@joshuafolkken/kit 1.1050.0 → 1.1128.0

package/SECURITY.md

@@ -4,7 +4,10 @@
 
 | Version | Supported          |
 | ------- | ------------------ |
-| 0.x     | :white_check_mark: |
+| Latest  | :white_check_mark: |
+| Older   | :x:                |
+
+Fixes ship in a new release rather than as patches to older versions; upgrade to the latest release to receive them. Each release's changes are listed on the [Releases](https://github.com/joshuafolkken/kit/releases) page.
 
 ## Reporting a Vulnerability
 
@@ -17,13 +20,41 @@ We aim to acknowledge reports within **48 hours** and provide a resolution or mi
 
 ## Published Package Contents
 
-The following files are intentionally included in the published npm package and distributed to consuming projects via `josh sync`:
+The `files` field of `package.json` decides what the published `@joshuafolkken/kit` package contains. None of it carries credentials or private URLs. `josh init` and `josh sync` copy or merge part of it into a project; [sync.md](https://github.com/joshuafolkken/kit/blob/main/docs/sync.md) lists which files are overwritten and which are merged.
+
+**Code that runs on your machine and in CI**
+
+- `dist/`, `scripts/` — the `josh` command and the scripts behind it. They read and write files in your repository, run `git`, `gh` and `pnpm`, and call the GitHub API and, when configured, the Telegram API.
+- `.claude/settings.json` — Claude Code hooks that run `pnpm josh …` commands on session start, on each prompt, before and after tool calls and on stop, plus a deny list for staging, committing, merging, force pushes and branch deletions. `josh sync` overwrites it.
+- `.codex/` — the same hooks for Codex (`hooks.json`) and its project settings (`config.toml`).
+- `.github/` — GitHub Actions workflows (CI, tagging, pull request classification, Dependabot auto-merge, SonarQube and production), composite actions, `dependabot.yml`, the pull request template and the release-notes config. `josh sync` overwrites the distributed workflows; their action references are pinned to commit SHAs. kit's own `publish.yml` ships in the package but is never copied into a project.
+- `lefthook/` — Git hook presets: lint before commit, checks before push.
+- The `preinstall` script `josh init` adds to `package.json` runs `pnpm dlx @aikidosec/safe-chain setup-ci` on every install, then a `node -e` check that only prints a warning, and only outside CI. The `prepare` script it adds runs `lefthook install` when Lefthook is present, then, when tsx is present, kit's `fix-gh-packages` script: only when the project `.npmrc` routes a scope to GitHub Packages, it reads a GitHub token (`NODE_AUTH_TOKEN`, a token in the project `.npmrc`, or `gh auth token`), queries GitHub Packages and rewrites tarball URLs in `pnpm-lock.yaml`.
+- `.pnpmfile.mjs` — a pnpm `beforePacking` hook that removes that safe-chain `preinstall` from the manifest `pnpm pack` and `pnpm publish` write, and changes nothing else. `josh init` and `josh sync` copy it only into a project whose `package.json` is not `private: true`, and never over a pnpmfile of the project's own.
+
+**Agent instructions and skills**
+
+- `CLAUDE.md`, `AGENTS.md`, `GEMINI.md`, `.cursorrules`, `prompts/` — the rules and procedures AI agents follow.
+- `.claude/skills/`, `.claude/.claude-plugin/`, `.claude-plugin/` — the Claude Code skills, shipped as the `kit` plugin and its marketplace entry.
+- `.mcp.json` — the MCP server configuration for the Svelte documentation server. It contains only the public endpoint `https://mcp.svelte.dev/mcp`.
+- `.coderabbit.yaml` — CodeRabbit review settings.
+
+**Configuration presets and templates**
+
+- `eslint/`, `prettier/`, `tsconfig/`, `cspell/`, `env/`, `ports/`, `playwright.config.ts`, `tsconfig.sonar.json` — shared configuration your project imports or receives.
+- `templates/` — the templates `josh init` and `josh sync` write from.
+- `.vscode/`, `.gitattributes`, `.ncurc.json`, `.prettierignore`, `pnpm-workspace.yaml` — editor, Git and package-manager settings.
+- `CODE_OF_CONDUCT.md`, `SECURITY.md` — community files.
+
+## Secrets
 
-- `.mcp.json` — MCP server configuration for the Svelte documentation server. Contains only the public endpoint `https://mcp.svelte.dev/mcp`. No credentials or private URLs.
-- `.coderabbit.yaml`, `.cursorrules` — AI tool configurations with no sensitive data.
-- `templates/`, `prompts/` — Markdown and config templates. No credentials.
+kit never stores a secret in a file it ships or commits. The ones it uses:
 
-Sensitive runtime secrets (`TELEGRAM_BOT_TOKEN`, `TELEGRAM_CHAT_ID`) are stored only in a local `.env` file which is gitignored and never committed to the repository.
+- `TELEGRAM_BOT_TOKEN`, `TELEGRAM_CHAT_ID` — Telegram notifications. Keep them in a local `.env` file, which is gitignored, or in the environment. Set `JOSH_NOTIFY=off` to need neither.
+- `SONAR_TOKEN` — a repository secret the distributed SonarQube workflow reads. Without it the workflow skips the scan.
+- `GITHUB_TOKEN` — supplied by GitHub Actions to the distributed workflows; nothing to configure.
+- `GH_TOKEN` — authenticates `gh` in CI and cloud sessions, where `gh auth login` is not available.
+- `NODE_AUTH_TOKEN` — the GitHub Packages credential, only for projects that still install kit from there ([authentication.md](https://github.com/joshuafolkken/kit/blob/main/docs/authentication.md)). The `prepare` script above reads it.
 
