agentsclimarketplace

Agent workshop onboard

Skill giostriquer/agent-workshop/plugins/agent-workshop/skills/agent-workshop-onboard

Set of plugins of agents and skills for daily ai work

Install
npx -y skills add giostriquer/agent-workshop --skill agent-workshop-onboard

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Use when onboarding Agent Workshop agents into a repository, planning or applying repo-local agent packs, auditing existing agent-workshop adoption, or explaining pack/profile choices.

SKILL.md

5.6 KB, ~1.2k tokens by cl100k_base, as published. Nobody here has run it

Agent Workshop Onboard

This skill is the only active plugin surface for Agent Workshop. It does not make the scaffold's agents global. It helps an operator decide which repo-local agents, wrappers, skills, conventions, and profile definitions a target project should adopt.

It adopts the scaffolding worth localizing — the agents and workflow skills that benefit from per-repo adaptation (profile slots, conventions, host wrappers), which is exactly the set in references/catalog.json and the bundled references/. The self-contained, direct-use skills (handoff-*, doc-to-html, claim-check, qa-sweep, code-quality-review) and the self-contained code-quality-reviewer agent are not bundled here — they ship ready-to-use in the separate toolkit plugin; point the operator there for those. (The review agents that are in the catalog can be adopted repo-local here, or used directly from toolkit — the operator's choice.)

Default to mode: plan. Only mode: apply may write files, and only after the operator approves a concrete plan from this session or supplies an approved plan file.

Inputs

  • mode: plan (default), apply, audit, or explain
  • target: target repository root; default to the current working directory
  • packs: optional requested packs from references/catalog.json
  • agents: optional requested agent ids from references/catalog.json
  • hosts: optional target hosts such as Claude Code, Codex, Gemini, or OpenCode
  • approved_plan: required for mode: apply; either the plan produced in this session or a path to an approved plan file

Mode: plan

Inspect the target repo and produce an adoption plan. Do not write files.

Read, when present:

  • AGENTS.md, CLAUDE.md, and host-specific workflow docs
  • .claude/, .codex/, .gemini/, .opencode/
  • docs/, especially conventions, setup, decisions, change logs, and testing docs
  • existing plugin or marketplace manifests
  • test tooling and quality policy docs when test-quality-reviewer is a candidate

Classify each candidate agent:

  • copy-as-is: canonical scaffold can be copied with only path-neutral wrapper updates
  • profile-required: suitable, but project profile slots must be filled first
  • needs-local-definition: suitable only after a narrower local variant or convention docs are written
  • defer: likely useful later, but the supporting workflow does not exist yet
  • skip: the project does not currently have the problem the agent solves

Return:

  • selected packs and agents
  • classification and rationale for each candidate
  • required profile slots
  • exact files proposed for create/modify
  • source template for each proposed file or reference path to the bundled template
  • validation checks to run after apply
  • risks, omitted agents, and follow-up decisions
  • explicit prompt asking whether to proceed with mode: apply

Mode: apply

Apply only an approved plan. Refuse to write if there is no concrete approved plan.

Before writing:

  1. Re-read the approved plan.
  2. Run git status --short in the target repo when it is a git checkout.
  3. Report unrelated dirty files.
  4. Refuse if the plan lacks exact file paths.
  5. Refuse if any target file has conflicting uncommitted changes not accounted for in the plan.

When writing:

  • write only files named in the approved plan
  • preserve unrelated sections in existing AGENTS.md, CLAUDE.md, and docs
  • prefer repo-local .claude/, .codex/, .gemini/, and .opencode/ files
  • never edit user/global host config
  • never copy private local project paths from examples into the target repo
  • do not commit unless the approved plan explicitly asks for a commit

After writing:

  • run the approved validation checks
  • report changed files, skipped files, validation output, and remaining profile gaps

Mode: audit

Inspect an existing Agent Workshop adoption and report drift. Do not write files.

Check:

  • installed agents versus references/catalog.json
  • local divergence from bundled references
  • wrapper pointers and host support
  • skill mirrors when multiple hosts are supported
  • required profile slots
  • agreement between AGENTS.md and CLAUDE.md
  • stale agents that no longer earn their keep
  • plugin-global agents accidentally replacing repo-local agents

Mode: explain

Answer questions about packs, profile slots, host wrappers, or agent boundaries. Use references/catalog.json and bundled docs as the source model.

Safety Rules

  • Default to read-only planning.
  • Do not install all agents by default.
  • Do not install pattern-reviewer as a gate until pattern domains and convention docs exist.
  • Do not install visual-implementer unless a visual baseline and proof path exist or are in the approved plan.
  • Do not install research unless a project-owned research brief schema exists or is in the approved plan.
  • Do not derive CRAP from coverage data unless complexity values are valid.
  • Treat high-impact surfaces as profile-specific; average coverage is not enough.
  • Do not overwrite existing local agents without showing the local diff and getting approval.

Bundled References

Bundled templates live under references/ inside this skill. They are source material for repo-local adoption, not active plugin agents or skills. Reference skill templates use references/skills/<skill>.md instead of nested SKILL.md filenames so plugin hosts do not auto-discover them as active skills.

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.