TechTide Harness Kit

TechTide Harness Kit

The enterprise skill platform for AI coding agents.
385 production skills. 348 specialist agents. 6 cloud providers. 7 harnesses. One repo.

Get Started · Harnesses · Cloud Coverage · Trust · Anatomy · Full Catalog · Contribute

Not an awesome-list. Every skill ships with YAML-frontmatter tool scoping, inline assessment questions, safety checklists with blast-radius checks, and 9 JSON Schema contracts enforcing shape at CI time. Built for teams that answer to auditors, not just developers.

Why this exists

What it does. Ships 385 skills and 348 specialist agents as a schema-validated catalog (skills/, agents/, catalog/, schemas/), exporters that package them for Claude Code, Codex, Cursor, Copilot, Gemini CLI, Kiro, and Lovable (scripts/export-marketplace-agents.mjs, plugins/, powers/, .claude-plugin/), and a npm run validate chain of 25+ gates that checks frontmatter, tool scopes, link syntax, asset integrity hashes, MCP trust matrices, and plugin manifests before the package is published to npm as @techtideai/harness-kit.

Who it serves. Platform, cloud, and security engineers who want agent skills with declared tool permissions, approval gates, and provenance they can show an auditor, and maintainers who need one source tree exported to several agent harnesses without copying files by hand.

Why Alex built it. The repository opened on 2026-05-18 as a "showcase release" (bbc5aad). AGENTS.md states the purpose: a "cloud and zero-trust agentic marketplace for skills, agents, rules, MCP references, and npm distribution" that optimizes for "evidence-backed security workflows: least privilege, source grounding, manifests, validation, and safe automation". TRUST.md records the reasoning behind the gates: "skills are treated as behavior packages that can influence tool use, so every public asset needs source evidence, scoped permissions, security notes, and human accountability." Four days in, a dedup pass removed 96 harness-clone skills and added a hash-collision CI gate (f93e8bb, 2026-05-22: "487 -> 391 unique production skills"), and the Agent Plugins 1.0 distribution landed on 2026-09-05 (894952c). Built by Alex Cinovoj, Founder & CTO of TechTide AI.

Real-world use case (2025-2026 evidence). Verified in this repository. A team adopts Agent Skills, the folder-plus-SKILL.md format Anthropic introduced on 2025-10-16[^1] that is now an open specification with required name and description frontmatter and an optional allowed-tools field[^2], and installs this catalog into Claude Code through a plugin marketplace, the distribution mechanism Anthropic shipped on 2025-10-09[^3]. Before every release npm run validate fails if any skill's allowed-tools, metadata, or integrity hash drifts, and the OpenSSF Scorecard workflow (.github/workflows/scorecard.yml; Scorecard v5.6.0 was released 2026-10-12[^4]) publishes the supply-chain posture reported by the badge above. The install paths and the packed npm artifact are exercised on every push by .github/workflows/install-paths-smoke.yml and .github/workflows/packed-artifact-smoke.yml, which is the evidence behind the verified label.

[^1]: https://github.com/techtideohio/techtide-harness-kit/blob/HEAD/Anthropic, "Introducing Agent Skills", https://claude.com/resources/articles/skills, published 2025-10-16. [^2]: https://github.com/techtideohio/techtide-harness-kit/blob/HEAD/Agent Skills, "Specification", https://agentskills.io/specification, updated 2026-08-04. [^3]: https://github.com/techtideohio/techtide-harness-kit/blob/HEAD/Anthropic, "Customize Claude Code with plugins", https://claude.com/resources/articles/claude-code-plugins, published 2025-10-09. [^4]: https://github.com/techtideohio/techtide-harness-kit/blob/HEAD/ossf/scorecard, release v5.6.0, https://github.com/ossf/scorecard/releases/tag/v5.6.0, published 2026-10-12.

 Works With

/plugin marketplace add
  TechTideOhio/techtide-harness-kit

git clone TechTideOhio/
  techtide-harness-kit
# Add as plugin directory

codex plugin marketplace add
  TechTideOhio/techtide-harness-kit

npm run lovable:write
# Upload ZIPs to Lovable workspace

 Get Started

npm install @techtideai/harness-kit

# Validate everything (25+ gates, no secrets required)
npm run validate

# Export agents for your harness
npx thk-export-agents --platform claude-code --provider aws --repo .
npx thk-export-agents --platform codex --role cloud-security-engineer --repo .
npx thk-export-agents --platform gemini --all --repo .

 By The Numbers

 Cloud Coverage

Every cloud has deep, provider-specific skills with inline assessment questions and validation checklists -- not generic templates with names swapped.

 Skill Anatomy

Every skill follows a consistent, agent-optimized structure with least-privilege tool scoping:

skills/<provider>/<skill-id>/
  SKILL.md            # YAML frontmatter + workflow + assessment questions + checklist
  metadata.json       # Schema-validated: id, version, harnesses, official_docs
  references/         # Safety checklists, official sources, extended guidance

---
name: techtide-aws-waf-reliability-review
description: "Review AWS workload reliability..."
allowed-tools: Read Grep Glob
metadata:
  author: "github: TechTide"
  version: "0.1.0"
  category: resilience
---

## Assessment Question Bank
### Compute and Autoscaling
1. Are all stateless tiers behind autoscaling?
2. Is scale-in protection set for in-flight?

## Validation Checklist
- [ ] Multi-AZ for all production compute
- [ ] Autoscaling configured for stateless
- [ ] Backup restore tested in last 30 days
- [ ] Circuit breakers on external calls

 What Ships

 Enterprise Trust

 Security

Report vulnerabilities through SECURITY.md. The findings below were open on the repository's Dependabot and code-scanning dashboards as of 2026-10-20. They are documented here rather than dismissed.

 vs Alternatives

 Common Commands

# Validate everything (25+ gates)
npm run validate

# Check trust + proof layer
npm run trust:check && npm run proof-layer:check

# Export agents
npm run agents:export -- --list
npm run agents:export -- --platform claude-code --provider aws --repo .
npm run agents:export -- --platform codex --all --repo .

# Regenerate artifacts
npm run proof-layer:write && npm run plugin-manifest:write

 Documentation