 ## MCP External Endpoints
 

package/cspell/index.yaml

@@ -41,6 +41,10 @@ ignorePaths:
   #   - '**/*.min.js'
   #   - '**/*.svg'
 words:
+  # pnpm's `.pnpmfile.mjs` hook file that strips the safe-chain preinstall at pack time
+  # (joshuafolkken/kit#2693). Distributed, because `josh init` / `josh sync` copy that file, which
+  # names itself, into a published consumer (joshuafolkken/kit#3110).
+  - pnpmfile
   # Git's `--show-toplevel` flag is added to every Codex hook command when `josh sync` copies the
   # hooks into a consumer. The consumer spell-checks that transformed file with this dictionary.
   - toplevel
@@ -59,7 +63,7 @@ words:
   # (joshuafolkken/kit#1299). Distributed, because `docs/josh-commands.md` names it in the phase list.
   - wrapup
   # What `josh time`'s bundling block calls a call that could have gone out beside another
-  # (joshuafolkken/kit#1607). Distributed, because `docs/josh-commands.md` and `turn-batching.md` name it.
+  # (joshuafolkken/kit#1607). Distributed, because `docs/josh-commands-automation.md` and `turn-batching.md` name it.
   - bundleable
   # The alias for `josh lane:launch`, the composite lane-start command (joshuafolkken/kit#2162).
   # Distributed, because `docs/josh-commands.md` and `backlogrun-lanes.md` name it and `josh sync`
@@ -124,6 +128,7 @@ words:
   - mischeck
   - multicriteria
   - narratively
+  - ncurc
   - networkidle
   - NOSONAR
   - noopener

package/dist/CLAUDE.basic.md

@@ -1,8 +1,8 @@
 # Project instructions
 
-- 既存のファイルと利用者の設定を尊重し、変更する前に内容を確認する。
-- 変更には検証を付け、実行した結果を報告する。
-- HTML、CSS、JavaScript のファイルがある場合は Prettier で整形する。
-- 画面に見える変更をした場合はブラウザーで表示と操作を確認する。
-- JavaScript または TypeScript の動作を変更した場合は、その動作を確認する自動テストを追加する。
-- Git や GitHub を使う場合だけ、それぞれのワークフローを利用する。
+- Respect existing files and the user's settings, and read a file's contents before changing it.
+- Verify every change, and report the results you actually observed.
+- Format HTML, CSS and JavaScript files with Prettier when the project has any.
+- After a change that is visible on screen, check its display and interaction in a browser.
+- After changing JavaScript or TypeScript behavior, add an automated test that checks that behavior.
+- Use the Git or GitHub workflow only when you use Git or GitHub.

package/dist/CLAUDE.md

@@ -1,8 +1,6 @@
 # Claude Code Instructions
 
-> **This file is the single source for every agent rule.** `AGENTS.md` and `GEMINI.md` are pointers to it and carry no rules of their own — a rule addition, spec change or wording fix is written **once**, here. History: `https://github.com/joshuafolkken/kit/blob/main/docs/maintainers/claude-md-history.md`.
->
-> Each rule is written as a **trigger plus a pointer** — enough to act safely with nothing else loaded, with the steps and rationale at the pointer.
+> **This file is the entry point for every agent rule.** `AGENTS.md` and `GEMINI.md` are pointers to it and carry no rules of their own — a rule's trigger is written **once**, here, and its procedure and any number it uses **once**, at its pointer. History, and the explanations kept out of this file: `https://github.com/joshuafolkken/kit/blob/main/docs/maintainers/claude-md-history.md`.
 
