souvik-skills
AGENTS.md1 starsChanged 5 months agoMITView on GitHub
This is the agent guide for Souvik Skills. Before creating or changing a skill, read guidance.md and NAMING.md. Do not create context.md.
- Commits and pushes
Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.
Of the AGENTS.md files we read, the 31,700 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.
In this repository19 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- agy-skill-installerSkill
- apex-bug-hunterSkill
- arsSkill
- browser-triageSkill
- claude-md-initSkill
- delink-githubSkill
- gemini-skill-packagerSkill
- gh-readySkill
- indian-legal-verificationSkill
- local-systematic-verification-ocrSkill
- magento-team-leadSkill
- merge-gateSkill
- motioncraftSkill
- pdf-vector-engineSkill
- presscraftSkill
- score2mdSkill
- skill-shipSkill
- wpcraftSkill
The file
Show its rules and text
We split this AGENTS.md into the rules it gives an agent
44 rules. About 1 is common, 20 are distinctive, and 23 are very specific to this project.
≈1
Common · estimate
Many files like this have a rule of this kind.
≈20
Distinctive · estimate
This author's own choices. Worth a look.
23
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 5
Before creating or changing a skill, read guidance.md and NAMING.md.
- Distinctive: line 20
Keep every skill folder self-contained and installable.
- Very specific: line 21
Keep SKILL.md concise and procedural.
- Very specific: line 22
Use only name and description in SKILL.md frontmatter.
- Very specific: line 24
Put detailed optional material in references/ and mention when to read it.
- Very specific: line 25
Put deterministic automation in scripts/.
- Very specific: line 26
Put reusable output assets in assets/.
- Very specific: line 27
Keep agents/openai.yaml aligned with SKILL.md.
- Very specific: line 28
Ensure interface.default_prompt mentions the skill as $skill-name.
- Very specific: line 29
Keep the README catalog and cross-agent NPX install matrix aligned with the skills/ directory.
- Distinctive: line 30
After creating or updating a skill for local Codex use, reinstall the updated skill into Codex and remind the user to restart Codex.
- Common: line 31
Do not add private traces, generated caches, local logs, or secrets.
- Distinctive: line 32
Do not commit changes unless the user explicitly asks.
- Distinctive: line 33
When the user explicitly asks to sync GitHub, update NPM/NPX, publish, release, or make the change globally available, follow the GitHub …
- Very specific: line 38
Before changing frontend routes, layouts, cards, install modules, copy buttons, logo assets, typography, animation, theme behavior, or …
- Very specific: line 39
Update DESIGN.md in the same change when the frontend visual language, typography, layout rules, color system, component styling, or …
- Very specific: line 40
Update MOTION.md in the same change when animation tokens, interaction feedback, route transitions, copy feedback, theme motion, …
- Distinctive: line 42
Do not hard-code current public skill names as the source of truth for site routing or catalog generation.
- Very specific: line 43
After adding, removing, renaming, publishing, or materially updating any public skill, inspect and update the marketplace frontend where …
- Very specific: line 44
Keep frontend catalog data derived from skills/, SKILL.md, and agents/openai.yaml.
- Very specific: line 46
Keep the install UI aligned with the Project, Global, and Try once command structure in guidance.md and README.md.
- Distinctive: line 47
Verify zero horizontal overflow at 360px, 390px, 430px, tablet, desktop, and wide desktop before finishing frontend work.
- Distinctive: line 48
After any public skill is added, removed, renamed, published, or materially updated, deploy the rebuilt marketplace to both production …
- Very specific: line 52
Run npm run validate:skills and npm run build after landing page or skill catalog updates.
- Distinctive: line 59
Add only resources the skill actually needs.
- Very specific: line 60
Update README.md with the skill catalog row, project install, global install, try-once command, and example $skill-name prompt.
- Distinctive: line 61
Check the frontend marketplace data and pages.
- Very specific: line 68
Run npm run build when the new skill changes the public marketplace output.
- Distinctive: line 70
Add tests or examples when the skill includes scripts or deterministic behavior.
- Distinctive: line 74
When changing a skill:
- Distinctive: line 77
Update README.md if the public catalog, install commands, or usage prompts change.
- Distinctive: line 78
Update the frontend helpers or pages if the skill update changes categories, resources, commands, public copy, routing, or install behavior.
- Very specific: line 79
Update package.json only when package metadata or CLI behavior changes.
- Distinctive: line 82
Run validation before finishing.
- Distinctive: line 86
Use this protocol when the user explicitly asks to commit, push, sync GitHub, publish, release, update NPM, update NPX, or make a skill …
- Distinctive: line 90
Keep the worktree scoped. Do not stage unrelated user files or untracked skill folders unless they are part of the request.
