From alive
Rebuilds context window from multiple sessions by dispatching subagents to analyze squirrel entries, files, logs, tasks, and merge conflicts. Use for resuming after breaks or parallel work.
How this skill is triggered — by the user, by Claude, or both
Slash command
/alive:session-context-rebuildThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
You're returning to work that spans multiple sessions — possibly conflicting ones. The context window is empty but the walnut isn't. This skill dispatches subagents to deeply reconstruct what happened, what changed, where conflicts exist, and what the merged state should be. Then you resume with full context as if you never left.
You're returning to work that spans multiple sessions — possibly conflicting ones. The context window is empty but the walnut isn't. This skill dispatches subagents to deeply reconstruct what happened, what changed, where conflicts exist, and what the merged state should be. Then you resume with full context as if you never left.
Dispatch one subagent to map the territory. It calls alive:session-history to get the lay of the land, then reports back:
working: field in squirrel YAML)╭─ 🐿️ context rebuild — scouting
│ Found 5 sessions on this walnut (last 2 weeks)
│ 3 sequential, 2 overlapping (possible conflicts)
│ 12 files touched, 2 bundles active
│
│ ▸ Rebuild scope?
│ 1. All 5 sessions (full rebuild)
│ 2. Last 3 only
│ 3. Just the 2 conflicting sessions
╰─
Based on scope, dispatch parallel subagents — one per session or one per domain:
Agent assignments:
_kernel/log.md, extracts all entries from sessions in scope. Returns decisions, rationale, timestamps.working: fields. Returns current state, what changed, any draft versions.tasks.py list to query task state. Returns what's done, what's in progress, what's blocked.Each agent returns structured findings. They don't summarise — they return the actual context with file paths, timestamps, and quotes.
Combine all agent findings into a single reconstructed context:
╭─ 🐿️ rebuilt context (5 sessions, 2 conflicts resolved)
│
│ Timeline:
│ +-- 2026-03-10 (session:a8c) — Started competitor research bundle
│ +-- 2026-03-12 (session:b4f) — Shortlisted 3 vendors, shared draft-01 with Sue
│ +-- 2026-03-14 (session:c7d) — CONFLICT with session:d9e (same day)
│ │ c7d: Changed pricing to per-seat model
│ │ d9e: Kept usage-based pricing, added enterprise tier
│ │ -> Recommended merge: per-seat with enterprise tier override
│ +-- 2026-03-18 (session:e2f) — Final draft-03 written
│
│ Current state:
│ Bundle: competitor-research (draft-03)
│ Key file: competitor-research/competitor-research-draft-03.md
│
│ Open tasks (4):
│ - [ ] Vendor site visits
│ - [ ] Final pricing comparison
│ - [ ] Sue's feedback on draft-03 (shared 2026-03-12, no response logged)
│ - [ ] Enterprise tier pricing model (from conflict resolution)
│
│ People: Sue Chen (shared draft), Ryn Okata (mentioned in vendor shortlist)
│
│ Unresolved conflicts (1):
│ - Pricing model: per-seat vs usage-based (sessions c7d/d9e)
│
│ ▸ How to proceed?
│ 1. Accept merged context, resume working
│ 2. Resolve conflicts first
│ 3. Show me the raw findings from each agent
╰─
On confirmation:
Two sessions conflict when:
_kernel/now.json with different next actions (path: _kernel/now.json)The conflict agent presents each side with evidence and recommends a merge. The human decides.
Don't over-swarm. Scale agents to the rebuild scope:
| Scope | Agents |
|---|---|
| 1 session | 2 — log agent + file agent (no conflict possible) |
| 2-3 sessions, no overlap | 3 — log + files + tasks |
| 2+ sessions with overlap | 4-5 — log + files + tasks + people + conflict |
| Full walnut rebuild | 5+ — all of the above, split file agents by bundle |
Each agent gets the specific file paths and session IDs it needs. No broad searches — targeted reads from known locations.
session-history lists what happened. This skill reads the actual files, dispatches agents to reconstruct state, and merges conflicts. History is a timeline. This is a context resurrection.
mine-for-context extracts knowledge from external source material. This skill reconstructs your own working state from your own session history and files. Different sources, different purpose, different agent strategy.
npx claudepluginhub alivecontext/alive --plugin aliveRevives previous sessions from YAML squirrel logs and transcripts to reconstruct context. Browses recent activity across walnuts, filters by topic/person/timeframe, flags unsaved sessions.
Restores context from a previous session so you can pick up where you left off, reading transcripts directly with no LLM calls or token cost.
Recovers session context after compaction by detecting in-progress work, loading knowledge artifacts, and summarizing what was being done and what's next.