Unify

• Every judgment rests on each input's own sources, quoted on both sides, and is made against a summit stated with its source.
• Real conflicts, drift, duplication and stragglers are told apart from tensions doing useful work, and only the problems get fixes.
• The result is a short ranked list of changes, each placed where it lands with the check that proves it, and no input is edited until the principal says yes.

Read LIFEOS/USER/CUSTOMIZATIONS/SKILLS/Unify/Config.md for the default LifeOS input set, the summit it is judged against, and any standing exclusions. CLAUDE_CONFIG_DIR overrides the installation root. When the user names no inputs and this file is missing, stop and say so, pointing at Setup.md; never guess a default set or summit. Named inputs need no config, but a summit still has to be stated from their own sources.

• Analysis means read-only. Unify recommends; it never edits an input, even an obviously broken one. Doctrine and identity changes, and anything a self-initiated analysis proposes, stay recommendations until the principal says yes (self-initiated or scheduled analysis loops recommend, and never change the system on their own).
• Routing-table entries are not sources. CLAUDE.md describes every subsystem in one line, and that line is often stale or marketing-shaped. Characterizing a component from it produces confident wrong findings; read its system doc or ISA.
• Vocabulary collisions look like conflicts (2026-10-08, first full LifeOS run). Two components using one word for different things (or two words for one thing) is drift, not opposed aims; that run's "spec" and "claim" findings were renames. Name it as drift; the fix is a rename, not a redesign.
• Long agent hand-backs get cut off. Brief each reader to write its full report to one scratchpad file and hand back only the path. If a reader is blocked mid-report, don't route the content around the block; name it in the report as a coverage gap.

Example 1: Default scope

User: "Run Unify"
→ Harmonize on the default LifeOS set from Config.md
→ Per-component aims, harmony map, conflicts vs tensions, unifying principle
→ Ranked recommendations, each filed as an Upgrades proposal on the principal's yes

Example 2: Named inputs

User: "Are the approval rules, the scheduling rule, and the monitoring layer in harmony?"
→ Harmonize on exactly those three
→ Where self-initiated loops, scheduled workflows and Vigil routing agree or collide
→ Recommendations scoped to those three

Example 3: Outside LifeOS

User: "Unify these three strategy docs" (paths given)
→ Harmonize on the documents
→ Same report shape, judged against the summit the docs share