User Wrap

user-wrap

A THIN front door for the sit-back-down moment. The operator has been away — lunch, overnight, a context switch — and wants one answer before touching anything: keep going, recycle the window, wrap it, or just close it? This skill composes three existing contracts and owns ZERO triage logic of its own:

• Orientation shape — /task-handoff --resume

([../task-handoff/SKILL.md](../task-handoff/SKILL.md) § --resume — in-window orient).

• Verdict + loss report — /session-wrap --advise

([../session-wrap/SKILL.md](../session-wrap/SKILL.md) § --advise — verdict + loss report) is the ONE owner of triage: signal collection, route scoring, the read-only salvage preview, the 4-verdict banner (KEEP GOING / RECYCLE WINDOW / WRAP & CLOSE / SAFE TO CLOSE), and the two-line loss report.

• State fields — current.md format, field semantics, path resolution, and

staleness are owned by .claude/references/task-state-schema.md (workspace reference, not published in this mirror) (cite it, never restate it).

Defect clause. Any triage constant appearing in this file or in user-wrap-authored output prose — a token threshold, a context ceiling, a staleness window, a route-scoring rule — is a DEFECT. Those numbers live in session-wrap's threshold table and the task-state schema doc; they reach user-wrap's output only inside the verbatim re-presented --advise report, never as this skill's own reasoning.

---

When to use / when NOT to use

Use at the return moment — the operator is sitting back down at an open window and wants a verdict before acting: "sitting back down", "should I keep going or wrap up", "am I safe to close this window". The name says wrap, but the skill serves the whole return moment — KEEP GOING is the most common verdict.

Do NOT use:

• Mid-task transition moments (a boundary just hit, context feels heavy, end of

day) → bare /session-wrap — it triages AND acts in one step; user-wrap would only add a detour.

• Checkpoint-only writes mid-loop → /task-handoff --loop.

---

Steps

Step 1 — Orient-lite (read-only)

Read /.claude/task-state/current.md (path resolution and field semantics per the schema doc above) plus quick git state: git log --oneline -5, git rev-parse --short HEAD, git status --porcelain (dirty/ahead counts per touched repo). Output the orientation block in task-handoff --resume's shape:

Resuming [Task field]: [Status] Next Action: [Next Action field verbatim] git: @ — dirty, ahead

One deliberate difference from --resume: do NOT auto-proceed yet. --resume proceeds without confirmation; user-wrap defers the proceed/stop decision to the verdict — a KEEP GOING verdict restores --resume's no-confirmation proceed in Step 4.

Absent or stale current.md (staleness per the schema doc — never restate its threshold here): say so in the orientation block ("no fresh task state — orientation from git only") and continue to Step 2; the --advise triage handles degraded signals itself.

Step 1 writes nothing and mutates nothing — no checkpoint, no commit, no current.md touch.

Step 2 — Delegate the verdict (never compute it)

Invoke session-wrap --advise via the Skill tool. That mode owns everything about the verdict — see session-wrap SKILL.md § --advise. user-wrap never computes a triage signal, salvage preview, or verdict itself, never re-scores the route, and never restates a threshold. If the delegated run fails, report the failure and stop — do not fall back to a self-computed verdict.

Step 3 — Present the verdict SUPER-clearly

Re-present --advise's report as the centerpiece of the output, verbatim and unmangled:

1. The advise triage line, as --advise printed it.
2. The ## banner — the first heading after the triage line, visually

unmissable.

1. Already durable: then Dies with this window: immediately under the banner, in

that order, with every SHA, count, timestamp, and path preserved.

Nothing may be inserted between the banner and the loss report, and nothing may paraphrase or summarize it — the operator's real question is "do I lose anything if I close this?", and the loss report exactly as --advise wrote it is the answer.

Step 4 — Act per verdict

Exactly one of four actions, keyed to the banner:

| Verdict | Action | |---|---| | KEEP GOING | Execute current.md's Next Action immediately — task-handoff --resume semantics: no confirmation ask, no "should I continue?". State the command being run, then run it. | | RECYCLE WINDOW | Invoke bare /session-wrap via the Skill tool — it re-triages and ACTS (durable state, git verb, Pick-up-here block ending in /clear). | | WRAP & CLOSE | Invoke /session-wrap --end via the Skill tool — the full end-window wrap. | | SAFE TO CLOSE | Print ONE line — everything is durable; the window can be closed — then STOP. No skill invocation, no write, no git verb. |

user-wrap itself never writes state or runs a git verb at any point: Steps 1–3 are read-only, and every Step 4 mutation happens inside the delegated skill (or, on KEEP GOING, inside the Next Action work itself).

---

Maintenance

The evals/ suite targets THIS front-door contract (created 2026-07-13, #305): 10 assertions across 3 categories in evals/evals.json (passing threshold 8/10), 4 scenarios in evals/test_scenarios.json (one per verdict path), and a golden corpus of 4 goods + 10 single-defect bads under evals/golden/ (manifest.json maps each bad to the one assertion it trips). Any edit that changes an output contract here (orientation block, re-presentation rules, per-verdict actions) must update the affected assertions and goldens in the same diff, keeping this footer's numbers equal to evals.json's. The verdict + loss-report CONTENT contract stays owned by session-wrap's --advise section — when that changes, re-check the re-presentation assertions here instead of duplicating the contract.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: aberson
• Source: aberson/claude-skills
• License: MIT

Install and usage instructions live in the source repository linked above.