Przejdź do treści
Biblioteka skilli
AI i agenci · Automatyzacja

scaffold-init

/scaffold-init

Inicjalizuje w repozytorium przestrzeń roboczą AI z przepływem spec-driven dla Claude Code: pliki, konfigurację i infrastrukturę workflow.

Repozytorium
jig ↗
Licencja
MIT
Ostatnia zmiana repozytorium
26.10.2026

Ocena i bezpieczeństwo

100/ 100Klasa: dobry
Format25 / 25
Frontmatter z nazwą i opisem, poprawna nazwa, opis do 1024 znaków, nazwa zgodna z folderem.
Odkrywalność25 / 25
Opis mówi, kiedy użyć skilla, ma sensowną długość i jest w 3. osobie.
Treść35 / 35
Co najmniej 150 słów i do 500 linii, nagłówki, listy kroków, przykłady albo kod.
Higiena15 / 15
Bez zaślepek typu TODO, bez ścieżek Windows, z licencją.
Ryzykobrak sygnałów w skanie
Pliki w paczce
6
Skrypty
5
Komentarze HTML
0
Adresy URL w SKILL.md
0
Przestarzałość

Brak sygnałów. Repozytorium zmienione 26.10.2026, bez odwołań do wycofanych modeli i API.

Licencja

MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.

Co sprawdzić przed instalacją
  • 5 skryptów w paczce. Przejrzyj je przed instalacją, bo Claude może je uruchomić.

Skrypty: copilot_hook_adapter.py, copilot_permissions_floor.py, governance.py, scaffold.py, stocktake.py

Co robi i kiedy

Inicjalizuje w repozytorium przestrzeń roboczą AI z przepływem spec-driven dla Claude Code: pliki, konfigurację i infrastrukturę workflow.

Oryginalny opis (EN)

Initialize an AI-native development workspace with spec-driven workflow infrastructure. Use when starting a new project, setting up Claude Code on a codebase for the first time, or when the user says scaffold, initialize, set up AI workflow, onboard this repo, or similar. Do not use for adding an individual skill or config to an already- scaffolded project — that is handled by the relevant tier skill directly.

Jak działa

  1. What this skill does
  2. How to use
  3. Codex custom-agent install
  4. Q&A flow (slice 001-05)
  5. Output
  6. Immediate next steps to surface to the user
  7. Constraints
  8. Gotchas
Sekcje z nagłówków SKILL.md, w oryginale · 156 linii, 1 160 słów
Podgląd SKILL.md
---
name: scaffold-init
description: >
  Initialize an AI-native development workspace with spec-driven workflow infrastructure.
  Use when starting a new project, setting up Claude Code on a codebase for the first
  time, or when the user says scaffold, initialize, set up AI workflow, onboard this
  repo, or similar. Do not use for adding an individual skill or config to an already-
  scaffolded project — that is handled by the relevant tier skill directly.
user-invocable: true
---

> Spec 001 is fully implemented: greenfield-scaffold, doc-content, signal-detection,
> deferred-decisions, and Q&A wizard.
> See [docs/specs/001-scaffold-init/spec.md](../../docs/specs/001-scaffold-init/spec.md).

## What this skill does

Generates an AI-native development workspace by copying templates from
`${CLAUDE_PLUGIN_ROOT}/templates/` into a target directory. Detects project
signals from the filesystem (LLM/agent files, CI, tests, team), runs an optional
Q&A flow to let the user override those signals, and selects tiers accordingly.
Tier 0 always installs; Tier 1 installs when test signals are present; Tier 2
is offered (not auto-installed) when LLM/agent signals are present.

## How to use

1. Determine the target directory. Default: the current working directory.
   If you're unsure, **ask the user once** before scaffolding.
2. Check if the target already has a `scaffold.json` or `docs/specs/` — if so,
   the project is already scaffolded. **Stop and tell the user** rather than
   overwriting.
3. **Run the Q&A flow** (see next section). Collect answers as flag values.
4. Invoke the wizard with the collected flags:
   ```bash
   python3 "${CLAUDE_PLUGIN_ROOT}/skills/scaffold-init/scaffold.py" \
     [--runtime <name>] [--team|--solo] [--has-ci|--no-ci] \
     [--has-tests|--no-tests] [--plans-ai|--no-ai] \
     [--in-repo] \
     <target-dir>
   ```
   With no machinery flag the wizard scaffolds **plugin mode** — the lean default
   (docs + primer only; jig runs from the installed plugin). Pass `--in-repo`
   only when the sixth question is answered "yes".
5. Read the wizard's stdout summary and report back to the user. List the files
   that were created and the immediate next steps.

