cell-divide

name	cell-divide
description	Reproductive function. Copies the root .claude/skills/ genome into a target section's .claude/skills/ location (or deeper), appends the parent commit SHA to each child SKILL.md lineage, and registers the new cell in lattice/knowledge/skills_registry. INERT in Phase 1 — do not execute recursive division yet.
arguments	[target_path] [parent_commit_sha] [section_name]
disable-model-invocation	true
user-invocable	true
source	lattice-internal (Phase 1 Amendment §2.2)
vendored_at	null
vendor_strategy	lattice-authored
local_adaptations	[]
lineage	[]

cell-divide — Polymorphic Skill Genome Reproduction

HARD CONSTRAINT — Phase 1: This skill is inert. The body below documents its eventual behavior. Do NOT invoke recursive division in this phase. Division begins only after Phase B (live execution + bootstrap verification on the M3 Max) completes successfully and a human approves a propose-decomposition output.

Purpose

Copy the genome (the contents of /.claude/skills/) into a target sub-tree (typically meta/harness/sections/<section>/.claude/skills/), preserving lineage and registering the new cell in the knowledge substrate.

Inputs

Argument	Required	Meaning
target_path	yes	Absolute path inside the repo where the new cell's .claude/skills/ should land. Typically meta/harness/sections/<section>/.claude/skills/.
parent_commit_sha	yes	The commit SHA of the parent genome being copied (root genome = first commit that introduced the skill tree; deeper cells = the commit at which the parent cell forked).
section_name	yes	Human-readable section name (e.g. schema, api, frontend, docs). Recorded in skills_registry.lineage_source metadata.

Behavior (Phase 2+)

1. Verify target_path is inside the repo and inside an meta/harness/sections/* subtree (defense-in-depth against arbitrary writes).
2. Snapshot the parent genome at parent_commit_sha. Use git archive or a plain recursive copy of the parent path's contents.
3. For each child SKILL.md, append parent_commit_sha to the YAML lineage: list. Order: youngest ancestor first → oldest last.
4. For each child SKILL.md that has no local_adaptations: key yet, add local_adaptations: [].
5. Insert one row per skill into lattice/knowledge/skills_registry (table already exists from migration 0015):
   • skill_name = <skill-name>
   • tool_name = "lattice-meta-harness"
   • skill_md_path = <target_path>/<skill-name>/SKILL.md
   • lineage_source = parent_commit_sha
   • content = <SKILL.md body>
   • created_at = NOW()
6. Commit the new cell to git as a separate commit with message feat(harness/<section>): cell-divide from <parent_sha[:7]>. Do NOT merge or push — the human reviews each division on a feature branch.

What this skill does NOT do

• It does NOT mutate the parent genome. Source is read-only.
• It does NOT invoke propose-decomposition. That is a separate, introspective skill.
• It does NOT recurse automatically. Each invocation copies exactly one level. To go deeper, the human must re-invoke after reviewing the previous level's health.
• It does NOT touch lattice/knowledge/skills_registry.committed_path if the row already has a non-null value. (Promotion paths preserve that field.)

Hard constraint reminder

Per Phase 1 Amendment §2.2 + §5:

• Do not run this skill in Phase 1.
• Do not run propose-decomposition in Phase 1.
• The skill exists on disk so its presence is auditable and its frontmatter is testable, but it remains inert until Phase B sign-off.