Repo Adoption Radar — Vatsal AI/Agent Stack

Decision matrix

Mobile-first answer: what to install, what to steal from, and what to leave alone.

agent-skills

67.5k★ · MIT

Install selectively

Production engineering skills for agentic coding. Low footprint, high leverage.

Medium

Very high

frontend-slides

23.6k★ · MIT

Install / port

Best fit for decks, visual explainers, content/report artifacts.

High

marketingskills

35.3k★ · MIT

Guardrails

Useful for growth/content, but load only when marketing mode is intended.

High

Medium

claude-code-best-practice

61.4k★ · MIT

Borrow only

Great reference; unsafe to copy included broad Claude settings wholesale.

Medium

High

DeepCode

15.9k★ · MIT

Sandbox

Full agentic coding platform. Interesting, but overlaps Cloidberg/Codex/Claude Code.

Low-med

Medium

spacedrive

38.4k★ · FSL

Watch later

Potential file OS layer, but not needed now and privacy surface is big.

Medium

Low

MediaManager

3.2k★ · AGPL

Skip

Media-library automation. Not relevant to AI/agent/build workflows.

Low

None

Fit by actual workstream

The useful view: which repo helps which part of your life.

Workstream	Best repo	Use it for	Avoid
Cursor / Claude Code / Codex	agent-skills	Specs, plans, test discipline, reviews, shipping checklists.	Loading every skill all the time.
Video/deck/content artifacts	frontend-slides	HTML slide decks, explainers, report visuals, polished presentation assets.	Running deploy scripts blindly.
TechWithVatsal growth	marketingskills	Landing pages, positioning, CRO, email/newsletter copy.	Letting marketing voice pollute technical analysis.
Agent ops / workflow design	claude-code-best-practice	Ideas for subagents, hooks, commands, MCP patterns.	Copying broad Bash/Edit settings.
Research-to-code experiments	DeepCode	Controlled bake-off on paper-to-code style tasks.	Main-machine install or broad filesystem mount.

Repo drilldowns

Each card answers: what it does, how, safety, telemetry, costs, and our verdict.

Install selectively

agent-skills

github.com/addyosmani/agent-skills

Prompt pack

Does

24 engineering skills.
Spec → plan → build → test → review → ship.
Claude/Cursor/Codex-friendly guidance.

Risks

No obvious telemetry.
Token/context bloat if overused.
Validation scripts exist but optional.

Verdict: install selectively and port the best parts into Hermes after dedupe.

Install / port

frontend-slides

github.com/zarazhangrui/frontend-slides

Visual artifacts

Does

Beautiful HTML slide decks.
PPTX extraction/export helpers.
Good for explainers and reports.

Risks

No telemetry found.
Shell deploy/export scripts need review.
Mostly model-token cost.

Verdict: high fit. This should become part of the content/report workflow.

Use with guardrails

marketingskills

github.com/coreyhaines31/marketingskills

Growth skills

Does

45 marketing/growth skills.
CRO, SEO, ads, copy, launch, positioning.
Useful for TWV/content packaging.

Risks

No repo telemetry.
References paid analytics/ad tools.
Can make voice too salesy.

Verdict: use only when the task is explicitly marketing/growth.

Borrow only

claude-code-best-practice

github.com/shanraisshan/claude-code-best-practice

Reference

Does

Claude Code tips/patterns.
Subagents, commands, MCP, hooks.
Good agentic engineering reading.

Risks

No telemetry found.
Included settings allow broad Bash/Edit/Write.
Event hooks execute Python.

Verdict: steal ideas. Do not copy the whole .claude folder.

Sandbox only

DeepCode

github.com/HKUDS/DeepCode

Agent platform

Does

Paper2Code/Text2Web/Text2Backend.
Multi-agent coding pipeline.
MCP filesystem/fetch/command tools.

Risks

API keys in config.
Command execution/filesystem access.
Multi-agent token burn.

Verdict: Docker sandbox bake-off only. Not a main-stack install.

Watch later

spacedrive

github.com/spacedriveapp/spacedrive

File OS

Does

Distributed cross-platform file explorer.
Rust/Tauri/React daemon + apps.
Indexing, P2P sync, cloud volumes.

Risks

README says no telemetry.
Large local-file privacy surface.
FSL license, not pure OSS permissive.

Verdict: interesting later for file intelligence. Test folders only.

Skip

MediaManager

github.com/maxdorninger/MediaManager

Media stack

Does

TV/movie library automation.
FastAPI/Postgres/web Docker stack.
qBittorrent/SABnzbd/Transmission.

Risks

No app telemetry init found.
TMDB/TVDB and media ops costs/risks.
AGPL obligations if distributing.

Verdict: irrelevant to our agent/build workflows. Skip.

Telemetry, security, and cost view

The boring table. Boring is where the landmines are.

Repo	Telemetry finding	Security concern	Extra costs
agent-skills	No conventional telemetry found.	Prompt steering/context bloat.	Model tokens only.
frontend-slides	No conventional telemetry found.	Deploy/export scripts need path review.	Model tokens only.
marketingskills	No repo telemetry found.	Can steer style/voice too aggressively.	Paid analytics/ad tools only if used.
claude-code-best-practice	No telemetry found.	Broad Claude permissions + executable hooks.	Model/tool usage only.
DeepCode	No obvious analytics found.	API keys, command executor, filesystem/MCP access.	Potentially high model/API spend.
spacedrive	README says no telemetry.	Indexes local/cloud files, P2P network surface.	None intrinsic; storage/cloud optional.
MediaManager	No app telemetry init found; Prometheus metrics exist.	Downloader/media stack exposure, AGPL, API keys.	TMDB/TVDB, Usenet/VPN/storage if used.

Method

Shallow cloned every repo, recorded commit/license/GitHub metadata, inspected READMEs/install docs/manifests, and ran static sweeps for telemetry strings, risky execution, secrets, and network behavior. This is an adoption-risk assessment, not a formal security audit. Anyone calling a grep pass “compliance” should be put in charge of the printer.