claudecodex

vendor-harness

Lifecycle management for LLM vendor harness artifacts — validate, diagnose, feedback, and keep current across Claude Code, Cursor, Codex, and Copilot CLI.

1 on GitHubby eyelockv0.3.1MITupdated 4 days ago
/

Plugin vendor-harness from marketplace eyelock-assistants in eyelock/assistants. Commands read from the repo's manifest — 4 days ago.

What it installs

Skills (12)

  • compose-feedback

    Compose a structured feedback report for a harness artifact problem — gathers required context and formats it clearly for submission as a GitHub issue, PR description, or message.

  • fetch-vendor-docs

    Fetch current documentation from LLM vendor sources — given a vendor or artifact type, retrieves up-to-date content from canonical URLs for comparison against stored references.

  • flag-vendor-gaps

    Maintain the vendor gap table — add new gaps, update existing entries, and mark gaps as resolved when vendor support improves.

  • hooks-artifact

    Validate, diagnose, and understand hook configurations across Claude Code, Cursor, Codex, and Copilot CLI — event coverage, hook types, format differences, and why hooks may not fire.

  • locate-artifact-source

    Determine where a harness artifact originated — tries multiple provenance strategies in order and returns the source repo, path, and ref when found.

  • mcp-artifact

    Validate, diagnose, and understand MCP server declarations and runtime behavior across Claude Code, Cursor, Codex, and Copilot CLI — transports, declaration format, and vendor support.

  • plugin-audit

    Audit an entire plugin or marketplace for cross-vendor readiness — walks all artifacts (skills, agents, MCP, hooks, startup context, manifests) and reports issues by vendor and priority. Use before publishing, or when asked whether a plugin or marketplace will install and work on a vendor.

  • skill-artifact

    Validate, diagnose, and understand Agent Skills (SKILL.md) across Claude Code, Cursor, Codex, and Copilot CLI — spec compliance, vendor loading behavior, and known quirks.

  • startup-context

    Validate and diagnose startup context files — CLAUDE.md, AGENTS.md, copilot-instructions.md, rules — that inject instructions at session start across all four vendors.

  • subagent-artifact

    Validate, diagnose, and understand subagent definitions (agents/*.md) — frontmatter format, delegation model, and vendor support across Claude Code, Cursor, Codex, and Copilot CLI. Use when a custom agent fails to load, or ignores its tools, model or skills, on a vendor.

  • submit-feedback

    Select the right feedback channel and submit — GitHub issue, GitHub PR, email, or message — based on provenance, contributor status, and local checkout availability.

  • vendor-adapters

    Master cross-vendor reference for Claude Code, Cursor, Codex and Copilot CLI harness formats — support matrix, manifest and marketplace paths, marketplace source types, documentation URLs, and the verified known-gaps table (including where a vendor's own docs are wrong). Use when comparing vendors,

Subagents (4)

  • feedback-composer

    Owns the feedback flow for harness artifact problems — diagnoses the gap between expected and actual behavior, selects the right submission channel, composes a structured report, and submits it.

  • harness-advisor

    Entry point for vendor harness problems — elicits what went wrong with a skill, agent, MCP server, hook, or startup context file, identifies the artifact type and vendor, then routes to the right specialist agent.

  • provenance-detective

    Determines where a harness artifact originated. Tries multiple strategies in order — YNH, native plugin directories, git log, manifest walk — then asks the user if all strategies fail.

  • vendor-sync

    Keeps vendor harness references current — fetches latest documentation from all vendor sources, compares against stored references, presents a diff of changes, and updates on confirmation.

View source on GitHub eyelock.github.io

More from eyelock/assistants

All 12 plugins →

ynh

claude

User skills for ynh: create harnesses, inspect projects, compress artifacts, validate, and set up team delegation.

20 skills · 1 command · 3 agents

1

Process music purchases and audio CDs: extract or securely rip, tag, import to Apple Music, stage MP3 and WAV to NAS, and repair split or half-done imports.

11 skills · 2 agents · 1 MCP server · hooks

1

Build a Wheelhouse: a bespoke setup of named agent seats for one space, each a ynh harness, sharing a Beads work graph and ynm memory, with handoffs instead of exits and Kudos for work well done.

13 skills

1

Development workflow skills: project setup, code quality, review, debugging, backend, UI, and security patterns.

11 skills

1

Triage skills: what is waiting on me, what am I waiting on, and what to act on now or soon, answered in one shape on any platform.

8 skills

1

gitflow

claude

Gitflow branching model: develop/main strategy, conventional commits, quality gate, tag-driven release automation, and push/PR workflow.

6 skills · 1 agent

1

Maintain vendor-harness?

This page updates itself from your repo. Add the badge to your README, or get it Featured to pin it above the organic results with a followed link to your site.

claude & codex plugins12 plugins