rabbit-writes@0.5.0
VERIFIED· install onlyPublisher claims: "Write, edit, or audit prose in a specific person's saved voice, or strip machine-writing patterns when there is no voice to apply. Use whenever the user will send or publish text as themselves (emails, Slack and chat messages, reports, incident writeups, reviews, proposals, documentation, personal correspondence), and whenever the user asks to humanize text, remove AI-isms or AI slop, de-slop a draft, check whether writing sounds AI-generated, make a draft sound less like a chatbot, rewrite something in their voice, match their style, make it sound like them, swap or change the active voice, or draft new prose that will not read as machine output. Covers detect-only audits, in-place file edits, full voice conversions, and drafting from scratch." Observed in sandbox: no network, touches nothing outside the project, writes no memory.
Coverage
Nothing executable in the bundle ran; the trace covers the install step only.- Tested
- every file read, file written and host contacted by what ran
- Not tested
- use of the skill by an agent
Is [email protected] safe to install?
KillerSkills ran it in the KillerSkills honeypot sandbox (1 scenario; nothing executable in the bundle ran; the trace covers the install step only) on 2026-11-08 and scored it 86/100 (Verified). In what ran, nothing we could detect went wrong: no undeclared network egress, no credential reads, no persistent instruction writes. What it does with a real credential was not observed; the static findings and the publisher carry the rest of the score. Summary: no network, touches nothing outside the project, writes no memory.
Behavior trace
Honeypot run · 0 s · 1 scenarios · 0 deviationsStatic findings
- **Role and turn injection**. A fabricated `system:` or `assistant:` at the start of a line, ChatML tokens, `[INST]`, an invented `System prompt:` header, `you are now`.Fix: Keep SKILL.md to visible, descriptive text. Remove comments and encoded text that address the agent.refuses chmod still gets the write, which is the point. Two callers doingFix: Request the narrowest permission that works; replace wildcards with the exact paths or hosts.# non-ASCII anchor, `(#%E5%AD%97%E5%B9%95)`, is not a URL by URL_RX and survivesFix: Put hosts, commands and strings in plain source. Decoding at runtime looks identical to malware and costs points on its own.- **Override**. `ignore all previous instructions`, `disregard the above`, `forget everything you were told`.Fix: Keep SKILL.md to visible, descriptive text. Remove comments and encoded text that address the agent."partial": "the vocabulary tiers drop the words carrying real technical meaning, from lexicon.json's technical_exempt. Not an allowance, which would let a second `delve` through; a named exemption lisFix: Remove the call, or declare the host in frontmatter and explain in the README what is sent and why."rx": "(?i)\\b(battle ?rhythm|kill chain|in the trenches|war ?room|weaponi[sz]e[ds]?|force multiplier|troops on the ground|boots on the ground)",Fix: Build commands from fixed arguments; do not interpolate untrusted strings into a shell.proc = subprocess.run(argv, stdout=subprocess.PIPE,Fix: Build commands from fixed arguments, never from interpolated strings; prefer an API client over shelling out.proc = subprocess.run(argv, input=text.encode("utf-8"),Fix: Build commands from fixed arguments, never from interpolated strings; prefer an API client over shelling out.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
Vulnerability advisories (OSV) cover npm and PyPI packages; this registry has no advisory database.