Skip to content

Srinitude Skills - Setup Guide

Source: srinitude/skills via skills.sh - 10.6K combined installs across 23 indexed listings; first seen Oct 9, 2026 (extended-query probe) GitHub: srinitude/skills (2 stars, MIT; pushed 2026-09-30; layout skills/<name>/SKILL.md with root plugin.json and mcp.json) Category: Skill Infrastructure / Agent Tooling Quality Tier: 🟡 Beta - Agent Plugins v1.0.0 package; MIT; evaluation-driven; mixed sampled verdicts - see Security

srinitude publishes 23 portable skills as a single Agent Plugins v1.0.0 package: one canonical skills tree, root plugin.json and mcp.json manifests, a bundled read-only local MCP server, and trigger, behavior, failure, recovery, and speed evaluation suites behind the skills. The catalog is small but almost every listing is heavily used: 14 of the 23 sit at 500+ installs, led by starting-point (714), reify (691), visual-design-system-extractor (682), and skill-factory (681).

The skills cluster into three groups. Agent-behavior gates sharpen how an agent works: starting-point checks whether a prescribed method can actually reach the user's outcome, outcome-bounded-work separates outcomes from recipes, logic-audit hunts contradictions, would-humans-actually and would-agents-actually test action-dependent claims, timebox bounds effort, and meaning-preserving-rewrite protects meaning during edits. A skill-authoring cluster (skill-factory, simplify-skill, dedupe, goal-prompt) covers building and maintaining skills themselves. A design and utility group spans visual-design-system-extractor, always-current-datetime (558), and the design skills in the tail. MIT licensed, though the repository itself is young (first indexed Oct 9, 2026).

Installation

Prerequisites: Node.js for the npx skills CLI; local development of the package targets Node 24+ and a mise toolchain.

# Inspect first, then install
npx skills add srinitude/skills --list
npx skills add srinitude/skills

Claude Code plugin route:

/plugin marketplace add srinitude/skills
/plugin install srinitude-skills@srinitude-skills

Gemini CLI route:

gemini extensions install https://github.com/srinitude/skills

The README documents per-client routes (Codex, ChatGPT, Cursor, opencode, Continue, Aider, and a Hermes Agent adapter). Note: the CLI reports anonymous install telemetry to skills.sh unless DISABLE_TELEMETRY=1 is set.

What It Provides

Install counts from the Oct 10, 2026 skills.sh snapshot:

Skill Installs Use For
starting-point 714 Start from the user's observable end state: pick the starting boundary and the completion proof before running a prescribed method
reify 691 Turn a vague idea, stray thought, or uncertain direction into a concrete outcome, tested design, decision record, or executable handoff
visual-design-system-extractor 682 Reverse-engineer reference images, screenshots, moodboards, or a live site URL into design tokens, art direction, motion rules, and a YAML style spec
skill-factory 681 Turn a workflow into a new agent skill: scaffolding, validation, evaluation, scripts, tests, and user-level or project-level variants
always-current-datetime 558 Refresh the current date and time when replying
outcome-bounded-work 555 Separate outcomes from recipes when instructions mix them
meaning-preserving-rewrite 545 Rewrite rules and instructions without losing meaning
timebox 544 Keep work inside a stated time limit
logic-audit 544 Find contradictions and reasoning gaps
would-humans-actually 543 Pressure-test claims that depend on people taking a real action
dedupe 541 Deduplicate bounded collections
simplify-skill 540 Simplify a skill without losing its behavior
would-agents-actually 536 Pressure-test claims that depend on an agent taking a real action
goal-prompt 534 Package source input for a standing goal

The remaining 9 indexed listings range from 1 to 315 installs, led by mobile-first-website-design (315), prompt-enhancer (313), and tool-call-configuration-for (306), and ending at figma-code-connect-design-system with a single install.

Why This Matters for Hermes Agents

Skills are only as good as their stopping conditions, and this package is built around exactly that failure mode: each skill encodes when it applies, when it does not, and what evidence counts as done. The evaluation discipline is the rare part: the repository ships trigger, behavior, failure, recovery, and speed suites behind the skills, though the README is precise that fixture results prove runner behavior only, not model behavior. The bundled MCP server is read-only by design (list_skills, search_skills, get_skill, get_reference, get_eval_manifest, validate_skill), confines paths to the repository skill tree, and has no write tool, telemetry, credentials, or network calls. For Hermes agents the value is reliability tooling: gates like starting-point and would-humans-actually catch wrong-metric and unproven-claim work early, while skill-factory turns repeated workflows into maintainable skills. One caveat: the sampled verdicts are mixed (a Socket Warn on starting-point; Socket and Snyk Warns on visual-design-system-extractor).

