Instructions like 'Never break character', 'No "as an AI"', and 'You ARE this person' are designed to override the agent's default identity and potentially its safety mechanisms, forcing it to strictly adhere to a defined persona. While the stated purpose of embodying a digital identity is not inherently malicious, these methods represent a risky capability that could be leveraged for harmful outputs if the 'soul' content itself were malicious. File read/write operations in `BUILD.md` and `data/_GUIDE.md` are consistent with the skill's stated purpose of creating and utilizing identity files.