From workbench-kit
Second step after interview-for-personalizing. Takes a target repo location or git URL and a .persona/ draft, then scaffolds a minimal workbench (persona-composed AGENTS.md + AGENTS.overlay.md + codebases.yaml + empty docs/) and discards .persona/. The engine (skills+utils) comes from the installed plugin, so the generated repo stays minimal. Use this whenever the user is ready to create their workbench repo from a persona, or asks to generate/scaffold a workbench.
How this skill is triggered — by the user, by Claude, or both
Slash command
/workbench-kit:generate-workbenchThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Bootstrap step 2. Turn a `.persona/` draft into a real, minimal workbench repo.
Bootstrap step 2. Turn a .persona/ draft into a real, minimal workbench repo.
The mechanical part — laying the scaffold and composing AGENTS.md from the framework
core + the persona overlay — is deterministic and identical every time, so it lives in
a bundled script: scripts/compose.sh. Your job in this skill is the judgment
around it: get the target repo, locate the framework core, run the script, then clean
up and hand off. Don't re-derive the scaffolding by hand — call the script.
.persona/ (the contract from interview).persona/overlay.md — filled AGENTS.overlay.md (the 7 persona cells).persona/meta.yaml — repo: owner/name or path: …, plus language:.persona/codebases.yaml — (optional) initial work targetsIf .persona/ is absent, run interview-for-personalizing first — don't invent a persona.
Obtain the target repo from .persona/meta.yaml.
repo: owner/name → create it if needed (gh repo create owner/name --private),
then clone, or use an existing local clone. Confirm with the user before creating
anything remote — repo creation is outward-facing.path: … → use that local directory (init git if empty).Locate the inputs — both bundled in this plugin (so it works when installed).
scaffold/ (with AGENTS.core.md, AGENTS.overlay.md, templates, empty docs,
codebases.yaml) ships inside this bootstrap plugin at ${CLAUDE_PLUGIN_ROOT}/scaffold/.
No cross-plugin lookup is needed.
Run the compose script:
"${CLAUDE_PLUGIN_ROOT}/skills/generate-workbench/scripts/compose.sh" \
--persona <.persona dir> \
--core "${CLAUDE_PLUGIN_ROOT}/scaffold/AGENTS.core.md" \
--scaffold "${CLAUDE_PLUGIN_ROOT}/scaffold" \
--out <target repo>
It lays templates/ .github/ docs/(empty) codebases.yaml, writes AGENTS.overlay.md
from the persona, and composes AGENTS.md = core + overlay (inline). It does not
copy the engine — that's the plugin's job.
The script self-checks: it rejects an empty/half-filled .persona/ up front, and
after composing it asserts post-conditions (AGENTS.md has both core+overlay markers,
overlay/docs/codebases present, engine not copied, CLAUDE.md symlink) — failing
loudly. So you don't eyeball the result; if compose.sh exits 0, the workbench is
well-formed. If it exits non-zero, read the FAIL lines and fix the input.
Discard .persona/ — it's temporary scratch; its content now lives in the repo.
Commit and hand off. Make the initial commit (in the user's output language per
the persona), then point the user to their first task: workbench:task-start <issue>
(engine plugin). Remind them the workbench plugin must be installed/enabled for the
engine commands to work.
my-workbench/
AGENTS.md composed: framework core + your persona (do not edit directly)
AGENTS.overlay.md your rules — edit here, then recompose
codebases.yaml your targets (or empty template)
docs/ empty knowledge skeleton (index + empty anchor tables + log)
templates/ .github/ profile defaults you can tune
No utils/, skills/, or bin/ — the engine is the installed workbench plugin.
That's the whole point: your repo carries only your stuff.
AGENTS.overlay.md) is how rules change later, and how
framework upgrades flow in (the script re-reads the plugin's new core).npx claudepluginhub yoogomja/workbench-kit --plugin workbench-kitFetches up-to-date documentation from Context7 for libraries and frameworks like React, Next.js, Prisma. Use for setup questions, API references, and code examples.
Applies a firm's KYC/AML rules grid to parsed onboarding records: assigns risk rating, checks required documents, outputs rule outcomes with citations, and routes for escalation.
Generates daily or weekly digests of activity from connected sources (chat, email, docs, tasks, CRM), highlighting action items, decisions, mentions, and project updates.