 ## Project
 
@@ -19,8 +17,8 @@ Stack: TypeScript · pnpm · SvelteKit · Vitest · Playwright · TailwindCSS ·
 - **Distinguish consultation from execution.** A question about approach or a goal statement ("どうすべき？", "〜したい") is a request for analysis only — act only on an explicit imperative or a workflow keyword. `node_modules/@joshuafolkken/kit/prompts/collaboration-workflow/principles.md` → "相談と実行".
 - **Route distributed-doc / config changes upstream to kit.** In a consumer repo never edit kit-distributed docs/config locally (`josh sync` overwrites them); propose the change for kit. In kit itself you are the source. `node_modules/@joshuafolkken/kit/prompts/collaboration-workflow/principles.md` → "配布ドキュメント".
 - **Latest-first, fix forward — pin back only as a last resort.** Adopt newest versions by default and resolve a breaking bump forward, recording why on the rare pin-back. `node_modules/@joshuafolkken/kit/prompts/collaboration-workflow/principles.md` → "最新優先".
-- **Output language follows `JOSH_SESSION_LANG`** (default `ja`; a hook injects it). It governs session-facing output and artifact prose (Issue bodies, comments, Telegram). English stays for Issue/PR titles, code comments/test titles/commit messages, and script-emitted strings. `node_modules/@joshuafolkken/kit/prompts/collaboration-workflow/overview.md`.
-- **Cite an Issue with a number-link and a short Japanese title in session-facing output** — `[#<N>](https://github.com/<owner>/<repo>/issues/<N>) — <短い要約>`. `node_modules/@joshuafolkken/kit/prompts/collaboration-workflow/issue-citation.md`.
+- **Output language follows `JOSH_SESSION_LANG`** (default `ja`; a hook injects it). It governs session-facing output and artifact prose (Issue bodies, comments, Telegram). English stays for Issue/PR titles, code comments/test titles/commit messages, and script-emitted strings. One language per doc layer (English here; mostly Japanese in the prompt files). `node_modules/@joshuafolkken/kit/prompts/collaboration-workflow/overview.md`.
+- **Cite an Issue with a number-link and a short title in the session language in session-facing output** — `[#<N>](https://github.com/<owner>/<repo>/issues/<N>) — <short summary>`. `node_modules/@joshuafolkken/kit/prompts/collaboration-workflow/issue-citation.md`.
 - **Durable rules belong in prompts/docs, not local MEMORY.** Encode a lasting rule as a change to kit's distributed prompts/docs, not auto-memory; propose the edit when the turn does not authorize one. `node_modules/@joshuafolkken/kit/prompts/collaboration-workflow/principles.md` → "恒久ルール".
 - **Never carry a file's new text inside a shell command.** Edit a region with the Edit tool, never a heredoc — unless the change is so scattered the partial edit would carry more text than the file. The criterion is whether the command carries the replacement wholesale, not the tool's name. `node_modules/@joshuafolkken/kit/prompts/collaboration-workflow/file-edits.md`.
 - **Never put a body in shell double quotes** — a backtick or `$` there is _executed_; pass the body by path. `node_modules/@joshuafolkken/kit/prompts/collaboration-workflow/shell-body.md`.
@@ -38,7 +36,7 @@ Classify each decision point into one tier and act; stop **only** when the choic
 
 ## Environment Variables
 
-Read from a `.env` file at the project root by the AI scripts, `josh port` and `playwright.config.ts`. `TELEGRAM_BOT_TOKEN` and `TELEGRAM_CHAT_ID` are required for notifications; `JOSH_SESSION_LANG`, `PORT_SEED` and `JOSH_REPO_PATHS` are optional and personal. Setup and full semantics: [docs/scripts-ai.md](https://github.com/joshuafolkken/kit/blob/main/docs/scripts-ai.md) and [docs/josh-commands.md](https://github.com/joshuafolkken/kit/blob/main/docs/josh-commands.md).
+A `.env` at the project root: `TELEGRAM_BOT_TOKEN` and `TELEGRAM_CHAT_ID` are required for notifications; `JOSH_SESSION_LANG`, `PORT_SEED` and `JOSH_REPO_PATHS` are optional and personal. Details: [environment-variables.md](https://github.com/joshuafolkken/kit/blob/main/docs/environment-variables.md), [josh-commands.md](https://github.com/joshuafolkken/kit/blob/main/docs/josh-commands.md).
 
 **GitHub operations are `gh api` (REST) — instructing prose included — and need `gh` installed; some environments lack it.** Auth: `gh auth login`; `GH_TOKEN` in CI/cloud. `node_modules/@joshuafolkken/kit/prompts/collaboration-workflow/gh-rest.md`.
 
@@ -66,7 +64,7 @@ Read from a `.env` file at the project root by the AI scripts, `josh port` and `
 ### Quality limits
 
 - Function complexity ≤5 · nesting ≤2 · function ≤25 lines · file ≤300 lines · params ≤4 · statements per function ≤10 · cognitive complexity ≤4
-- **Those line counts are code lines, not physical lines** (`max-lines` / `max-lines-per-function` run with `skipBlankLines` + `skipComments`) — judge by what `pnpm josh lint` reports, never `wc -l`. Test files (`*.test.ts` / `*.e2e.ts`) allow 35 code lines per function.
+- **Those line counts are code lines, not physical lines** — judge by what `pnpm josh lint` reports, never `wc -l`. Test files (`*.test.ts` / `*.e2e.ts`) allow 35 code lines per function.
 - Magic numbers: extract all literals except `0`, `1`, `-1` to named `UPPER_CASE` constants
 - No `any` · no unused vars · no floating promises · type assertions (`as`) are restricted
 - All function params and return types must be explicitly typed
@@ -96,18 +94,20 @@ Read from a `.env` file at the project root by the AI scripts, `josh port` and `
 
 - Before building a feature, check whether a well-maintained package already solves the problem rather than writing original code first, and propose replacing hand-rolled implementations with a suitable package when it improves maintainability.
 - Prefer modern, actively-maintained packages (maintenance, popularity, bundle size, TypeScript support, license, fit). **If one is clearly the best fit, select it** (Tier A — log it); **only when two or more are genuinely close**, present about three ranked options and let the user choose.
-- **Measure the choice before you judge it — `pnpm josh pkg:scout <keywords>`.** It ranks the candidate packages by downloads, last publish, bundled types, license and install size, and prints `clear` when the leader is ahead by at least the near-tie threshold (select it, Tier A) or `close` when the top two are within it (ask, Tier B) — so "clearly best" and "genuine toss-up" are read off the output rather than decided by impression.
+- **Measure the choice before you judge it — `pnpm josh pkg:scout <keywords>`:** `clear` → select it (Tier A), `close` → ask (Tier B). Metrics and threshold: `https://github.com/joshuafolkken/kit/blob/main/docs/josh-commands-automation.md`.
 
