grimoire-construct

[Required Reading]

1. Read ${CLAUDE_PLUGIN_ROOT}/GRIMOIRE-CONVENTIONS.md. The sections that apply to this skill are: § Construct (the single source of truth for the construct artifact — definition, folder layout, CONSTRUCT.md format, writer contract, self-containment), § .grimoire/ layout (the constructs/ folder structure), § Sub-agent spawning (research runs in sub-agents; the orchestrator stays lucid), § Skill composition, § Commits (the docs(grimoire): construct <name> commit), § User-facing language, § Pause-point pattern (the clarifying-questions and draft-review pauses), § Project context (read PROJECT.md; never write it), and § IDE-aware review (the draft surface). Not applicable: § TDD, § Historic, and the Spec → Plan → Execute page lifecycle do not apply — this skill is orthogonal (same family as grimoire-init / grimoire-note / grimoire-deps) and writes no page, no SPEC.md, and no HISTORIC.md entry. Do not restate any § Construct rule in this body — reference it.
2. If .grimoire/PROJECT.md exists in the project root, read it for project context. If it does not exist, proceed without it and, at the end of the run, suggest the user run grimoire-init once to enrich future runs (non-blocking per § Project context).
3. If .grimoire/HISTORIC.md exists, read it read-only for recent-page recency context. Absence is non-blocking — this skill does not write or depend on HISTORIC.md.
4. If invoked with a handoff brief — a path to (or pasted content of) a .grimoire/bag/drafts/construct-brief-<epoch>.md written by grimoire-spec's size-evaluation pause — read it as the seed of the objective. Memorize its exact path for the Phase 6 cleanup.

[Objective] Turn the user's larger-than-a-page objective (or the seeding handoff brief) into a construct: a flat, ordered list of page-sized reference-seeds persisted under .grimoire/constructs/NNN-<name>/, per § Construct. Each reference is a focused seed that grimoire-spec will later expand — through its normal full interview — into one real page. The construct is the roadmap-of-record; the pages it spawns are the actual work.

User's objective (or handoff-brief path / content):

$ARGUMENTS

[Hard constraints — orthogonal skill]

• Never write to .grimoire/HISTORIC.md. Never write to .grimoire/PROJECT.md. Never create a page folder under .grimoire/pages/.
• Authoring and rewriting construct files (CONSTRUCT.md and the references/ detail files) belongs to this skill only. grimoire-spec may only flip a single reference's status in place when it consumes one — never author. This is the § Construct writer contract; do not duplicate it here.
• Keep the construct flat and self-contained per § Construct: no nested constructs, no sub-references, no HISTORIC.md entry for the construct itself.

[Phase 1: Research] Per § Sub-agent spawning, run the research inside sub-agent(s) so the orchestrator's context stays lucid. The orchestrator stays in charge of the interview, the NNN resolution, and the construct writes.

• Brief each research sub-agent with a self-contained prompt naming the objective (or the handoff brief's content) and the project context from PROJECT.md. Equip it with Read, Grep, Glob, read-only Bash, and — when the objective's established shape depends on facts outside the repo — WebSearch / WebFetch.
• The research goal is the decomposition: analyze the codebase (and, when relevant, the established external shape of the objective) to propose a flat, ordered list of reference-seeds where each reference is sized to fit one normal page. A reference too big to spec in one page interview must be split; trivial slivers should be merged.
• The sub-agent returns the proposed ordered reference list (each with a short description and the rationale for its size/ordering) plus an explicit list of anything it was uncertain about. Do not invent facts — uncertainties become questions for Phase 2.