Art-Project for Claude Code

    

> 한국어 README · 📖 Wiki: English / 한국어

A Claude Code plugin: a pre-studio articulation scaffold for practice-based artistic research. Not an ideation engine. The plugin accepts the Penny / Ingold / Borgdorff critique that artistic ideation is non-linguistic, material, and inseparable from making — and scopes itself to the propositional articulation work around ideation: grant applications, doctoral expositions, residency proposals, collaborator briefings. The actual ideation happens in your studio, with material.

---

Companion plugin

art-project and art-paper are two sibling Claude Code plugins for practice-based artistic research, by the same maintainer, covering opposite ends of a project's life:

| | Plugin | Phase | What it scaffolds | |---|---|---|---| | ← you are here | art-project | before the work | Pre-studio articulation — impulse surfacing, tradition-tagged provocations, lineage positioning, a Concept Brief, self-critique rehearsal | | | art-paper | after the work | Practice-based art-paper authoring — inquiry, drafting, ACM citation, a SIGGRAPH Asia jury, acmart LaTeX → PDF |

The arc: art-project articulate the concept → make the work in your studio → art-paper write the juried paper. Each stands alone; together they span concept-to-publication.

> Lineage: both descend from academic-research-skills (Cheng-I Wu). art-paper forked the suite into the art-paper genre; art-project then pivoted from art-paper into the pre-studio phase.

---

Install (30 seconds)

/plugin marketplace add joonhyungbae/art-project
/plugin install art-project

That gives you one skill (art-ideation) with six modes, accessed via /art-project:* slash commands. Outputs are Markdown; no toolchain required.

> Plan note. Three of the six commands (socratic, rehearsal, ideate) request Claude Opus because they carry the load-bearing IRON RULES (intent classification, persona stability, multi-session synthesis). The other three (provoke, lineage, brief) default to Claude Sonnet. If your Claude plan does not include Opus access, socratic / rehearsal / ideate fall through to your plan's default model — most rules still apply, but persona and intent-classification discipline is weaker. The wiki has a model-tier note.

Try one of:

• /art-project:socratic — "I have a vague pull toward something."
• /art-project:provoke — "I'm stuck. Throw constraints at me."
• Natural language: "Guide me through a new project." The plugin auto-routes by intent and announces the routing transparently.

👉 Wiki — apesuite.org/plugins/#/art-project and QUICKSTART.md — full walkthroughs.

---

What it does

One skill (art-ideation), six modes.

| Slash command | What it produces | Load-bearing rule | |---|---|---| | /art-project:socratic | Concept Pull Map (impulse / fragments / constraints / refusals / residue) | No auto-convergence under exploratory intent — HARD-bound by per-3-turn intent classification | | /art-project:provoke | 8–20 tradition-tagged provocations with an Authentic Practice Boundary per method | Preserved unhelpfulness — silence after delivery, no auto-interpretation, no ranking | | /art-project:lineage | Lineage Map extending artist-supplied initial candidates (kin / opposition / blind-spot / unexpected-neighbor) | Mandatory training-data bias header; Korean / East-Asian default routing on KO sessions; honest self-description as retrieval, not ideation | | /art-project:brief | 10-field Concept Brief (proposition / anti-proposition / disconfirmation condition / Frayling-type declaration / …) | Stay-rough default — voice preserved; no auto-completion of gaps | | /art-project:rehearsal | 4-persona self-critique (Curator + Practitioner-peer + Theorist + Devil's Advocate) | Formative not decisional; mandatory disclaimer; consultative friction backed by ~/.art-project/rehearsal-log.jsonl | | /art-project:ideate | Long-running project file at ~/.art-project/projects/<codename>/project.md, across weeks (Smith & Dean iterative cyclic web) | One mode per session; no single-session pipelining |

Each mode wires to a 25+ entry methodology reference layer (Frayling, Borgdorff, Sullivan, Smith & Dean, Eno & Schmidt, LeWitt, Cage, Bogart, Bauhaus, Manovich, Penny, Dunne & Raby, plus Korean / East-Asian and HCI prior-art). Full reference at the wiki.

---

Design rationale

Cognitive-scaffold position (Clark & Chalmers 1998; Malafouris 2013; Penny 2017). Neither inert tool nor co-author. Five architectural commitments (generation–evaluation separation; tension-over-ranking; lineage-with-opposition; formative-not-decisional rehearsal; tradition-tag-with-Authentic-Practice-Boundary) operationalise specific practice-based research positions; tradition-tag attribution comes with explicit Authentic Practice Boundaries naming what each cited method requires that the plugin does not simulate.

> Example. For Cage chance operations, the plugin proposes the procedure (which I Ching method, which dice protocol); the artist throws the dice. The time the artist spends performing the procedure is part of the work. The plugin does not execute the chance operation, because doing so would re-route the constitutive feature of the method.