- Distinctive: line 91
Run the relevant checks before publishing:
- Distinctive: line 100
Run any relevant skill-specific audit or script, such as:
- Distinctive: line 107
If the change is meant for immediate local Codex use, reinstall the changed skill and remind the user to restart Codex:
- Distinctive: line 113
Create a signed Conventional Commit when signing is configured, verify the local signature, then push to main.
- Very specific: line 121
After pushing, open or inspect the GitHub commit and confirm GitHub marks it as verified.
- Very specific: line 123
When the user has asked for NPM/NPX/package sync, merge the clean Release Please PR after checks pass.
- Very specific: line 137
If the package changed but Release Please does not open a release PR, check the commit type.
- Very specific: line 141
Do not commit NPM, GitHub, Vercel, Cloudflare, or other secrets.
≈ “Common” is an estimate: at least 1 in 5 of the 40,246 AGENTS.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.
Files
AGENTS
This is the agent guide for Souvik Skills.
Before creating or changing a skill, read guidance.md and NAMING.md. Do not create context.md.
Repository Facts
- Brand: Souvik Skills
- Owner: Souvik Dey
- GitHub owner:
imsovikde - Repository:
imsovikde/souvik-skills - NPM package:
@imsovikde/skills - License: MIT
- Current public skill names are derived from the
skills/directory, not hard-coded in this guide.
Agent Rules
- Before creating, updating, publishing, or deploying a public skill, read this entire
AGENTS.md, then read the repository Markdown guidance that applies to the change: root-level*.md,docs/**/*.md, the changed skill'sSKILL.md, and any changed or referenced skillreferences/**/*.mdorassets/**/*.mdtemplates. For broad catalog or deployment work, also inspect all publicskills/**/SKILL.mdfiles so the catalog remains coherent. - Keep every skill folder self-contained and installable.
- Keep
SKILL.mdconcise and procedural. - Use only
nameanddescriptioninSKILL.mdfrontmatter. - Make the skill folder name exactly match the frontmatter
name. - Put detailed optional material in
references/and mention when to read it. - Put deterministic automation in
scripts/. - Put reusable output assets in
assets/. - Keep
agents/openai.yamlaligned withSKILL.md. - Ensure
interface.default_promptmentions the skill as$skill-name. - Keep the README catalog and cross-agent NPX install matrix aligned with the
skills/directory. - After creating or updating a skill for local Codex use, reinstall the updated skill into Codex and remind the user to restart Codex.
- Do not add private traces, generated caches, local logs, or secrets.
- Do not commit changes unless the user explicitly asks.
- When the user explicitly asks to sync GitHub, update NPM/NPX, publish, release, or make the change globally available, follow the GitHub And NPM Sync Protocol end to end instead of stopping after local edits.
- Every repository commit must be verified before reporting completion: create a signed commit when signing is configured, confirm
git log -1 --show-signaturelocally, confirm the GitHub commit page shows a verified signature after push, and confirm the pushed branch contains the intended tree. Do not substitute unsigned API commits for signed local commits when the user asks for verified commits.
Landing Page Maintenance
- Before changing frontend routes, layouts, cards, install modules, copy buttons, logo assets, typography, animation, theme behavior, or responsive behavior, read
DESIGN.mdandMOTION.mdcompletely and treat them as the implementation contract. If a user saysnotion.mdwhile discussing site motion, read and updateMOTION.md. - Update
DESIGN.mdin the same change when the frontend visual language, typography, layout rules, color system, component styling, or responsive behavior materially changes. - Update
MOTION.mdin the same change when animation tokens, interaction feedback, route transitions, copy feedback, theme motion, reduced-motion behavior, or motion QA rules materially change. - The Next.js marketplace derives public skill pages from the
skills/directory at build time. - Do not hard-code current public skill names as the source of truth for site routing or catalog generation.
- After adding, removing, renaming, publishing, or materially updating any public skill, inspect and update the marketplace frontend where needed so
/,/skills,/skills/<skill-name>,/install, and/docsreflect the currentskills/content and install model. - Keep frontend catalog data derived from
skills/,SKILL.md, andagents/openai.yaml; add or adjust shared inference/helpers when a new skill domain, resource shape, or install surface is introduced. - Preserve independent pages for
/,/skills,/skills/<skill-name>,/install,/docs, and/motion. - Keep the install UI aligned with the Project, Global, and Try once command structure in
guidance.mdandREADME.md. - Verify zero horizontal overflow at 360px, 390px, 430px, tablet, desktop, and wide desktop before finishing frontend work.