 ## Code Change Rules
 
 For every code modification, in order:
 
-0. **Work summary + test declaration** _(before writing any implementation code)_: a two-layer summary — a plain-language overview a non-programmer can follow (three lines, **Now / Change / Check**, one sentence each, in the session language; name the concrete subject in each line — subject-less prose is not acceptable; no file paths, function or type names, or CLI option flags — only internal identifiers are banned, not names the reader sees on screen), **Details**, then **every change with its test** (`<what changes> — Test: <Unit|E2E> — <file path> — <what it verifies>`). Ordinary markdown, **never wrapped in a code fence**, once per Issue before implementation — required in fullrun/halfrun/backlogrun even when the Issue body was filled (`kickoff` is exempt); completion reports lead with **Cause / Fix / Result**. `node_modules/@joshuafolkken/kit/prompts/collaboration-workflow/report-format.md`; `pnpm josh report:lint` checks the form, and the `UserPromptSubmit` hook points here. Presentation only, never a confirmation stop.
+0. **Work summary + test declaration** _(before writing any implementation code)_: once per Issue, present a two-layer summary in ordinary markdown, **never wrapped in a code fence** — required in fullrun/halfrun/backlogrun even when the Issue body was filled (`kickoff` is exempt). Presentation only, never a confirmation stop. `node_modules/@joshuafolkken/kit/prompts/collaboration-workflow/report-format.md`; `pnpm josh report:lint` checks the form.
 
+   - **Overview** — three lines, **Now / Change / Check**, one sentence each, in the session language, that a non-programmer can follow; name the concrete subject in each line — subject-less prose is not acceptable. Use no file paths, function or type names, or CLI option flags — only internal identifiers are banned, not names the reader sees on screen.
+   - **Details**, then **every change with its test**: `<what changes> — Test: <Unit|E2E> — <file path> — <what it verifies>`. Completion reports lead with **Cause / Fix / Result**.
    - **Tests are required for ALL changes** — zero tests without explicit approval is a violation (bug fixes, timing/animation fixes and refactors included): bug fix → regression test; UI/animation/timing → E2E for the observable change; logic/utility → unit; refactor → tests pinning existing behavior BEFORE the change (`node_modules/@joshuafolkken/kit/prompts/refactoring.md`).
-   - **Non-runtime updates (pre-approved manual-only exception)**: a change touching no executable runtime code path may proceed with manual verification — declare it in Step 0 and state why. It covers docs, non-executable config, editor/IDE files, and cosmetic asset swaps with no selector/path change; the mechanically exempt (`*.md`, `.editorconfig`, `.idea/**`, `.vscode/**`, `prompts/**`) are recognized by the command below, and the rest stay the exemption a person declares. A test infeasible for a runtime change needs its reason stated and user approval.
-   - **basic の手動確認例外:** `package.json` に `josh.profile: basic`（改名前の `static` も同じ）が記録され、変更が kit がテストできない言語のファイル（HTML/CSS や Lua などのソース）だけなら、自動テストの代わりに手で確認する。HTML/CSS はブラウザーで対象ページを開き、想定する画面幅でレイアウト・リンク・HTML 内の操作を確認する。そのほかのソースは変更した処理を実際に動かす。どちらも観察結果を報告する。JavaScript と TypeScript のファイルには同時に変更した場合も自動テストが必要。`pnpm josh test:declared` はこの場合だけ理由を添えて `exempt` と表示する。
-   - **The commit stage answers this mechanically**: `pnpm josh test:declared` prints `required` / `exempt` / `satisfied` from the changed paths, and a runtime change with no test is refused at `pnpm josh git -y` on `required` — add the declared test or declare the non-runtime exception, then reissue (`node_modules/@joshuafolkken/kit/prompts/collaboration-workflow/rule-delivery.md`).
+   - **Non-runtime updates (pre-approved manual-only exception)**: a change touching no executable runtime code path — docs, non-executable config, editor/IDE files, cosmetic asset swaps with no selector/path change — may be verified by hand; declare it in Step 0 with the reason. The mechanically exempt (`*.md`, `.editorconfig`, `.idea/**`, `.vscode/**`, `prompts/**`) need no declaration. A runtime change whose test is infeasible needs the reason and user approval.
+   - **Basic-profile manual-check exception:** under `josh.profile: basic` (or `static`), a change touching only languages kit cannot test (HTML/CSS, Lua …) is verified by hand — a browser check of layout, links and interactions in the HTML, or actually run the changed code — and reported; JavaScript and TypeScript still need tests. `node_modules/@joshuafolkken/kit/prompts/testing-guide.md`.
+   - **`pnpm josh test:declared`** answers `required` / `exempt` / `satisfied` from the changed paths; a runtime change with no test is refused at `pnpm josh git -y` on `required`.
    - **Read a target's line headroom before editing: `pnpm josh lines <path>`** — near the limit, `Target` carries the splitting plan.
 
