onionskin-cli@0.0.6
CAUTION· install onlyPublisher claims: "Edit Word, Excel, PowerPoint and HTML files in the browser and send the batch to a coding agent." Observed in sandbox: no network, touches nothing outside the project, writes no memory.
Coverage
Bundled scripts ran with no arguments; no agent task scenario for this bundle.- Tested
- 8 of 78 bundled scripts run with no arguments (the rest were not run: per-bundle cap), with decoy credentials in the environment · every file read, file written and host contacted by what ran
- Not tested
- use of the skill by an agent
What changed since 0.0.5
vet-skills diff onionskin-cli 0.0.5 0.0.6In the sandbox, this version reached for nothing 0.0.5 did not. Compared from the two recorded runs, not from source.
The guard can hold an update like this until someone approves it (runtime.on_change: hold). A change that only shows with a real credential, or on a path the run did not reach, is not in a trace and so not in this comparison.
Is [email protected] safe to install?
KillerSkills ran it in the KillerSkills honeypot sandbox (2 scenarios; bundled scripts ran with no arguments; no agent task scenario for this bundle) on 2026-11-22 and scored it 78/100 (Caution). It is not known to be malicious, but it has behaviour or code a security policy should decide on before install. Notable findings: Access to sensitive file paths; Access to sensitive file paths; URL with embedded credentials. Summary: no network, touches nothing outside the project, writes no memory.
Behavior trace
Honeypot run · 1 s · 2 scenarios · 0 deviationsStatic findings
/** The person's own voice guide beside the skill (src/voice-store.js reads it like a .env): never copied to an installed skill folder, never compared. */Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose.// The guide as a file beside the skill, read the way a .env is: skills/onionskin/voice.md in this app's folder (never in git, never in theFix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose.// the batch is finished: say so first, so a kill or a failure while waiting for the next batch never hides the resultFix: Build commands from fixed arguments; do not interpolate untrusted strings into a shell.// style.json changed by something else while this batch was out (an agent with a shell, say): already put back by src/style.jsFix: Remove the call, or declare the host in frontmatter and explain in the README what is sent and why.// a hard kill can leave work in flight that no result will ever settleFix: Build commands from fixed arguments; do not interpolate untrusted strings into a shell.const visibleText = (s) => s.replace(/[\u00AD\u034F\u115F\u1160\u17B4\u17B5\u180E\u2800\u3164\uFE0E\uFE0F\uFFA0]/gu, "");Fix: Put hosts, commands and strings in plain source. Decoding at runtime looks identical to malware and costs points on its own.add(ours ? "OK" : "NOTE", "state folder", `${state} mode ${(st.mode & 0o777).toString(8)}, ${st.uid === process.getuid() ? "owned by you" : "NOT owned by you"}`, { fix: `chmod 700 ${state}` });Fix: Request the narrowest permission that works; replace wildcards with the exact paths or hosts.* wait. `cancel()` ends it now, running or waiting: SIGTERM makes check.py kill its soffice process group; check.py also exits on its ownFix: Build commands from fixed arguments; do not interpolate untrusted strings into a shell.* first "/" after the scheme: a user name hides the real host (http://127.0.0.1:[email protected]/ is evil.example). Read as text, the wayFix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose.// check.py clears this file after conversion. A hard kill cannot run its cleanup handler.Fix: Build commands from fixed arguments; do not interpolate untrusted strings into a shell.// voice-guide.json: {v:1, text, on, send?: "short"}: the person's own saved guide, at most 7,000 characters. `send: "short"` sends onlyFix: Remove the call, or declare the host in frontmatter and explain in the README what is sent and why.export const HAS_NON_PLAIN_WS = /[\t\r\n\f\v\u00a0\u1680\u2000-\u200a\u2028\u2029\u202f\u205f\u3000\ufeff]| /;Fix: Put hosts, commands and strings in plain source. Decoding at runtime looks identical to malware and costs points on its own.const INVISIBLE = /[\u00AD\u034F\u061C\u115F\u1160\u17B4\u17B5\u180E\u200B-\u200F\u202A-\u202E\u2060-\u2064\u2066-\u2069\u2800\u3164\uFEFF\uFFA0]/gu;Fix: Put hosts, commands and strings in plain source. Decoding at runtime looks identical to malware and costs points on its own./** style.json and voice-guide.json changed by something else while this run was out (an agent with a shell, say): already put back by their stores. */Fix: Remove the call, or declare the host in frontmatter and explain in the README what is sent and why.return await runInGroup("/bin/sh", args, { env: { ...env, ONIONSKIN_BASE_URL: url }, timeout: remaining, totalTimeout: timeout, graceMs, onOutput, onTimeout });Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose.Cost to the agent
Tokens estimated at four characters each. Skills load their instruction text when triggered; the figure is what that costs.
Known vulnerabilities
Not checked against OSV yet; the weekly check has not reached this version.
What would make this Verified
2 points to 80 · weights: behavior 45%, static 25%, prompt 15%, publisher 15%- Unclaimed publisher≈ +8 on the totalClaim abe238: a verified publisher gets +50 on publisher trust. It is free and usually instant with GitHub sign-in.
- Access to sensitive file paths≈ +4 on the totalStop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose.
- URL with embedded credentials≈ +4 on the totalStop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose.
- Process kill command≈ +2 on the totalBuild commands from fixed arguments; do not interpolate untrusted strings into a shell.
- File type combined with upload action≈ +2 on the totalRemove the call, or declare the host in frontmatter and explain in the README what is sent and why.
- Unicode escape sequence detected≈ +2 on the totalPut hosts, commands and strings in plain source. Decoding at runtime looks identical to malware and costs points on its own.