94/100 (Very Low)

OpenClaw： benign
VirusTotal： benign
StaticScan： unknown

OpenClaw： benign

The skill's instructions, lack of installs, and no requested credentials are coherent with its stated purpose of autonomous code work; it asks the agent to read and modify the codebase and to verify c...

[内容已截断]

VirusTotal： benign VT 报告

静态扫描： unknown

README 未提供

{
    "latestVersion": {
        "_creationTime": 1772116727585,
        "_id": "k976tt495sknwbb4w3epa34jv181xm0n",
        "changelog": "Here's what changed and why each patch matters:\nRule 0 — Non-Destructive Default addresses the \"unsupervised reckless changes\" concern directly. The skill now requires the AI to classify every planned action as Additive or Destructive before touching anything. Creating files, adding components, wiring up new code — all safe, proceed. Deleting files, replacing whole systems, schema changes — hard stop, confirm with user first. This makes autonomy safe by design, not by luck.\nRule 3 — The Sobriety Protocol is the core of what you described. The LLM's deepest failure mode isn't laziness, it's building on its own lies. It says \"I set up AuthContext in step 2\" and then imports it in step 5 — except it never actually wrote the file, it just narrated writing it. Now the AI must maintain three distinct states for every task: DONE-VERIFIED (I re-read the file, it's real), DONE-UNVERIFIED (I wrote it but haven't confirmed), and PENDING. The rule is: you cannot chain work onto anything that isn't DONE-VERIFIED. The lie breaks because it can't get past the verification gate.\nThe Known LLM Self-Deception Patterns table is a mirror held up to the model's own predictable failures — the exact lies it tells itself, why they happen, and how to catch them. Naming them explicitly forces recognition.\nThe idea is: a sober LLM isn't one with more tools or better vision. It's one whose confidence is always traceable back to something it actually read, not something it remembers saying.",
        "changelogSource": "user",
        "createdAt": 1772116727585,
        "version": "1.0.2"
    },
    "owner": {
        "_creationTime": 0,
        "_id": "publishers:missing",
        "displayName": "ubuntume",
        "handle": "ubuntume",
        "image": "https:\/\/avatars.githubusercontent.com\/u\/17444017?v=4",
        "kind": "user",
        "linkedUserId": "kn71tc7wrhjmpyz8gaegg4wyj5811n24"
    },
    "ownerHandle": "ubuntume",
    "skill": {
        "_creationTime": 1772116126547,
        "_id": "kd7dnyywvhq2x5wg2d97s95da581wagc",
        "badges": [],
        "canonicalSkillId": "kd73901q9dnva9xt43z13fg89x81xmjs",
        "createdAt": 1772116126547,
        "displayName": "Just Keep Working",
        "forkOf": {
            "at": 1772116126547,
            "kind": "duplicate",
            "skillId": "kd73901q9dnva9xt43z13fg89x81xmjs"
        },
        "latestVersionId": "k976tt495sknwbb4w3epa34jv181xm0n",
        "ownerUserId": "kn71tc7wrhjmpyz8gaegg4wyj5811n24",
        "slug": "just-keep-working",
        "stats": {
            "comments": 0,
            "downloads": 259,
            "installsAllTime": 0,
            "installsCurrent": 0,
            "stars": 1,
            "versions": 3
        },
        "summary": "Autonomous programming mode for openclaw.ai. Use this skill whenever a user requests any code change, feature addition, refactor, bug fix, or project task —...",
        "tags": {
            "latest": "k976tt495sknwbb4w3epa34jv181xm0n"
        },
        "updatedAt": 1772120527217
    }
}