 1. **Refactor first** — before any refactoring, read and follow `node_modules/@joshuafolkken/kit/prompts/refactoring.md`; apply high/medium refactoring to all new/modified code and do not proceed until none remain.
@@ -131,7 +131,11 @@ Run the full verification set **in order**; do not skip, reorder, or report comp
 
 ## Pre-commit Self-Review (mandatory)
 
-Before every `git commit` (follow-up commits included), self-review against `node_modules/@joshuafolkken/kit/prompts/review.md` — a subagent runs `/code-review`, never a main-line `Skill` load: the staged diff, and `git diff main...HEAD` before opening or updating a PR. Level from `pnpm josh review:brief --level-only`; resolve all high/medium findings; iterate — **at most two reviews in total**, the second a verification pass over the fix delta. Then route each remaining non-High finding through the three-way disposition: fix in place, file it as a follow-up Issue only when it is a confirmed defect that reaches a runtime path — `pnpm josh disposition <path>` answers `runtime` / `non-runtime`, so the only judgement left is whether the defect is confirmed — or drop it with a one-line PR note (the default). **A filed finding is run through `pnpm josh epic:bundle <new>` before the current Issue closes.** A confirmed High blocks rather than buying a third round; CI runs no Claude review, so this pass is authoritative. `node_modules/@joshuafolkken/kit/prompts/review.md` → "Review round cap".
+Before every `git commit` (follow-up commits included), self-review against `node_modules/@joshuafolkken/kit/prompts/review.md` — a subagent runs `/code-review`, never a main-line `Skill` load. `node_modules/@joshuafolkken/kit/prompts/review.md` → "Review round cap".
+
+- **Scope** — the staged diff, and `git diff main...HEAD` before opening or updating a PR; level from `pnpm josh review:brief --level-only`.
+- **Rounds** — resolve all high/medium findings, **no further than the round cap**, the last round a verification pass over the fix delta. A confirmed High blocks rather than buying a round past the cap.
+- **Each remaining non-High finding** — fix in place, file it as a follow-up Issue only when it is a confirmed defect that reaches a runtime path (`pnpm josh disposition <path>`), or drop it with a one-line PR note (the default). **A filed finding is run through `pnpm josh epic:bundle <new>` before the current Issue closes.**
 
 ## Doc Sync Rules
 
@@ -139,20 +143,20 @@ Before every `git commit` (follow-up commits included), self-review against `nod
 
 ## Git Rules
 
-- **No commits** unless explicitly requested. **No PR merges, branch deletions, force pushes or other shared-state mutations** unless explicitly requested in the current turn — the default end state is PR still OPEN. **Exception**: invoking `fullrun` authorizes the merge, via `pnpm josh followup`. `.claude/settings.json` denies these, but the deny is narrower than the rule — never read "the tool let me" as permission. `node_modules/@joshuafolkken/kit/prompts/collaboration-workflow/operating-rules.md` → "指示されていない行動は取らない".
+- **No commits** unless explicitly requested. **No PR merges, branch deletions, force pushes or other shared-state mutations** unless explicitly requested in the current turn — the default end state is PR still OPEN. **Exception**: invoking `fullrun` or `backlogrun` authorizes the merge, via `pnpm josh followup` — `prrun` does not. `.claude/settings.json` denies these, but the deny is narrower than the rule — never read "the tool let me" as permission. `node_modules/@joshuafolkken/kit/prompts/collaboration-workflow/operating-rules.md` → "指示されていない行動は取らない".
 - **Never stage or mutate the git index on your own.** `git add` / `git rm --cached` / `git restore --staged` / `git commit` overwrite the user's snapshot and are denied by settings; inspection is read-only. Staging is allowed only on an explicit current-turn instruction or an authorized commit flow (`pnpm josh git`). `node_modules/@joshuafolkken/kit/prompts/collaboration-workflow/operating-rules.md` → "git index を勝手に変更しない".
 - **For git operations use `pnpm josh git`.** After a failed push, fix, push manually, then `pnpm josh pr` — **never** `gh pr create` directly, which bypasses `closes #N` generation so the Issue will not auto-close.
 - **Start-of-conversation git status is a stale snapshot.** Before acting on any working-tree / index / stash / branch assumption, run `git status` (and `git stash list`) live first.
 
