Adds a pure-click meta-composer that converts a free-text task description into the right durable KeiSeiKit artefact — agent, skill, hook, rule, or block — by composing existing primitives: - Phase 1 intake (one free-text DESC + target-type click) - Phase 2 wave-based decomposition (delegates to /research for heavy tasks) - Phase 3 prior-art grep sweep across _blocks/, _manifests/, skills/, _bridges/, hooks/ + optional PROJECT-E + external docs - Phase 4 gap analysis (AskUserQuestion multi-select) - Phase 5 math-first architecture proposal (derive-first, "what is unnecessary" pass, Constructor-Pattern check) - Phase 6 block augmentation — drafts + persists new _blocks/<slug>.md on user click; kit gets smarter each session - Phase 7 recipe assembly — branches by type, hands off to /new-agent (agent) or /escalate-recurrence (hook / rule) Split into SKILL.md (121-LOC index) + seven phase-*.md files (each <160 LOC) to stay under the 500-LOC/file Constructor-Pattern limit. Minimum 6 AskUserQuestion calls per session enforced by the pipeline; every referenced block / skill / manifest / bridge path verified to exist on disk before commit (RULE 0.4 — no fabricated references). Writes only to _blocks/<slug>.md and skills/<slug>/SKILL.md; handoffs to /new-agent and /escalate-recurrence own their own files.
2.3 KiB
2.3 KiB
Phase 5 — Architecture proposal (math-first)
Compose the architecture by following _blocks/rule-math-first.md.
5a — Expression first
One to three lines describing which primitives combine, in which order, with which invariants. Use this shape:
artefact = compose(block_A, block_B, ..., block_N)
where block_* ∈ {_blocks/, newly drafted, skills/, _manifests/}
invariant: <one-line, e.g. "every cube <200 LOC, every handoff verified">
5b — What is UNNECESSARY?
For each block listed, justify why it's in. If a block can be removed
without losing the user's goal — remove it. Derive-first: explicit claim
"this is the minimal decomposition, nothing removable". Follow the checklist
from _blocks/rule-math-first.md:
- Learned parameters / free knobs? WHY? Determined by input?
- Separate blocks for similar concerns? WHY? Can a single block cover both?
- Gate / wrapper layers? WHY? Is a direct reference enough?
5c — Constructor Pattern check
Each output cube must be single-concern, file < 200 LOC, function < 30 LOC. If the proposed assembly violates this, split before proceeding.
5d — Count
Show the numbers explicitly in the preview:
- New files: N
- Edits to existing files: M
- Total lines of markdown to be written: L
5e — Preview + confirm
Preview as plain text in chat, then:
{
"questions": [
{
"question": "Architecture OK?",
"header": "Architecture",
"multiSelect": false,
"options": [
{"label": "Confirm", "description": "Proceed to Phase 6 block augmentation (if any gaps) then Phase 7 assembly"},
{"label": "Revise component N", "description": "One component's decomposition or reuse choice is wrong — reply with one free-text line"},
{"label": "Remove something", "description": "You see a block that's not strictly necessary — reply which one"},
{"label": "Abort", "description": "Stop — nothing gets written"}
]
}
]
}
On Revise / Remove → ONE free-text prompt, regenerate the architecture,
re-preview.
Verify-criterion
- User clicked Confirm.
- The expression (5a) is present and < 3 lines.
- The "what is unnecessary" pass (5b) has been applied and is visible in the preview.
- Constructor Pattern check (5c) passed.