Agentsmith

Agent rules and reusable skills from my daily setup, with a separate workspace for plugin packaging.

Install

Install selected skills:

npx skills add Sarthib7/agentsmith

Install everything globally:

npx skills add Sarthib7/agentsmith --all -g

Read one skill without installing it:

npx skills use Sarthib7/agentsmith@caveman

Browse the collection on Skills.sh or open the full generated skill catalog.

Changes and new skills: CONTRIBUTING.md.

Start here

Read in this order:

1. rules/ — the standing instructions every session runs under. Everything else assumes them.
2. my-skills/ — the skills written here, then the full collection via the skill catalog.
3. workflows/ — the operating workflows the rules link to.

Repository layout

agentsmith/
├── AGENTS.md          rules for contributing to this repository
├── CONTRIBUTING.md    contributor workflow and validation steps
├── rules/             global AGENTS.md snapshot and harness config examples
├── workflows/         reusable operating workflows linked by agent rules
├── skills/            collected skills used in my setup
├── my-skills/         skills with verified sarthib7 authorship
├── plugin/            agents, commands, hooks, and manifests
├── scripts/           catalog generation and validation
├── data/              shared skill reference data
├── SKILLS.md          generated skill catalog
└── skills.sh.json     generated skills.sh grouping manifest

Ownership and purpose stay separate. skills/ answers what I use. my-skills/ answers what I wrote. ATTRIBUTION.md records known authors, licenses, and source gaps.

My skills

These eight skills declare author: sarthib7 in their source:

Skill	Purpose
adhd	Parallel divergent ideation for coding agents
bench-it	Compare performance claims against a named baseline
deterministic-code-review	Review diffs with frozen scope, exact anchors, and a falsification pass
followup-review	Review fixes made after an earlier code review
fresh-eyes	Get an independent second opinion without shared context
git-worktree-runner	Run isolated agent tasks through git gtr worktrees
prove-it	Verify that a check passes and can fail
research-url	Research a URL with one subagent per URL into a fixed-format report plus learnings

Authorship stays conservative. A skill remains in skills/ when its origin is uncertain.

How the pieces fit

rules/ contains standing instructions loaded for every session. One AGENTS.md snapshot serves every harness: Oh My Pi and Codex read AGENTS.md natively, and Claude Code reads the same policy from its live CLAUDE.md source. workflows/orchestration.md records how delegation and parallel agents run on top of those rules.

skills/ and my-skills/ contain task procedures. Each skill has a SKILL.md entry point and may include references, scripts, templates, or assets.

plugin/ is the packaging workspace. Its capability split follows the Vercel plugin: specialist agents, commands, hooks, and platform manifests remain separate from skill source. The plugin is a scaffold today and has no published manifest.

DYOR

Treat every skill as untrusted until you review it. Inclusion in this collection is not a security review or endorsement.

Before installing a skill:

• Read its SKILL.md, scripts, hooks, and requested permissions.
• Check its source and license in ATTRIBUTION.md.
• Record the commit SHA you reviewed when reproducibility matters.
• Test unfamiliar skills in an isolated repository or worktree.

Known limitations

• Some Solana and DeFi skills reference ../../../data/.... Those paths depend on the installation layout.
• skills/spec-build/ declares name: build because the skills CLI treats a directory named build as generated output.
• skills/tempo-request/ declares name: tempo, matching its upstream source.
• Provenance is incomplete for skills that arrived without author metadata, a license, or a recorded source.

Generated files should not be edited by hand. Follow CONTRIBUTING.md when changing the collection.