 ## Collaboration Workflow
 
-- For the issue-driven proposal/plan/execution/notification flow, follow `node_modules/@joshuafolkken/kit/prompts/collaboration-workflow/` (indexed by `node_modules/@joshuafolkken/kit/prompts/collaboration-workflow.md`; each topic is its own file).
+- The issue-driven flow is `node_modules/@joshuafolkken/kit/prompts/collaboration-workflow/`, indexed by `node_modules/@joshuafolkken/kit/prompts/collaboration-workflow.md`.
 - **Before writing a rule as prose, place it by `node_modules/@joshuafolkken/kit/prompts/collaboration-workflow/residency.md`** — its single source. A computable answer goes to `pnpm josh oracle:list` and an ordering rule to `pnpm josh run:step`, never into `CLAUDE.md` or a prompt file.
-- **Count the target repository's open Issues before filing; with more than 30 open, close one first.** Nothing honestly closable means do not file. A filing the run is blocked by is exempt, and so is an **interrupt** — three tests decide that, never judgement: a verification answers wrongly, a documented workflow cannot complete, or data is lost or written outside the repository; both exemptions proceed, stating the overage. Meeting none of the three, a finding stays discretionary. `pnpm josh rule:guard` states it again at the call that files. The count command and both procedures: `node_modules/@joshuafolkken/kit/prompts/collaboration-workflow/wip-cap.md`.
+- **Count the target repository's open Issues before filing; above the WIP cap, close one first.** Nothing honestly closable means do not file. A filing the run is blocked by is exempt, and so is an **interrupt** — three tests decide that, never judgement: a verification answers wrongly, a documented workflow cannot complete, or data is lost or written outside the repository; both exemptions proceed, stating the overage. Meeting none of the three, a finding stays discretionary. `pnpm josh rule:guard` states it again at the call that files. The count command and both procedures: `node_modules/@joshuafolkken/kit/prompts/collaboration-workflow/wip-cap.md`.
 
 ### Shorthand Commands
 
-`kickoff`, `fullrun`, `halfrun` and `backlogrun` are the Issue-driven shorthand commands. **Their procedures are not resident** — they live in the `workflow-commands` skill. **What stays here is decided by one question: must the rule fire on a turn where no skill was loaded?** Explicit invocation, the mid-workflow stop notification, the `overrides` / `devEngines` prohibitions, the UI-verification gate and the three `epic:*` rules below all do; everything a run reaches only after reading the skill is routed to, never restated (`node_modules/@joshuafolkken/kit/prompts/collaboration-workflow/residency.md` → "第 2 問").
+`kickoff`, `fullrun`, `prrun`, `halfrun` and `backlogrun` are the Issue-driven shorthand commands. **Their procedures are not resident** — they live in the `workflow-commands` skill. **What stays here is decided by one question: must the rule fire on a turn where no skill was loaded?** Explicit invocation, the mid-workflow stop notification, the `overrides` / `devEngines` prohibitions, the UI-verification gate and the three `epic:*` rules below all do; everything a run reaches only after reading the skill is routed to, never restated (`node_modules/@joshuafolkken/kit/prompts/collaboration-workflow/residency.md` → "第 2 問").
 
 **Read the skill before running any part of a command — including the first `gh` call.**
 
@@ -160,36 +164,38 @@ Before every `git commit` (follow-up commits included), self-review against `nod
 | ------------------------ | --------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ---------------------------------------------------------------- |
 | `kickoff [#N \| new]`    | Plan only — normalize title, post the plan, notify, **stop**                                                                                                          | `workflow-commands` skill + `kickoff.md` + `split-assessment.md` |
 | `fullrun [#N \| new]`    | Plan → implement → gate → PR → **merge** → notify                                                                                                                     | + `fullrun.md` + `split-assessment.md` + `chain-rule.md`         |
+| `prrun [#N \| new]`      | `fullrun` up to a green, mergeable PR, then **stop** for a person to merge by hand                                                                                    | + `prrun.md` + the `fullrun` set                                 |
 | `halfrun [#N \| new]`    | Implement + gate, then **stop before commit** for manual verification                                                                                                 | + `halfrun.md` + `split-assessment.md`                           |
 | `backlogrun [#N1 #N2 …]` | Run named Issues and epics in order (an epic runs its children first), then drain the opted-in backlog — dependency order, lanes. `--only` stops after the named list | + `backlogrun.md` + `split-assessment.md` + the `fullrun` set    |
 
-**`queue` and `epicrun` were removed** — tell a user who types `queue` to run `backlogrun #N1 #N2 …`, and one who types `epicrun` to run `backlogrun #E --only` (drop `--only` to also drain the backlog).
+**`queue` and `epicrun` were removed** — point `queue` to `backlogrun #N1 #N2 …` and `epicrun` to `backlogrun #E --only`.
 
-**Three rules decide what a run does when the work turns out not to be one Issue** — the split assessment every entry point applies identically (**its default is not to split**: separability **and** a scope clearly exceeding one verification gate — about 10 changed files, about 400 changed lines (test files excluded) — must both hold), a prerequisite discovered mid-run (a dependency, not parked), and a named `backlogrun` item that is not an epic. All bind only after a command has started, so all are read from the `workflow-commands` skill: `split-assessment.md`, `SKILL.md` → §2d, `backlogrun-child.md` → "When `#N` is not an epic", and `fullrun.md` / `halfrun.md` / `backlogrun.md` for the branch each entry takes.
+**The `josh epic:*` commands have their own skill** — `epic:audit`, `epic:next`, `epic:bundle`, and how an epic spans repositories. **Read the `epic-commands` skill before running any of them, before writing an epic that tracks a child in another repository, and right after filing an issue.** Three rules bind outside those commands:
 