- After any public skill is added, removed, renamed, published, or materially updated, deploy the rebuilt marketplace to both production targets and verify the changed skill appears on Vercel and Cloudflare Pages:
- Vercel:
https://souvik-skills.vercel.app - Cloudflare Pages:
https://souvik-skills.pages.dev - Required routes:
/,/skills,/skills/<skill-name>,/install, and/docs.
- Vercel:
- Run
npm run validate:skillsandnpm run buildafter landing page or skill catalog updates.
Creating A New Skill
- Choose a name using
NAMING.md. - Create
skills/<skill-name>/SKILL.md. - Create
skills/<skill-name>/agents/openai.yaml. - Add only resources the skill actually needs.
- Update
README.mdwith the skill catalog row, project install, global install, try-once command, and example$skill-nameprompt. - Check the frontend marketplace data and pages. Update shared frontend helpers when the new skill adds a category, nested resources, install behavior, or public content shape not already represented.
- Run:
npm run validate:skills- Run
npm run buildwhen the new skill changes the public marketplace output. - Deploy and verify both hosted sites when the skill is public: Vercel and Cloudflare Pages must show the new skill before the task is done.
- Add tests or examples when the skill includes scripts or deterministic behavior.
Updating Existing Skills
When changing a skill:
- Preserve its public install name unless the user explicitly approves a rename.
- Update
README.mdif the public catalog, install commands, or usage prompts change. - Update the frontend helpers or pages if the skill update changes categories, resources, commands, public copy, routing, or install behavior.
- Update
package.jsononly when package metadata or CLI behavior changes. - Reinstall the updated skill into Codex when the change is meant for immediate local use.
- Redeploy and verify both hosted sites when the update changes public skill content or marketplace pages.
- Run validation before finishing.
GitHub And NPM Sync Protocol
Use this protocol when the user explicitly asks to commit, push, sync GitHub, publish, release, update NPM, update NPX, or make a skill globally available.
This is a professional release workflow, not a memory-only habit. Future agents must use it whenever package sync is requested so GitHub, GitHub Releases, NPM, GitHub Packages, NPX install paths, and the local Codex copy stay aligned.
- Keep the worktree scoped. Do not stage unrelated user files or untracked skill folders unless they are part of the request.
- Run the relevant checks before publishing:
npm run validate:skills
npm test
npm run build
npm pack --dry-run- Run any relevant skill-specific audit or script, such as:
npm run audit:gh-ready
npm run audit:motioncraft- If the change is meant for immediate local Codex use, reinstall the changed skill and remind the user to restart Codex:
node bin/souvik-skills.cjs install <skill-name> --force- Create a signed Conventional Commit when signing is configured, verify the local signature, then push to
main. Example:
git commit -m "feat: upgrade <skill-name> ..."
git log -1 --show-signature
git push origin main- After pushing, open or inspect the GitHub commit and confirm GitHub marks it as verified. Then verify GitHub Actions for
CI,CodeQL, andRelease Please. - Release Please opens or updates a release PR. Inspect it before merging. It should only contain release metadata such as
package.json,package-lock.json,.release-please-manifest.json, andCHANGELOG.md. - When the user has asked for NPM/NPX/package sync, merge the clean Release Please PR after checks pass. Resolve generated changelog conflicts carefully and keep one clean release entry.
- Confirm the release automation publishes successfully to NPM and GitHub Packages. Verify with:
npm view @imsovikde/skills version
npx -y @imsovikde/skills@latest list- Smoke-test the published package when a skill changed:
npx -y @imsovikde/skills@latest install <skill-name> --dest <temp-dir> --force- If the package changed but Release Please does not open a release PR, check the commit type. Use a release-triggering Conventional Commit such as
feat:orfix:when the package contents must reach NPM/NPX. - Deploy the current
mainbuild to Vercel and Cloudflare Pages when public frontend or skill catalog output changed. Manually verify both hosted domains include the changed skill on/,/skills,/skills/<skill-name>,/install, and/docs; use a browser or HTTP fetch against the live domains, not only local build output. - Finish only after reporting the signed GitHub commit and verification status, release PR or release URL, NPM version, package publish status, Vercel deployment status, Cloudflare Pages deployment status, local Codex reinstall status, and any check that could not be run.
Do not commit NPM, GitHub, Vercel, Cloudflare, or other secrets. Use the existing GitHub Actions release automation; manual npm publish is only a fallback when the user explicitly asks and automation cannot complete.
Release Readiness
A change is release-ready only when:
- Skill validation passes.
- Relevant tests pass.
- Install commands still work.
- Documentation reflects the current skill names.
- Vercel and Cloudflare Pages show the current public skill catalog when frontend or skill content changed.
- License and ownership information remain intact.
- Release Please and package publishing are verified when the user requested NPM/NPX sync.
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
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.

