askit-init-plugin
Purpose
Onboard a maintainer and scaffold a starting point that is gradeable from the first commit. Three modes: interview runs a live conversational Q&A, then scaffolds; questionnaire emits the structured onboarding template (with per-section maintainer-feedback and agent-response blocks) for the maintainer to fill async, then processes it; hybrid pre-fills that questionnaire with suggestions from chat context, still leaving the feedback blocks. All three scaffold the Bronze seed anatomy from templates/seed-plugin/ (library.json with the five required fields + AGENTS.md + a minimal .claude-plugin/plugin.json (name, version, description only, ADR 0043) + README/CHANGELOG starters) so the new plugin passes the conformance core immediately AND installs on Claude Code and Codex from the first commit. Fill name/description identically in library.json and .claude-plugin/plugin.json, or U8 (manifest-drift) warns on the mismatch. interview mode additionally asks for an optional author, so a plugin scaffolded that way carries a real attribution and passes claude plugin validate --strict outright; declining leaves the manifest exactly as minimal as the template, still --strict-warning and correctly so (ADR 0043). Mode and anatomy detail are in references/onboarding-modes.md.
When to use
When starting a brand-new plugin from scratch, onboarding a maintainer, or generating a scaffolding questionnaire. (To adopt an EXISTING repo instead, use askit-migrate.)
interview mode
- Ask the maintainer the onboarding questions live (theme/scope, target agents, target tier, first components, and an optional author: a name, and optionally a url/email).
- Synthesize a starting config and scaffold from
templates/seed-plugin/, filling name/description in BOTH library.json and .claude-plugin/plugin.json (the same values in both) and AGENTS.md. If the maintainer supplied an author, add author: { name, [url], [email] } (only name is required by the vendor's schema) to BOTH library.json (its existing OPTIONAL field, Standard sec 5.1, so a later Silver-tier regeneration via gen-manifest.mjs does not silently drop it) and .claude-plugin/plugin.json. If the maintainer declined or skipped, emit NO author key in either file - never a placeholder (ADR 0043; the same class of defect U5 already penalizes in descriptions).
- Hand off:
askit-build-skill for the first skill, askit-build-docs for docs, askit-evaluate to confirm Bronze.
questionnaire mode
- Emit
templates/onboarding-questionnaire.template.md for the maintainer to fill async.
- On return, process the maintainer-feedback blocks into a config and scaffold as above.
hybrid mode
- From chat context, emit a tailored questionnaire pre-filled with suggested answers (keeping the feedback blocks).
- Process the returned, corrected questionnaire and scaffold.
Scope
The seed scaffold is the minimal Bronze anatomy (the structural match the asserted test enforces, ADR 0023, not a byte-exact diff): it passes every Universal check with 0 errors. The minimal .claude-plugin/plugin.json is install recognition ONLY, not Silver's S6 (per-target emission): it does not carry agent-targets, prefix, a components index, or a second native manifest, so the seed does not claim Silver behavior it has not built (ADR 0043). The richer surfaces (a real README via askit-build-docs, CHANGELOG via askit-release, ADRs via askit-decision, backlogs via askit-backlog, and the full per-target manifests via scripts/generators/gen-manifest.mjs at Silver) are scaffolded by their own skills, so init composes the toolkit rather than duplicating it. Marketplace scaffolding is askit-init-marketplace.
1---2name: askit-init-plugin3description: Creates a starting plugin that satisfies the Bronze anatomy and onboards the maintainer, in three modes (interview, questionnaire, hybrid). Use when starting a new plugin from scratch, onboarding a maintainer, or generating a scaffolding questionnaire.4---56# askit-init-plugin78## Purpose9Onboard a maintainer and scaffold a starting point that is gradeable from the first commit. Three modes: `interview` runs a live conversational Q&A, then scaffolds; `questionnaire` emits the structured [onboarding template](../../templates/onboarding-questionnaire.template.md) (with per-section maintainer-feedback and agent-response blocks) for the maintainer to fill async, then processes it; `hybrid` pre-fills that questionnaire with suggestions from chat context, still leaving the feedback blocks. All three scaffold the Bronze seed anatomy from [`templates/seed-plugin/`](../../templates/seed-plugin/) (`library.json` with the five required fields + `AGENTS.md` + a minimal `.claude-plugin/plugin.json` (`name`, `version`, `description` only, ADR 0043) + README/CHANGELOG starters) so the new plugin passes the conformance core immediately AND installs on Claude Code and Codex from the first commit. Fill `name`/`description` identically in `library.json` and `.claude-plugin/plugin.json`, or U8 (manifest-drift) warns on the mismatch. `interview` mode additionally asks for an optional author, so a plugin scaffolded that way carries a real attribution and passes `claude plugin validate --strict` outright; declining leaves the manifest exactly as minimal as the template, still `--strict`-warning and correctly so (ADR 0043). Mode and anatomy detail are in [references/onboarding-modes.md](references/onboarding-modes.md).1011## When to use12When starting a brand-new plugin from scratch, onboarding a maintainer, or generating a scaffolding questionnaire. (To adopt an EXISTING repo instead, use `askit-migrate`.)1314## interview mode151. Ask the maintainer the onboarding questions live (theme/scope, target agents, target tier, first components, and an optional author: a name, and optionally a url/email).162. Synthesize a starting config and scaffold from `templates/seed-plugin/`, filling `name`/`description` in BOTH `library.json` and `.claude-plugin/plugin.json` (the same values in both) and `AGENTS.md`. If the maintainer supplied an author, add `author: { name, [url], [email] }` (only `name` is required by the vendor's schema) to BOTH `library.json` (its existing OPTIONAL field, Standard sec 5.1, so a later Silver-tier regeneration via `gen-manifest.mjs` does not silently drop it) and `.claude-plugin/plugin.json`. If the maintainer declined or skipped, emit NO `author` key in either file - never a placeholder (ADR 0043; the same class of defect U5 already penalizes in descriptions).173. Hand off: `askit-build-skill` for the first skill, `askit-build-docs` for docs, `askit-evaluate` to confirm Bronze.1819## questionnaire mode201. Emit `templates/onboarding-questionnaire.template.md` for the maintainer to fill async.212. On return, process the maintainer-feedback blocks into a config and scaffold as above.2223## hybrid mode241. From chat context, emit a tailored questionnaire pre-filled with suggested answers (keeping the feedback blocks).252. Process the returned, corrected questionnaire and scaffold.2627## Scope28The seed scaffold is the minimal Bronze anatomy (the structural match the asserted test enforces, ADR 0023, not a byte-exact diff): it passes every Universal check with 0 errors. The minimal `.claude-plugin/plugin.json` is install recognition ONLY, not Silver's S6 (per-target emission): it does not carry `agent-targets`, `prefix`, a `components` index, or a second native manifest, so the seed does not claim Silver behavior it has not built (ADR 0043). The richer surfaces (a real README via `askit-build-docs`, CHANGELOG via `askit-release`, ADRs via `askit-decision`, backlogs via `askit-backlog`, and the full per-target manifests via `scripts/generators/gen-manifest.mjs` at Silver) are scaffolded by their own skills, so init composes the toolkit rather than duplicating it. Marketplace scaffolding is `askit-init-marketplace`.
product-on-purpose