-**The `josh epic:*` commands have their own skill** — `epic:audit`, `epic:next`, `epic:bundle`, and how an epic spans repositories. **Read the `epic-commands` skill before running any of them, before writing an epic that tracks a child in another repository, and right after filing an issue.** Three rules stay here because they bind outside those commands: **recording a decision removes that child's `needs-decision` label** (Tier A); **fixing what the audit finds is Tier A**, park only when the contradiction is a design choice nobody has made; and **an epic in another repository is referenced as `owner/repo#N`** (a bare `#N` resolves to this repository's issue). Canonical reference: `.claude/skills/epic-commands/SKILL.md`.
+- Recording a decision removes that child's `needs-decision` label (Tier A).
+- **Fixing what the audit finds is Tier A**; park only when the contradiction is a design choice nobody has made.
+- **An epic in another repository is referenced as `owner/repo#N`** — a bare `#N` resolves to this repository's issue.
 
 #### Explicit invocation required (MANDATORY)
 
-Never start a `kickoff` / `halfrun` / `fullrun` / `backlogrun` workflow (including their `#N` and `new` variants) unless the user has typed the keyword in the **current turn's prompt**.
+Never start a `kickoff` / `halfrun` / `prrun` / `fullrun` / `backlogrun` workflow (including their `#N` and `new` variants) unless the user has typed the keyword in the **current turn's prompt**.
 
 - Conversational requests like "implement X", "fix Y", "open a PR for Z" are **NOT** implicit invocations — do not infer authorization from the request shape.
 - Do **NOT** ask confirmation questions like "Shall I run `fullrun`?". A confirmation prompt is not a substitute for explicit invocation.
 - Instead, **prompt the user to type the command themselves**, with the exact phrasing: "Please run \`<command>\` to start this task." (e.g. "Please run \`halfrun new\` to start this task.").
 - This applies even when the user authorized a related workflow in an earlier turn — each invocation must be re-typed in the current turn.
 
-**A session cut inside a declared budget is not a new invocation** — a `backlogrun` cut and resumed is the one invocation a person typed, and it continues rather than waiting to be retyped; **`backlogrun` alone**, though — a `fullrun` cut still waits for the keyword. Single source: `backlogrun-steps.md` → "The session cut is inside the invocation".
+**A session cut inside a declared budget is not a new invocation — for `backlogrun` alone**: a resumed `backlogrun` continues, while a `fullrun` cut still waits for the keyword. `backlogrun-steps.md` → "The session cut is inside the invocation".
 
-**A `backlogrun` parks a child instead of stopping the run**, and its procedure is read from `backlogrun-park.md` → "park and continue", which is that rule's single source.
+**A `backlogrun` parks a child instead of stopping the run** — `backlogrun-park.md` → "park and continue".
 
 #### Mid-workflow stop notification (`confirmation`)
 
-Whenever you pause **any** run mid-execution to wait for the user — a `kickoff` / `halfrun` / `fullrun` stop, an upstream-Issue interrupt, a Tier C confirmation — you MUST send a Telegram notification **before** stopping, so the user is alerted off-screen. `halfrun`'s built-in stop before commit is a confirmation pause and follows this same rule.
+Before pausing **any** run to wait for the user — a command's stop (`halfrun`'s stop before commit included), an upstream-Issue interrupt, a Tier C confirmation — you MUST send this Telegram, so the user is alerted off-screen:
 
 ```bash
 pnpm josh notify --task-type confirmation --issue-url "<issue-url>" --body=$'<one-line reason>\n<what is needed from the user>'
 ```
 
-- Use `--body=...` (single token) when the body starts with `-`, otherwise `parseArgs` rejects it
-- Send only once per stop — do not spam if you re-evaluate within the same pause
-- Skip the notification when the stop was explicitly requested by the user in the same turn (they already know)
+- Send it once per stop; skip it when the user asked for the stop in this same turn.
+- Pass `--body=...` as one token when the body starts with `-`, otherwise `parseArgs` rejects it.

package/dist/CLAUDE.static.md

@@ -1,8 +1,8 @@
 # Project instructions
 
-- 既存のファイルと利用者の設定を尊重し、変更する前に内容を確認する。
-- 変更には検証を付け、実行した結果を報告する。
-- HTML、CSS、JavaScript のファイルがある場合は Prettier で整形する。
-- 画面に見える変更をした場合はブラウザーで表示と操作を確認する。
-- JavaScript または TypeScript の動作を変更した場合は、その動作を確認する自動テストを追加する。
-- Git や GitHub を使う場合だけ、それぞれのワークフローを利用する。
+- Respect existing files and the user's settings, and read a file's contents before changing it.
+- Verify every change, and report the results you actually observed.
+- Format HTML, CSS and JavaScript files with Prettier when the project has any.
+- After a change that is visible on screen, check its display and interaction in a browser.
+- After changing JavaScript or TypeScript behavior, add an automated test that checks that behavior.
+- Use the Git or GitHub workflow only when you use Git or GitHub.

package/dist/config-merge/index.js

@@ -56,15 +56,39 @@ var string_array_schema = z.array(z.string());
 var string_record_schema = z.record(z.string(), z.string());
 
