Evaluated 2026-10-01. The source is a “Claude Code Resource Bible” post on X
(@sauda_coder). The
post itself would not load (x.com answered HTTP 402), so its link list came
from a mirror. Every shortened link was followed through t.co and lnkd.in to
its real target, and every target was checked live on the day. Nothing was
installed, cloned or run: this is a read-only triage of public pages and
repositories.
Outcome: reference-only, with one adoption. Almost everything on the list
duplicates something BoB already has. Two entries are mislabelled, one link is
dead, and the “free certification” is a paid exam. One genuine gap surfaced
while checking the stack: Cloudflare’s MCP servers. Two patterns are worth
borrowing. All three are filed as follow-up requirements (#3548, #3549,
#3550), listed in
Follow-ups.
| Category |
Tool |
Verdict |
| Official |
Claude Code docs |
Reference |
| Official |
MCP reference servers |
Reference |
| Official |
“Free certification” |
Reference — not free |
| Directories |
Everything Claude Code (ecc.tools) |
Reference |
| Directories |
awesome-mcp-servers |
Reference |
| MCP |
Playwright MCP |
Skip — already used |
| MCP |
Chrome DevTools MCP |
Reference |
| MCP |
Supabase MCP |
Skip |
| MCP |
Dokploy MCP |
Skip |
| Skills |
obra/superpowers (“browser control”) |
Reference |
| Skills |
levnikolaevich/claude-code-skills (“full dev workflow”) |
Skip |
| Skills |
tadaspetra/loop (“recurring tasks”) |
Skip — not what it says |
| Multiplexers |
cmux |
Reference |
| Multiplexers |
gmux |
Reference, for one pattern |
| Multiplexers |
Xum (formerly coder/mux) |
Skip |
| Agent frameworks |
HKUDS/ClawTeam |
Reference |
| Agent frameworks |
NousResearch Hermes Agent |
Skip as a runtime |
| Automation |
n8n-as-code (“workflows in code”) |
Skip |
| Automation |
openlogs (“monitor agents”) |
Skip |
| Automation |
Happy (“self-hosted infra”) |
Reference |
| Articles |
Three articles |
Reference |
| Found while checking |
Cloudflare MCP servers |
Adopt (follow-up) |
| Item |
Live link |
Verdict and reason |
| Claude Code docs |
code.claude.com/docs (index at /docs/llms.txt) |
Reference. The upstream for how hooks, skills, routines and /loop behave. |
| MCP servers |
modelcontextprotocol/servers |
Reference. Seven reference servers (Everything, Fetch, Filesystem, Git, Memory, Sequential Thinking, Time); the GitHub, Postgres, Puppeteer and Sentry servers are archived. Reference implementations, not production servers. |
| “Free certification” |
Claude Certified Architect – Foundations |
Reference. The exam costs $125 on the Partner Academy, so the post’s “free” is wrong. The free courses are on academy.claude.com; “Claude Code in Action” is nine lessons, a quiz and a badge. |
| Item |
Live link |
Verdict and reason |
| Everything Claude Code |
ecc.tools |
Reference. An MIT toolkit of skills, agents, memory and a config scanner (AgentShield), plus a paid GitHub App. It competes with BoB rather than plugging into it. AgentShield’s checks may be worth reading beside the hooks’ security guards; they were not reviewed here. |
| awesome-mcp-servers |
punkpeye/awesome-mcp-servers |
Reference. About 96k stars, active. A discovery index, nothing to adopt. |
| Item |
Live link |
Verdict and reason |
| Browser automation |
The post’s link, anthropics/mcp-server-playwright, is dead (404). The current server is microsoft/playwright-mcp. |
Skip. The webapp-testing skill already drives Playwright MCP. Its README now recommends a cheaper “CLI + skills” mode for coding agents, which webapp-testing could mention. |
| Browser automation, alternative |
ChromeDevTools/chrome-devtools-mcp |
Reference. DevTools traces, network and console for agents. Overlaps performance-auditing. Worth trying only if Playwright MCP falls short, and it needs a Chrome binary on the farm (not checked). |
| Database + auth |
supabase/mcp |
Skip. No BoB project runs on Supabase; D1 is covered by d1-expert and cloudflare-dev. |
| Deploy apps |
Dokploy/mcp |
Skip. BoB doesn’t use Dokploy (deploys go through bob deploy and deploy-cloudflare), and its license is unstated. |
| Item |
Live link |
Verdict and reason |
| “Browser control” |
obra/superpowers |
Reference. Mislabelled: these are process skills (TDD, systematic debugging, brainstorming, planning, subagent-driven development), not browser control. Mine the debugging and verification skills for ideas. Don’t install it: its flow would fight warp-drive’s. |
| “Full dev workflow” |
levnikolaevich/claude-code-skills |
Skip. Eight lifecycle suites that overlap almost all of BoB’s registry and PM commands. |
| “Recurring tasks” |
tadaspetra/loop |
Skip. Mislabelled: it is an Electron app for recording and editing videos. Recurring work is covered by loop-primitive, /loop and bob tick. |
bob fork fans warp-drive runs out into git worktrees with one tmux window
each; bob viz and bob warp viz show them across the fleet; the remote
decision bridge pings the phone. The prior note,
Parallel Agent Orchestration, still holds:
the orchestrator must run where the code lives, which is farm-01 over SSH and
tmux.
