agentleFS
imsovikdeimsovikde/souvik-skills/CLAUDE.md

souvik-skills

CLAUDE.md1 starsChanged 5 months agoMITView on GitHub

Concise guide for Claude (and any AI agent) working in this repository. For deep detail, defer to the files named below instead of duplicating them here. - Brand: Souvik Skills — owner Souvik Dey (GitHub imsovikde). - Repo imsovikde/souvik-skills; npm @imsovikde/skills; MIT. - Site: https://souvik-skills.vercel.app (mirror: https://souvik-skills.pages.dev). - Skills (each skills/<name>/ = SKILL.md + agents/openai.yaml + optional references/, scripts/, assets/): claude-md-init, delink-github, gh-ready, magento-team-lead, motioncraft, score2md. - Static Next.js (output: export) marketplace built from skills/ at build time via lib/skills.js…

The file

We split this CLAUDE.md into the rules it gives an agent

9 rules. About 0 are common, 3 are distinctive, and 6 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 6

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Do not create context.md.

    line 27
  2. Distinctive:

    Every commit in this repo — from any environment (Claude Code on the web / cloud, local, or CI) — must satisfy all of the following:

    line 31
  3. Very specific:

    Author is only Souvik: Use Souvik Dey <imsovikde@gmail.com> (the GitHub-verified email).

    line 34
  4. Very specific:

    No AI attribution: Never add Co-Authored-By: trailers for Claude or any AI tool, and never add Claude-Session: or other assistant-identity …

    line 42
  5. Distinctive:

    Every commit that lands on main must show GitHub's green Verified badge.

    line 49
  6. Distinctive:

    Do not substitute an unsigned API/direct commit on main when a verified commit was requested.

    line 59
  7. Very specific:

    Never point a marketplace entry at the repo root ("./") and never add a root .claude-plugin/plugin.json — the root holds the Next.js site, …

    line 71
  8. Very specific:

    When a skill is added, removed, or renamed, update .claude-plugin/marketplace.json, codex-plugin.json, README.md, package.json, and …

    line 81
  9. Very specific:

    Run the repo's own checks: npm run validate:skills, npm test, npm run build, and npm pack --dry-run when packaging matters.

    line 90

≈ “Common” is an estimate: at least 1 in 5 of the 26,980 CLAUDE.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
CLAUDE.md

CLAUDE.md

Concise guide for Claude (and any AI agent) working in this repository. For deep detail, defer to the files named below instead of duplicating them here.

Repository Snapshot

  • Brand: Souvik Skills — owner Souvik Dey (GitHub imsovikde).
  • Repo imsovikde/souvik-skills; npm @imsovikde/skills; MIT.
  • Site: https://souvik-skills.vercel.app (mirror: https://souvik-skills.pages.dev).
  • Skills (each skills/<name>/ = SKILL.md + agents/openai.yaml + optional references/, scripts/, assets/): claude-md-init, delink-github, gh-ready, magento-team-lead, motioncraft, score2md.
  • Static Next.js (output: export) marketplace built from skills/ at build time via lib/skills.js — never hard-code the skill list; it is derived from the skills/ directory.
  • Automation: release-please (versioning/changelog) → npm-publish.yml (npm + GitHub Packages with provenance + OIDC trusted publishing) → CodeQL + Dependabot.

Read Before Working

Load these as they apply to the change, rather than guessing:

  • AGENTS.md — how to create/maintain skills, the GitHub/NPM sync protocol, release readiness.
  • guidance.md — skill directory rules and install standards. NAMING.md — naming rules.
  • DESIGN.md and MOTION.md — frontend/motion contracts (before any site change).

Do not create context.md.

Commit Identity & Attribution (hard rule)

Every commit in this repo — from any environment (Claude Code on the web / cloud, local, or CI) — must satisfy all of the following:

  1. Author is only Souvik. Use Souvik Dey <imsovikde@gmail.com> (the GitHub-verified email). Set it at the start of every session:

    git config user.name "Souvik Dey"
    git config user.email "imsovikde@gmail.com"
  2. No AI attribution. Never add Co-Authored-By: trailers for Claude or any AI tool, and never add Claude-Session: or other assistant-identity trailers to commit messages. Only Souvik's name may appear on the GitHub contribution graph. If the harness would append such trailers by default, strip them and commit a clean message.

Verified Commits (hard rule)

Every commit that lands on main must show GitHub's green Verified badge.

  • From cloud (Claude Code on the web): the container holds no private signing key, so a direct cloud commit cannot be cryptographically signed. Therefore: commit to a feature branch with Souvik's identity and a clean message, push, then merge through the GitHub web UI. GitHub signs the resulting merge/squash commit with its own key → the landed commit is Verified and attributed to Souvik. This is the required path for cloud work.
  • From local: sign with Souvik's own GPG or SSH signing key (git config commit.gpgsign true), confirm git log -1 --show-signature, and confirm the GitHub commit page shows Verified after push. See docs/local-commit-setup.md for the exact one-time setup commands.
  • Do not substitute an unsigned API/direct commit on main when a verified commit was requested.
  • To apply this same CLAUDE.md pattern (repo snapshot + these two hard rules) to another project, use the claude-md-init skill — it scaffolds or updates that project's CLAUDE.md idempotently.

Marketplace Surfaces (keep in sync when skills change)

  • Claude Code / Cowork: .claude-plugin/marketplace.json — one entry per skill, each sourced at its own ./skills/<skill-name>, and each of those folders carries its own skills/<skill-name>/.claude-plugin/plugin.json. This mirrors Anthropic's own Cowork marketplace (anthropics/knowledge-work-plugins), where every plugin source directory is self-describing. Install: /plugin marketplace add imsovikde/souvik-skills then /plugin install <skill>@souvik-skills. Never point a marketplace entry at the repo root ("./") and never add a root .claude-plugin/plugin.json — the root holds the Next.js site, docs, and tooling, so it is not a plugin directory. npm run validate:skills fails if either is reintroduced, if a per-skill entry's source isn't its own ./skills/<skill-name>, or if a skill's plugin.json name/version/ description drifts from its marketplace entry.
  • OpenAI Codex: per-skill agents/openai.yaml + root codex-plugin.json. Package install: npx @imsovikde/skills install <skill-name>.
  • npm / npx: @imsovikde/skills (published with provenance via GitHub Actions).
  • Web: the Vercel/Cloudflare site, auto-derived from skills/.

When a skill is added, removed, or renamed, update .claude-plugin/marketplace.json, codex-plugin.json, README.md, package.json, and release-please-config.json together — npm run validate:skills enforces that the manifests match the skills/ directory and the package version. Every manifest version is bumped at release time by release-please via extra-files, so a new skill needs its two extra-files entries (its marketplace.json plugin index and its own plugin.json) or the release will fail validation on main.

Before Finishing

Run the repo's own checks: npm run validate:skills, npm test, npm run build, and npm pack --dry-run when packaging matters. Report what ran and what could not.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.