Migration & Wiring
## §6 Wiring After user.md and personality.md are created, `~/.claude/CLAUDE.md` must reference them in the Context Loading table so they are loaded at session boot. Patch to add to the `## Context Loading` table in `~/.claude/CLAUDE.md`: ```diff | Before... | Load this | +| Any session (user context) | `~/.claude/user.md` — who Alem is, constraints, trust model | +| Any session (agent identity) | `~/.claude/personality.md` — who John is, hard constraints, decision tree | | Any task | `~/system/agents/specialist-mapping.json` | ``` The two new rows are inserted at the top of the table (load first, most general). No other changes to CLAUDE.md in this MC. **Session boot sequence after wiring:** 1. `~/.claude/CLAUDE.md` loads (always, auto-loaded by Claude Code) 2. user.md loaded via Context Loading reference (Alem profile, constraints, trust) 3. personality.md loaded via Context Loading reference (John identity, hard constraints) 4. `bash ~/system/boot.sh` runs (tool-verified system state) 5. `node ~/system/tools/email-inbox.js pending` runs (CEO emails) Wiring does NOT change the existing session start steps — it prepends the two context files to the existing load order. --- ## §7 Migration Plan **Phase 1 — Create canonical shells (this MC's build task)** Create `~/.claude/user.md` as an empty structured file with section headers only. Create `~/.claude/personality.md` as an empty structured file with section headers only. Wire both into CLAUDE.md Context Loading table. No persona files modified. No content auto-generated. Deliverable: two skeleton files + 3-line CLAUDE.md patch. Estimated 30 minutes of build work. **Phase 2 — AI interview (live with Alem, ~20 minutes)** John runs the interview from §5 in a dedicated session. Alem answers all 13 questions. John captures verbatim then composes user.md and personality.md prose. Both files reviewed by Alem before saving as canonical. Deliverable: populated user.md + personality.md (60-80 lines each), Alem-approved. This is the only step that creates actual identity content. **Phase 3 — Persona reference update (separate MC, low priority)** After Phase 2, update 78 persona files to reference `~/.claude/personality.md` as the canonical John identity contract. Each file gets one added line: "For John's operating identity: see `~/.claude/personality.md`". No persona behavior changes in this step. Estimated effort: automated grep-and-append, ~1 hour. Do NOT do this in the same MC as Phase 1 or 2. No persona logic is altered — only a documentation pointer is added. --- ## §8 Acceptance Criteria The Pillar #1 build MC is complete when all of the following are true: AC1. `~/.claude/user.md` exists and contains populated content in all §3 sections (verified by wc -l >= 40). AC2. `~/.claude/personality.md` exists and contains populated content in all §4 sections (verified by wc -l >= 40). AC3. `~/.claude/CLAUDE.md` Context Loading table contains two new rows referencing user.md and personality.md (verified by grep). AC4. Alem has reviewed and approved both files in a live session (verbal confirmation captured in HiveMind post). AC5. Session boot on a clean Claude Code launch loads both files without error (verified by running a test session and confirming context is present in first turn). AC6. BookStack page published under Agentic OS shelf describing user.md and personality.md structure and how to update them. AC7. No persona file content was auto-modified in this MC — confirmed by git diff showing zero changes to ~/system/agents/definitions/*. --- ## §9 Risks and Non-Goals ### Risks **R1: Interview capture drift.** If John paraphrases Alem's answers rather than recording verbatim, user.md becomes a hallucination of Alem's preferences. Mitigation: record raw answers first, compose prose second, show Alem the draft before saving. **R2: CLAUDE.md already does this.** CLAUDE.md contains identity information. Adding user.md and personality.md risks duplication and divergence. Mitigation: Phase 2 of the build MC explicitly reviews CLAUDE.md and moves identity content to the new files, leaving CLAUDE.md as routing + ZAKONs only. This must be an explicit build step, not an afterthought. **R3: 131 persona files ignored.** If Phase 3 never runs, the 131 persona files continue to infer John and Alem identity independently. Mitigation: Phase 3 MC is filed as M-priority during Phase 1 build so it is not forgotten. **R4: Over-engineering the interview.** 13 questions risks Alem losing patience at question 7. Mitigation: questions are ordered by priority. First 6 questions in each bank are sufficient for a working user.md and personality.md. Q7 and Q6 (in their respective banks) are optional depth questions. ### Explicit Non-Goals - NOT changing behavior or content of any of the 131 persona files in this MC. - NOT building any automation that reads user.md or personality.md programmatically. - NOT generating user.md or personality.md content without the AI interview. - NOT modifying any ZAKONs in CLAUDE.md as part of this work. - NOT creating a new agent or skill for identity management. - NOT making any changes to ~/system/agents/definitions/ in this MC. --- ## §10 Cost Estimate **Design phase (this MC #99291):** $0 implementation cost. Spec writing only. **Build phase (separate MC):** - Phase 1 — shell files + CLAUDE.md patch: $0 (file creation, no inference needed) - Phase 2 — AI interview + composition: ~$0.05-0.10 (one Claude session, Sonnet, ~20 min) - Phase 3 — persona pointer update (separate MC): ~$0.02 (automated grep-and-append) - AC5 session boot test: ~$0.01 **Total estimated build cost: < $0.15** No infrastructure changes. No new services. No API calls beyond normal session cost.