Detail: cognitive scaffold, Frayling typology, Authentic Practice Boundaries, measured harms, and the v0.2 synthesis spec.

---

Who it's for

For artists where propositional articulation is a bottleneck:

• early-career artists who have not yet absorbed artist-statement conventions
• artists writing across languages (e.g. a Korean-first artist writing English grant applications)
• doctoral candidates preparing expositions for venues such as the Journal for Artistic Research
• artists under grant or residency deadlines
• collectives needing a shared articulation document

Not for artists whose articulation is already fluent (use Claude directly), nor for traditions where articulation is constitutively unwanted (improvisational, ritual, oral). Naming the boundary is part of the design — see philosophy / measured harms §6.

Bilingual: English default, Korean / East-Asian routing on Korean sessions. Honest disclosure: the East-Asian section of the reference layer is the most under-developed (3 entries as of v0.1.0; expansion pending library work). On Korean sessions, lineage surfaces a training-data bias header and may still fall back to anglophone Korean-studies sources rather than Korean-language primary sources. Treat the Korean experience as seeded rather than fully parallel to the English one until v0.2 reference expansion.

---

License & attribution

CC-BY-NC 4.0. Share + adapt + attribute, non-commercial use only.

art-project (Version 0.1.0).
Pivoted from art-paper v0.1.0; ultimately forked from
Academic Research Skills (Cheng-I Wu) v3.9.4.2.
https://github.com/joonhyungbae/art-project

---

Provenance

Maintainer. Joon-Hyung Bae (joonhyungbae on GitHub; jh.bae@kaist.ac.kr). The v0.2 design synthesis (Frayling layered hybrid self-positioning, cognitive-scaffold framing, tradition-tag-with-Authentic-Practice-Boundary architecture, six-mode reshape, measured-harm disclosure) is the maintainer's work.

Lineage. academic-research-skills v3.9.4.2 (Cheng-I Wu) → art-paper v0.1.0 → art-project v0.1.0. The genre-neutral safety machinery (L3 citation-faithfulness gate, Concession Threshold Protocol, intent detection, routing discipline) is inherited unchanged from ARS. The maintainer keeps a local pristine ARS clone under ref/academic-research-skills/ for diffing; that directory is gitignored and not part of the published plugin.

Four-agent design critique (2026-05-24) — the v0.2 design was synthesised from four specialist agent critiques: artistic-research methodologist (PaR / Frayling / Borgdorff / Sullivan), HCI / AI-creativity researcher (Shneiderman / Cherry-Latulipe / Davis / Wordcraft-Sparks), practising-artist studio-side review, Devil's Advocate (Penny / Ingold / Borgdorff / Wittgenstein / Illich attacks on the premise).

---

Repo layout

art-project/
├── art-ideation/
│   ├── SKILL.md                       # the single skill spec
│   ├── agents/                        # 4 v0.2-aligned + 4 archived under deprecated/
│   ├── references/ + templates/       # in-skill references and (deprecated) templates
│   └── examples/                      # in-skill examples (v0.1-deprecated; see wiki)
├── commands/                          # 6 slash-command files (bare names: brief, ideate, …)
├── shared/references/                 # methodology reference + glossaries + routing
├── docs/                              # design spec / audits / verification / ops
├── ref/academic-research-skills/      # pristine ARS reference (gitignored; local-only)
├── hooks/ + scripts/                  # SessionStart hook + announce script
├── .claude-plugin/{plugin,marketplace}.json
└── README{,.ko-KR}.md, LICENSE, NOTICE, SECURITY, CHANGELOG, MODE_REGISTRY,
    POSITIONING, QUICKSTART

---

Changelog (recent)

See CHANGELOG.md for full history.

• v0.1.0 (2026-05-24, pivot from art-paper) — drops paper-authoring scope; pivots art-inquiry → art-ideation with six modes; rebuilds the reference layer (positionality + tensions + Authentic Practice Boundaries + restored critical edge on Penny / Borgdorff); adds HCI prior-art section; ships measured-harm disclosure.
• v0.2 internal milestones (2026-05-30) — runtime-readiness honesty sweep; intent-detection HARD-bound in socratic; Dialogue Health Indicator demoted to Devil's-Advocate sub-heuristic inside rehearsal; rehearsal friction backed by real ~/.art-project/rehearsal-log.jsonl log (was honour-system); ideate full-mode persistence at ~/.art-project/projects/<codename>/project.md (was artist-managed); v0.1-drift agents archived under art-ideation/agents/deprecated/. Documented in docs/V0.2-DESIGN-DECISIONS.md and verified in docs/V0.2-VERIFICATION.md.
• Wiki canonical → apesuite (2026-06-12) — the user wiki now lives at apesuite.org/plugins/; the MkDocs mirror previously deployed to GitHub Pages has been removed.