 // scripts/config-merge/parse-jsonc.ts
-import strip_json_comments from "strip-json-comments";
+import { parseTree, printParseErrorCode } from "jsonc-parser";
+function property_key(property) {
+  return String(property.children?.[0]?.value);
+}
+function node_value(node) {
+  if (node === void 0) return void 0;
+  const children = node.children ?? [];
+  if (node.type === "array") return children.map((child) => node_value(child));
+  if (node.type !== "object") return node.value;
+  return Object.fromEntries(
+    children.map((property) => [property_key(property), node_value(property.children?.[1])])
+  );
+}
 function parse_jsonc(content) {
-  const stripped = strip_json_comments(content, { trailingCommas: true });
-  return json_object_schema.parse(JSON.parse(stripped));
+  const errors = [];
+  const tree = parseTree(content, errors, { allowTrailingComma: true });
+  const [first_error] = errors;
+  if (first_error !== void 0) {
+    throw new SyntaxError(
+      `Invalid JSONC: ${printParseErrorCode(first_error.error)} at offset ${String(first_error.offset)}`
+    );
+  }
+  return json_object_schema.parse(node_value(tree));
 }
 
 // scripts/config-merge/patch-json-key.ts
-import { createScanner, findNodeAtLocation, getNodeValue, parseTree } from "jsonc-parser";
-import strip_json_comments2 from "strip-json-comments";
+import {
+  createScanner,
+  findNodeAtLocation,
+  getNodeValue,
+  parseTree as parseTree2,
+  stripComments
+} from "jsonc-parser";
 
 // scripts/config-merge/json-format.ts
 var PRINT_WIDTH = 100;
@@ -124,7 +148,6 @@ function format_package_json(value) {
 `;
 }
 var json_format = {
-  inline_short_arrays,
   format_json,
   format_package_json
 };
@@ -147,7 +170,7 @@ function render_value(key, value, has_trailing_comma) {
   return document.slice(prefix.length, document.length - suffix.length);
 }
 function parse_root(content) {
-  const root = parseTree(content);
+  const root = parseTree2(content);
   return root?.type === "object" ? root : void 0;
 }
 function replace_range(content, range, text) {
@@ -193,7 +216,7 @@ function separator_offset(content, offset, limit) {
 function has_separator_after(content, offset, limit) {
   return separator_offset(content, offset, limit) !== -1;
 }
-function property_key(property) {
+function property_key2(property) {
   const key = property.children?.[0]?.value;
   return typeof key === "string" ? key : void 0;
 }
@@ -203,7 +226,7 @@ function target_value_end(property) {
 }
 function contains_comment(content, node) {
   const text = content.slice(node.offset, node.offset + node.length);
-  return strip_json_comments2(text) !== text;
+  return stripComments(text) !== text;
 }
 function rebalanced_value(content, key, value_node, has_trailing_comma) {
   if (contains_comment(content, value_node)) return void 0;
@@ -217,7 +240,7 @@ function rebalanced_value(content, key, value_node, has_trailing_comma) {
   return rendered === current ? void 0 : rendered;
 }
 function rebalance(content, property, has_trailing_comma) {
-  const key = property_key(property);
+  const key = property_key2(property);
   const value_node = property.children?.[1];
   if (key === void 0 || value_node === void 0) return content;
   const rendered = rebalanced_value(content, key, value_node, has_trailing_comma);
@@ -378,7 +401,20 @@ var json_list = {
   patch_json_list_field
 };
 
-// scripts/yaml/yaml-document.ts
+// scripts/lib/json-value.ts
+function parse_or_undefined(text) {
+  try {
+    return JSON.parse(text);
+  } catch {
+    return void 0;
+  }
+}
+function is_record(value) {
+  return typeof value === "object" && value !== null && !Array.isArray(value);
+}
+var json_value = { parse_or_undefined, is_record };
+
+// scripts/lib/yaml-document.ts
 import { load, loadAll, YAMLException } from "js-yaml";
 var EMPTY_YAML_REASON = "expected a document, but the input is empty";
 function load_yaml_or_empty(content) {
@@ -395,7 +431,7 @@ function parse_yaml(content) {
   return json_object_schema.parse(raw);
 }
 function is_mapping_document(value) {
-  return typeof value === "object" && value !== null && !Array.isArray(value);
+  return json_value.is_record(value);
 }
 function parse_yaml_documents(content) {
   if (content.trim().length === 0) return [];

package/dist/hooks/batch-guard.js

@@ -9,10 +9,10 @@ import {
   load_environment_file,
   notice_refusal_path,
   refusal_path
-} from "./chunk-GQ7KP45M.js";
-import "./chunk-NJQOBE6L.js";
-import "./chunk-WQ4OSBSS.js";
-import "./chunk-7VPDUBRN.js";
+} from "./chunk-XYUIQOW3.js";
+import "./chunk-QVLT52OV.js";
+import "./chunk-MKVGBQPA.js";
+import "./chunk-4JZDP5X5.js";
 import "./chunk-WPALEIRX.js";
 import "./chunk-RKUAD3I6.js";
 export {