## Codex custom-agent install

Codex scaffold mode (`--host codex`) writes project-local custom agents as
TOML under `.codex/agents/`. For Codex plugin users who want jig's role agents
globally available, run the explicit post-install helper:

```bash
python3 "${CLAUDE_PLUGIN_ROOT}/skills/scaffold-init/scaffold.py" --install-codex-agents
```

The default destination is `~/.codex/agents`. Use
`--codex-agents-dir <dir>` to target a different Codex agents directory.
The helper refuses to overwrite user-owned `jig-*.toml` files unless
`--force` is passed.

## Q&A flow (slice 001-05)

Ask each question in order. **Each question is independently skippable** — if the
user says "skip", "I don't know", "unsure", or similar, do not pass the flag
(the wizard's filesystem inference handles it).

1. **Runtime/language** — "What runtime or language is this project?
   (e.g. Python, TypeScript, Go, Rust, mixed, unsure)"
   → `--runtime <name>` if answered; omit if skipped or unsure.
2. **Team context** — "Solo project or team setting?"
   → `--team` for team, `--solo` for solo; omit if skipped (uses git-author detection).
3. **Existing CI** — "Does the project already have CI configured?"
   → `--has-ci` for yes, `--no-ci` for no; omit if skipped.
4. **Existing tests** — "Does the project already have a test suite?"
   → `--has-tests` for yes, `--no-tests` for no; omit if skipped.
   This affects whether tier-1 (`tdd-loop` and friends) is auto-installed.
5. **LLM/agent work planned** — "Will this project involve LLM or agent development?"
   → `--plans-ai` for yes, `--no-ai` for no; omit if skipped.
   This affects whether tier-2 is offered.
6. **Machinery vs. plugin** — "Will this project ever run jig where the plugin may
   NOT be installed — CI, cloud agents, or teammates without jig? (yes → copy
   jig's machinery into the repo; no → use the installed plugin)"
   → `--in-repo` for yes; **omit for no** (plugin mode is the default). Skipping
   selects plugin mode.
   This is the one architectural question in the flow (it decides repo topology —
   what gets committed), so it won't appear in the project's own docs. Default to
   plugin mode (lean repo, jig updates flow from the plugin); choose `--in-repo`
   only for the self-contained cases above. See
   [ADR-0041](../../docs/decisions/adr-0041-scaffold-defaults-to-plugin-mode.md).

Skipping every question is the legitimate "pure inference" mode (slice 001-03
behavior) — the wizard infers from filesystem signals alone, and defaults to
plugin mode. Do not invent answers when the user is unsure.

## Output

After running, the target directory contains (plugin mode — the default):
- `CLAUDE.md` (with Hot Cache section, project name substituted)
- `docs/` (architecture, workflow, conventions, refinement-todo, inbox, memory/, specs/, decisions/)
- `.claude/hooks/` (empty — project-specific gates can go here)
- a **project-scoped permissions file**, on hosts that provide one — seeded with
  the ADR-0013 destructive-command deny floor (`git push --force`,
  `git reset --hard`, `rm -rf`). Hosts with no project-scoped permission surface
  get no such file and no deny floor.
- `.gitignore` (secret-ignore floor)
- `scaffold.json` (install-state manifest; `scaffold_mode: "plugin-only"`)

> For what your project actually received: `scaffold.json` records the mode and
> host, and your host's own settings file (if it has one) carries the deny
> floor. (This section is deliberately host-neutral — it is machine-translated
> per host.)

In plugin mode jig's skills, agents, and hooks stay under the installed plugin
and run from the plugin root — no *machinery* is copied into the repo. The one
exception is the permissions file above: `permissions.deny` lives in the
project's own settings and no plugin mechanism can inject it, so the scaffold
writes it in both modes wherever the host supports it (ADR-0041 OQ1). It
carries **no** hook registrations.

The wizard's stdout summary states the mode and why.

With `--in-repo`, the target additionally gets a self-contained copy of jig's
skills, agents, hook scripts, and templates under the host's runtime directory,
plus the host's own hook-registration file (`scaffold_mode: "in-repo"`). Choose
it only for CI, cloud agents, or teammates without jig installed.

