Floom Starter Pack

Quickstart

The simplest path is to paste a single prompt into your agent.

Set up Floom Starter Pack: https://floom.dev/starter

Your agent reads the manifest at that URL and installs the right files for its runtime. No account needed. Nothing to configure.

If you prefer to run it yourself:

npx @floomhq/starter install

The CLI detects which agent config files exist on your machine and writes only to the appropriate paths.

What gets installed

The pack writes three things per supported agent.

Total size on disk: approximately 120 KB for the full pack. Individual skill files average 1.2 KB.

Supported agents

The pack supports five agents out of the box. Each agent reads skill context from a different path.

When you run the installer, it detects which of these directories exist and writes only to the ones that do. If none are found, it prompts you to choose.

Updating

There is no auto-update. To pull the latest manifest and skills, re-run the install command with the @latest tag.

npx @floomhq/starter@latest install --profiles core,dev

@latest fetches the newest package version from npm and refetches the public manifest. The operation is idempotent: skills you have already customised are skipped. The collision check in CLI 0.2.1+ preserves your versions.

Uninstalling

To remove everything the pack installed:

npx @floomhq/starter remove --all

To remove specific skills:

npx @floomhq/starter remove --skills pr-review,brand-voice

To remove an entire profile:

npx @floomhq/starter remove --profiles dev

remove deletes the SKILL.md files it installed, clears the Floom activation block from CLAUDE.md / AGENTS.md, and deletes ~/.floom/manifest.json. Files the collision check skipped (your custom versions) are not touched.

Listing what's installed

To see every skill installed via Floom, which profile it belongs to, and which agent paths it was written to:

npx @floomhq/starter@latest list

Privacy

The Floom Starter Pack is a local install. After the initial download, it operates entirely on your machine.

• No telemetry. No analytics. No usage data is collected or transmitted.
• No account required. No sign-in. No API key.
• The installer downloads skill files from GitHub. After install, there are no ongoing network calls.
• The skill files themselves are static text. They contain no code that executes automatically.

License

The Floom Starter Pack installer and tooling is MIT licensed. You can use, modify, and redistribute it freely.

Individual skills in the pack carry their own licenses. The breakdown:

Each skill file includes a license header. If you redistribute individual skills, preserve the header.

Architecture

The full technical shape of the pack: how the installer resolves targets, what files get written, how agents discover installed skills, and what V0 guarantees.

Mental model

The package contains a manifest plus bundled skill folders. The installer resolves selected profiles, detects local agents, writes skills into their native roots, writes a local index, and adds instructions that teach agents to search locally.

Install flow

When the user runs install, explicit targets win. If no target is passed, the CLI detects local agent config directories and installs only to those agents.

npx @floomhq/starter install --profiles core,dev,writing --yes

The decision tree:

1. If --harness is provided, use the requested agents.
2. Otherwise, detect local agent config directories: ~/.claude, ~/.codex (or CODEX_HOME), ~/.cursor, ~/.config/opencode, ~/.kimi.
3. If any are found, install only to detected agents.
4. If none are found, ask for explicit --harness.

Files written

Each target receives skill folders in its native root, a harness instruction file, and a shared local index at ~/.floom/packs/starter-index.json.

Conflict protection

Every installed skill gets a provenance file. Managed pack skills can be replaced by later installs; untracked user-created folders are protected by default.

1. Plan a skill write. If the destination does not exist, copy the skill folder.
2. If it does exist, check for a .floom-pack.json from @floomhq/starter.
3. If the provenance file exists, replace the managed copy.
4. If it does not, refuse to overwrite unless --force is passed.

Skill discovery

No MCP is required in V0. The discovery loop is local: injected instructions point the agent at the starter index and the local-find-skills skill.

The agent reads the injected instructions, searches the local starter-index.json via local-find-skills, picks the matching SKILL.md files, and uses them for task execution.

Data model

The manifest links profiles to skills, and skills to their upstream source records, keeping provenance legible and auditable.

Next curation pass planned sources:

Verified behaviors

Behaviors confirmed across all five launch targets. These are the acceptance criteria for any release.

• Manifest references existing skill folders.
• Every bundled skill has frontmatter and description.
• Dry-run writes nothing.
• Temp-root install writes skills, index, provenance, and instructions.
• Target autodetection works.
• Missing detected targets produces a clear error.
• --harness claude,codex,cursor,opencode,kimi writes all five launch agents.
• Untracked existing skills are not overwritten.

Methodology

Why we built it this way: curation, activation, and what Floom adds versus upstream catalogs.

How we curated 65 skills from 91,000

Skills.sh indexes 91,035 skills. We curated a focused subset of 65. The selection rules:

The four source tiers, in priority order:

How activation works (the 100% pattern from Vercel)

Vercel published research showing that agents skip 56% of installed skills by default. Even with an explicit "use these skills" prompt, invocation only reaches 70%. The AGENTS.md activation pattern achieves 100%.

The pattern works by embedding per-skill trigger conditions directly into the agent's context file (CLAUDE.md or AGENTS.md). Instead of telling the agent "you have many skills available," it tells the agent: "when the user says X, invoke Y."

Source for the 100% claim: Vercel's agent evals post. The AGENTS.md pattern is also the basis for Anthropic's own Claude Code setup.

Why curation beats quantity

The SkillsBench paper (arxiv.org/abs/2602.12670) tested agent performance across three skill installation strategies.

The mechanism: irrelevant skills confuse the agent about what to use. A kitchen-sink install forces the agent to evaluate 200+ options per task, increasing both latency and error rate. The sweet spot is 2-3 well-matched skills per task, which is exactly what profile-based curation produces.

This is why the pack uses profile tags. When you label yourself as a developer, you get the dev skills in your activation block, not every skill in the catalog. The agent doesn't know about skills outside your selected profiles unless it invokes find-skills specifically.

What Floom adds to skills.sh skills

Floom doesn't author most of the skills in this pack. We curate them from the open ecosystem (mostly skills.sh) and add five things on top.

When you install via Floom, you're getting the original author's skill, with the activation, distribution, and update layer Floom provides on top. We link every skill to its source repo. Floom doesn't intermediate the code.

Why Floom doesn't pay for inference

Skills run inside your agent, on your API key, using your token budget. Floom is the distribution layer. We curate, package, and activate. Your agent executes.

This is intentional. Skills that run on a third-party runtime can't access your local filesystem, your git repo, or your project context. Skills that run in your agent can do all three. The power of skills is that they run where your agent runs.

Floom's business is the distribution infrastructure: the manifest, the installer, the activation pattern, and the curation. Not the inference.