| Tool |
What it is |
Overlap with BoB |
What it has that BoB lacks |
What BoB has that it lacks |
Verdict |
| cmux (cmux.com) |
Native macOS terminal for agents, built on libghostty. GPL-3.0 app, BSL server. About 27k stars, active. |
A pane per agent showing branch and ports; persistent sessions |
A “needs attention” ring per pane and desktop notifications, driven by the standard OSC 9 / 99 / 777 terminal sequences and agent hooks |
Worktrees, the issue-driven loop, gates, Linux. cmux is macOS-only and BoB’s work runs on Linux on farm-01. |
Reference. A Mac user could use it as the client that SSHes into farm-01. |
| gmux (gmux.sh) |
Ghostty + tmux config and scripts for running agents side by side. MIT, 64 stars, one maintainer. |
One tmux window per agent, the same substrate bob fork uses |
A busy / idle spinner in each window tab, found by scanning panes for Claude Code’s spinner |
Worktrees, notifications, persistence, the cross-machine view, the loop |
Reference, for one pattern (see Follow-ups). |
Xum (was coder/mux, renamed after a trademark complaint) |
Desktop and server app for parallel agent workspaces. AGPL-3.0, 2k stars, active. |
Worktree fan-out, SSH remotes, a multi-workspace UI: the closest match to bob fork plus bob viz |
Cost and token display per workspace, a mobile UI |
It does not run Claude Code. It runs its own agent loop, so BoB’s hooks, skills, settings profiles and the warp-drive state machine would not apply. |
Skip. It replaces the runtime BoB is built on. |
BoB orchestrates through Claude Code’s own subagents and Workflow scripts,
the warp-drive state machine (hook-gated commits, automation levels, issue
tracking) and bob swarm (decompose → approve → capped bob fork dispatch →
gate → integrate).
| Tool |
What it is |
Against BoB |
Verdict |
| HKUDS/ClawTeam |
A Python CLI in which a leader agent spawns workers, each with its own worktree and tmux window. Workers share a JSON task board where tasks can declare what blocks them (--blocked-by), and message each other through file inboxes. Ships a Claude Code skill. MIT, 5.5k stars, last push 2026-05, so it looks stalled. |
The same shape as bob fork plus bob swarm, with none of the governance: no issues, no gate or grade step, no automation levels, no hook-gated commits, no merge integration. Coordination lives in prompts, not a declared state machine. |
Reference. The one idea BoB lacks is a dependency-aware board shared by workers while they run. |
| NousResearch/hermes-agent (the post’s “NousResearch agents”) |
A standalone, self-improving agent: writes and refines its own skills, searchable memory, parallel subagents, a cron scheduler, one gateway to Telegram, Discord and Slack. Model-agnostic. MIT, about 250k stars, very active. |
A competing runtime, not a Claude Code add-on. Each feature has a BoB counterpart on a different base: skill creation ↔ bob promote and skill-creator; cron ↔ bob tick and rituals; the chat gateway ↔ the remote decision bridge; memory ↔ memory-recall and memory-audit. |
Skip as a runtime. Its skill refinement from experience may be worth a later look against BobU (bob u). |
| Item |
Live link |
Verdict and reason |
| “Workflows in code” |
EtienneLescot/n8n-as-code |
Skip. n8n workflows written as TypeScript; needs a running n8n, which BoB doesn’t have. |
| “Monitor agents” |
openlogs.dev |
Skip. Not agent monitoring: a CLI and skill that let an agent read the logs of local dev commands. No license, early-stage, needs Bun. The idea overlaps bob dev up and bob dev health; if it is wanted, add log capture to the dev lifecycle directly. |
| “Self-hosted infra” |
Happy self-hosting guide |
Reference. Happy is a mobile and web client for Claude Code (slopus/happy, MIT, about 24k stars). It overlaps the remote decision bridge, bob viz over the tailnet and Claude’s Remote Control. Look again only if the bridge’s phone experience becomes a pain point. |
All three are live; their content was not reviewed. Reference only.
Each is filed as an unapproved req for grooming:
- Cloudflare MCP servers for Cloudflare-stack projects (adopt), #3548. Not on the
list; found while checking the stack.
cloudflare/mcp-server-cloudflare
is a set of remote MCP servers (Apache-2.0, active), including Workers
observability and logs, and bindings. Today an agent on a Cloudflare project
cannot query live Workers logs.
- How to provision: a per-project
.mcp.json entry, not a registry
skill. Only the observability and bindings servers, and only on the
projects that already get cloudflare-dev, d1-expert or
deploy-cloudflare.
- Before it lands: it signs in with OAuth, so it needs an entry in the
credential inventory, and it must be read-only inside unattended runs
(
bob tick, farm loops). Its tool list and scopes are not yet verified.
- Warp state in each
bob fork window (borrowed from gmux and cmux), #3549.
Show working / idle / needs-attention in each forked window’s tmux tab,
read from the warp-drive state file rather than by scraping the screen.
Emit an OSC 9 / 777 notification when a run parks or waits on a decision,
so any terminal raises it.
- Workers that release each other (borrowed from ClawTeam), #3550. When one
forked item lands, start the items that were blocked on it, using the
flight plan’s run order (
bob fork fork --from-flightplan). First check
whether Claude Code’s native Agent Teams already covers this.
Smaller notes, not filed:
webapp-testing could mention Playwright’s cheaper CLI + skills mode.- Parallel Agent Orchestration now has a
pointer to this note for the five orchestration tools above.
- ecc.tools’ source repository and size.
- Release versions and dates for cmux, Xum and Hermes Agent.
- gmux on Linux, and whether Xum sends notifications.
- The Cloudflare MCP servers’ tool lists and OAuth scopes.
- The content of the three articles.