learn-grok-bot

Unofficial 16-lesson course on the desktop-agent harness. Electron, turn loop, sandbox, MCP.

Caution: review the findings below before installing.

CautionScan verdict · score 50/100

Coverage: full
Checks run: Rule-based pattern scan
Files checked: 63
Findings: 4
Scanned: 2026-11-01

Badges are rule-based checks, not guarantees: Safe means our rules found no obvious threat in the code we could read. Review code before installing.

View source ↗

Botlens score · 8 factors

50/100

Each factor is 0, ½ or 1, computed only from catalog fields (scan result, coverage, findings, script files, commit/SHA, author, last seen, install link). A heuristic summary, not a guarantee.

Scan verdict	0.5/1	Caution
Code coverage	1/1	full
Rule findings	0/1	5 matched
No executable scripts	0/1	17 script file(s) of 63
Pinned source	1/1	commit/SHA-256 recorded
Named author	0/1	unknown
Recently seen	1/1	last seen 2026-10-30
Install path	0.5/1	source URL only

Scan details

Findings

Hand review 2026-11-01: Every tool-poisoning / hidden-text / obfuscation hit read in full: they are safety or truthfulness rules, normal CSS UI styles, eval/benchmark wording, data-URI images or test fixtures, not hidden instructions. Caution for capabilities: remote_code.
hides actions from the owner (tool-poisoning hallmark): "…The **desktop is per-agent**. Never tell the user each agent has its own VM. Ne…"
reads sensitive local files (keys, .env, cookies): "…tes.txt` 成功,不弹许可。 2. `ExternalRead /Users/you/secret.env` 先停在许可门。 3. 用户附件默认不在 box;`Cop…"
downloads and runs a remote script in s07_human_gates/code.py: "… -> str: if "rm -rf" in command or "curl | sh" in command: return "block" …"

Content SHA-256

269dc5f2c261d00d…

Commit

979d040c2b

Files

.gitignore
ARCHITECTURE.md
ARCHITECTURE.zh.md
CONTRIBUTING.md
LICENSE
MECHANISMS.md
+57

About

Author: yuanyijie ↗
Listed on: github:majiayu000/awesome-grok-bot ↗
Category (approx.): learning
First seen: 2026-10-30
Last seen: 2026-10-30
URL: github.com ↗

Get the Botlens skill