Usage

You say What happens
"A/B test the signup button color" starting-point checks whether the prescribed method reaches the real outcome (conversion) before doing the request as written
"Turn this rough idea into something concrete" reify converts the fragment into an outcome, a tested design, a decision record, or an executable handoff
"Extract a design system from these screenshots" visual-design-system-extractor produces design tokens, art direction, and a YAML style spec
"Make a skill for this workflow" skill-factory scaffolds, validates, and evaluates the new skill, including variants
"Rewrite these rules more clearly, same meaning" meaning-preserving-rewrite guards against meaning loss during the edit
"Will people actually use this feature?" would-humans-actually tests the people-action assumption; would-agents-actually covers agent actions
"Keep this investigation inside 30 minutes" timebox bounds the work and forces a stopping point

Verification

# Confirm installed (paths depend on your agent)
npx skills list | grep -i srinitude

# Review the flagship skill straight from GitHub before installing (returns 200)
curl -s https://raw.githubusercontent.com/srinitude/skills/main/skills/starting-point/SKILL.md | head -20

Package layout check: skills/<name>/SKILL.md for each skill, plus root plugin.json and mcp.json for Agent Plugins v1 clients. For MCP-capable clients, build the bundled server with mise run build-mcp from a checkout.

Security

skills.sh verdicts for sampled skills (verified Oct 10, 2026) - re-check the security pages on skills.sh before production use:

Skill Gen Agent Trust Hub Socket Snyk
starting-point Pass Warn Pass
design-system-extractor - - -
visual-design-system-extractor Pass Warn Warn

Mixed results: both skills with full verdicts carry a Socket Warn, and visual-design-system-extractor also carries a Snyk Warn. The bundled MCP server's design is explicitly read-only with paths confined to the skill tree, but the third-party verdicts above are what matter for supply-chain review.

Integration with CorpusIQ

CorpusIQ is the connector layer that lets AI assistants read the systems a business already runs on - accounting, payments, CRM, email, files and analytics - and answer with the source named. It is read-only by design: it retrieves data and never writes to, or acts in, a connected system, which is what makes it safe to point at live accounts.

This page is part of the CorpusIQ directory, the reference we maintain for the MCP server and agent-skill ecosystem. Run alongside CorpusIQ, a third-party server like this one can be combined with a company's own business data in a single answer.

See what CorpusIQ connects

Limitations

  • Mixed sampled verdicts: starting-point (Socket Warn) and visual-design-system-extractor (Socket Warn + Snyk Warn).
  • Small repository: 2 GitHub stars, independent publisher, last pushed Sep 30, 2026.
  • Eval honesty: fixture results prove runner behavior only; the README is explicit that they are not evidence about a language model.
  • Local development workflow assumes Node 24+ and mise (mise run bootstrap, mise run ci); a plain skills CLI install needs less.
  • The skills CLI reports anonymous install telemetry to skills.sh unless DISABLE_TELEMETRY=1 is set.
  • Snapshot data, verified Oct 10, 2026: 10.6K combined installs across 23 indexed listings; 2 GitHub stars; MIT; last pushed Sep 30, 2026. Counts drift over time.

Pro Tips

  • Inspect before installing with npx skills add srinitude/skills --list; the catalog is small enough to read in one pass.
  • The package is deliberately single-source: every client route loads the same canonical SKILL.md files, so behavior stays consistent across Claude Code, Codex, Gemini CLI, Cursor, and Hermes.
  • The bundled MCP server is for discovery and validation only (six read-only tools); it never writes and makes no network calls.
  • Skills declare metadata.scope: "user" so they are available across projects by default; project scope exists as a variant.
  • Run the local gate (mise run validate, mise run eval, mise run ci) when authoring or modifying skills from a checkout.

← Skills Catalog | Skills Marketplace → Powered by CorpusIQ

AI answers on your data. See where every number comes from.

Connect your tools and ask in ChatGPT, Claude, or Perplexity. Enterprise-grade answers in minutes. From $29.95/month.

Start free, no credit card
Contact Terms Privacy Pricing