Every scaffolded doc carries `Status: Draft (wizard-generated)`.
`docs/memory/people.md` is NOT created (solo-project default — team detection is slice 001-03).

## Immediate next steps to surface to the user

After scaffolding succeeds, tell the user:
1. Open `CLAUDE.md` and fill in the Hot Cache section with project-specific terms.
2. Open `docs/refinement-todo.md` to see what was deferred.
3. Your first spec will be `001` — reserve it with `/jig:spec-workflow new <slug>`
   (or write `docs/specs/001-<feature>/spec.md` by hand). The scaffold seeds only the
   genesis worked example `000-adopt-jig`; there is no placeholder spec to replace.
4. **Note:** `docs/conventions.md` is now gated. Edits require
   `JIG_CONVENTIONS_APPROVED=1` in the shell session.

## Constraints

- Do not invoke this skill in a directory that is already scaffolded (has `scaffold.json`).
- Do not overwrite an existing `CLAUDE.md` without explicit user confirmation.
- The wizard is deterministic — do not edit the generated files yourself before
  reporting back. The user should see exactly what `scaffold.py` produced.

## Gotchas

- The spec-gate hook for `docs/conventions.md` activates AFTER scaffold-init completes.
  It cannot gate its own creation (bootstrap paradox — documented and intentional).
- `templates/CLAUDE.md.template` is the source template; do NOT use the jig repo's own
  `CLAUDE.md` as a template — the two diverge over time.
- `${CLAUDE_PLUGIN_ROOT}` is the right env var inside the plugin. Don't confuse it with
  `$CLAUDE_PROJECT_DIR` (which is the target project's root after install).
- Signal detection (existing CI, LLM/agent files, team size) is deferred to slice 001-03.
  Until then, the wizard installs default tiers regardless of project context.
- **scaffold-init refuses if the target looks spec-driven but lacks `scaffold.json`.**
  Slice 008-05 introduced a second pre-flight check: if ≥3 of the four migrate
  triggers (`docs/specs/`/`docs/slices/`, `docs/decisions/`/`docs/adrs/`,
  `docs/workflow.md`, `docs/architecture.md`) are present without a
  `scaffold.json`, the helper raises `LooksAlreadySpecDrivenError` (exit 2)
  and points at `/jig:migrate`. Pass `--force` to scaffold over the existing
  tree anyway (NOT recommended — overwrites docs).
Pliki w paczce (6)
  • SKILL.md 9 KB
  • copilot_hook_adapter.py 19 KB
  • copilot_permissions_floor.py 8 KB
  • governance.py 19 KB
  • scaffold.py 202 KB
  • stocktake.py 5 KB

Wersje

RepozytoriumWynikLicencja
Ta strona · ramboz/jig100MIT
ramboz/jig100MIT
ramboz/jig100MIT

Identyczna treść występuje jeszcze w 1 innych folderach na GitHubie. Tu pokazujemy jedną, najlepiej ocenioną kopię.

Migawka bazy z 28.10.2026 · wersja paczki fd77b7c2f14e (skrót SHA-256 całej paczki)

Pobrania w ostatnich 30 dniach: 0 · ranking

Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.

Podobne skille

  • 90Project Scaffolding
    Przygotowuje Claude do projektu: tworzy CLAUDE.md, katalog .claude i workflow AI dla nowego lub istniejącego repozytorium.
  • 100ctx-init
    Inicjalizuje w repozytorium kodu pliki AGENTS.md, CLAUDE.md i katalog rules/ w trybie od zera, np. przy zakładaniu kontekstu projektu.
  • 100init
    Inicjalizuje projekt potoku benchmarku użyteczności agentowej dla SDK: tworzy config.json i szkielet nowej ewaluacji.
  • 100init
    Inicjalizuje repozytorium: tworzy AGENTS.md, wytyczne dla współtwórców i dokumentację nastawioną na agentów dla bazy kodu.
  • 89init
    Zakłada nowy projekt i tworzy plik CLAUDE.md z zasadami rozwoju, trasowaniem skilli, protokołem poprawiania błędów i dyscypliną dokumentacji.
  • 89nacl-init
    Przygotowuje start nowego projektu: plik CLAUDE.md z regułami rozwoju, wyborem skilli, protokołem poprawek i zasadami dokumentacji.

Uwagi i zgłoszenia

Wiadomość trafi do Bartosza Sokolińskiego. Adres e-mail jest potrzebny tylko do odpowiedzi. Szczegóły w polityce prywatności.