| name | wrap-up |
| description | Session wrap-up sync for ADLC (Agentic Development Life Cycle) vaults. Before ending a session, checks git changes across the vault and co-located code repos, injects updates into the relevant code wikis, reflects shipped features / resolved gaps / new requirements into the product wiki, then refreshes hot.md and appends log.md. Uses context already loaded in the session, no full re-scan. Triggers on: "wrap up the session", "end session", "wrap up", "session end", "session wrap-up", "sync before exit", "/wrap-up".
|
wrap-up: Session sync for ADLC vaults
Run this when the operator signals the session is ending. The goal is to leave every wiki in the multi-wiki set coherent before exit, cheaply, using what is already in context.
When to run
- The operator says "wrap up", "end session", "session end", or similar.
- Best in an ADLC vault (product wiki plus code wikis under
services/), but the hot.md + log.md refresh applies to any vault.
Steps
- Use loaded context first. Do not re-scan the whole repo. Work from what this session already read and changed. Only run targeted git commands to confirm.
- Detect changes. In the vault and in each
services/<checkout>, run git status --short and git diff --name-only HEAD to see what changed this session. Note new or edited code, docs, and wiki pages.
- Down-propagate (code wikis). For each code repo that changed, update its Mode B code wiki: impl specs, ADRs, plans, module / flow notes affected by the change. If a feature was specified this session, write the impl spec into that repo's wiki so its agent can build it. If a changed repo carries a code graph (
graphify-out/graph.json) and its source changed this session, run /graphify-update on it (or flag it as a follow-up if time-boxed) — a stale graph silently misleads every worker that orients with it next session.
- Up-propagate (product wiki). Reflect the session's outcomes into the ADLC wiki: shipped features (
features/ proposed to shipped), resolved or new gaps (gaps/), new or changed requirements (requirements/, with stable IDs), stories and tests if generated. Keep traceability links current. Reconcile the defect route: every verifier FAIL this session must have both its bugs/ page and a backlog item tracing to it (see technical-planning.md dispatcher rules) — file whichever is missing.
- Archive delivered plans. Point-in-time planning docs rot into traps: code drifts until the plan misleads the next agent that reads it. For every feature that shipped this session, flip its spec / plan / PRD-like pages to
status: delivered (with a one-line "as-built diverges: see [[page]]" note when relevant), and flip or supersede any verification contract sections the ship made stale. Maintained as-built pages compound; stale plans must be visibly closed.
- Reconcile rollups.
index.md, overview.md, and any readiness dashboard must agree with the detailed records this session changed. In ADLC vaults this includes the two metrics-seam pages (see wiki/references/mission-control.md): reconcile meta/mission-control.md against the verdicts / FAILs / ships this session produced, and refresh meta/ba-activity.md from produced_by / feature / effort_estimate frontmatter (grep the frontmatter; do not LLM-read the vault). Refresh stale counts and claims everywhere (production vaults drifted to "scaffold only" overviews while 20 features were release-ready). For pages a decision made stale (data model, architecture overview, verification contracts), add an explicit "superseded by [[ADR-xxx]]" pointer or flip the stale flags — never leave a silent contradiction.
- Refresh caches. Overwrite
hot.md (under 500 words, hot-cache format) and append a dated entry to the TOP of log.md recording trigger, scope, skills run, counts, and follow-ups. Log entries are pointers — full records (verification, review, retro) belong on their own pages. If wiki/index.json exists, regenerate it (skills/wiki-query/scripts/build_index_json.py, see the wiki-query skill's generator section) — a stale locator misroutes every large-vault query next session. On Claude Code hosts, also refresh the token-usage rollup: python3 <this-skill-dir>/scripts/usage_report.py . regenerates wiki/meta/usage.md (cumulative per-session ledger + top-consumer / per-model / per-subagent-type breakdowns) from the local session transcripts, so the operator can see what drives consumption. Host-local derived view (see wiki/references/mission-control.md § usage): on hosts without ~/.claude/projects/ the script exits cleanly — skip without error.
- Report. Summarize what was synced across which wikis, and flag anything left for the human (sign-offs, prod verification).
Rules
- Markdown is the system of record. Do not generate Office files here (that is the export step, see
wiki/references/ba-suite-pipeline.md).
- Append-only
log.md: newest at top, never edit past entries.
- Keep PII out of the committed wiki; sensitive staffing / context stays in gitignored private notes.
- Do not commit or push unless the operator asks.