session-handover

name	session-handover
description	Use at the START of a session to bootstrap from the previous session's state, and at the END of a substantial session (or when the user says "hand over", "wrap up", "write a handover", "what's the state for the next session", or work spans devices/agents). This repo's continuity runs on a chain of handover docs (SESSION-HANDOVER-*.md, HANDOVER.md) plus the failure ledger; reading the newest one first prevents redoing finished work, and writing one at the end keeps the next Claude/Copilot/GLM/human session from starting blind.
metadata	{"short-description":"Bootstrap from / write the cross-session handover docs"}

Session handover

See also spark-cluster-ops §1: SESSION-COORDINATION.md is the live, untracked cross-session ownership ledger (CLAIM/OWNS/DONE) — read + append it before launching expensive work to avoid double-launch across concurrent agents. It complements (does not replace) the dated SESSION-HANDOVER-*.md durable state.

Work here is picked up by many sessions across devices and agents. Continuity is carried by handover documents, not memory. This skill standardizes reading the latest one at the start and writing the next one at the end.

At session start — bootstrap (read, newest wins)

1. Newest master handover — the most recent SESSION-HANDOVER-YYYY-MM-DD*.md at repo root. NEVER trust a hardcoded filename (a stale pin in this very skill once pointed sessions at a 2-week-old handover): derive it — data/knowledge_graph/CURRENT.json → handovers.readFirst (regenerate with python tools/build_current_pointer.py), or ls SESSION-HANDOVER-*.md | sort. It has: git state at handover, what was proven, the exact next step, and "read first" pointers.
2. HANDOVER.md — HISTORICAL consolidation handover (2026-06-26); read only for history, never for current state.
3. agi-proof/failure-ledger.md — what is NOT yet proven (the real open-work list).
4. CHANGELOG.md top + recent main log (git log --oneline -15 origin/main) — what landed since the handover was written (the handover may already be partly done).
5. Note any caveat about local vs remote in the handover — origin/main is the source of truth; the local main ref has been a stale divergent lineage in this container before.

Reconcile: if git log shows the handover's "next" work already merged, don't redo it — move to the next open item. (This pairs with the git-discipline skill: fetch before trusting state.)

At session end — write the handover

Create/refresh SESSION-HANDOVER-YYYY-MM-DD.md (today's date) at repo root. Keep the proven structure of the existing file so the next session can scan it fast:

1. Git/repo state at handover — branch, HEAD vs origin/main, what's pushed, working tree.
2. What this session did — commits/PRs with numbers, and what each one proves within the no-overclaim ceiling (state verdicts as GO/NO-GO/candidate/VALIDATED, never overclaim).
3. Proven vs still open — and add a agi-proof/failure-ledger.md entry for anything measured that is not yet fully validated (single-seed, no third-party, etc.).
4. ▶ Next step — the single most valuable next action, with exact commands + pass/fail bars.
5. Read-first list — the files the next session should open before acting.
6. Don't-break list — the CI gates that must stay green (see the ci-artifact-drift skill).

Then update CHANGELOG.md if the session landed work, and confirm the doc is committed on the session branch (handover docs are tracked, not throwaway).

Rules

• Faithful reporting: if a step was skipped or a run failed, say so with the numbers. The repo's credibility is its discipline — never report a number without its CI, seeds, and judge families.
• A handover is a pointer to state, not a duplicate of it. Link to the artifacts (published-results.json, gate JSONs, ledger), don't paste large results.
• The newest dated handover is authoritative; older ones are history — don't edit them to "fix" the past, write a new one.