version: "1.0.2" name: next description: Use at the start of a work session, or any time the question is "what should I pick up now" across the whole project. Triggers — "/engineer.next", "what's next", "what should I work on", "what should I do next", "where do I pick up".

next

The session-start skill — the human's re-entry point. Surveys every source of DAE state and answers one question: what should I pick up now?

Read-only and advisory. It changes nothing, produces no artifact, and emits no handoff (the recommendation is the whole output; a handoff would just restate it). This is a deliberate exception to the agentic-summary contract — next is a query, not a task that changes state.

next is the read-side bookend to session-summary: session-summary writes "here's where I left off, next tasks" at session end; next consumes that — and everything else — at session start.

When to use

Start of a session, or any "what now?" moment. Project-scope; surveys the whole project — there is no narrower scope.

Not for: loading context on a feature you've already chosen (prime-context); checking artifact consistency (consistency-check); a per-feature next-step (each skill's handoff already carries recommended_next).

Workflow

Step 1 — Resolve and survey

Resolve the methodology root + manifest via ${CLAUDE_PLUGIN_ROOT}/scripts/dae_resolve.py (see references/resolving.md). Then read, read-only:

• In-flight features — every features/*/feature.md (status) + progress.md (current checkpoint; whether it's blocked or ready to advance)
• Consolidation backlog — .engineer/consolidation.md if present (coverage backlog + triage order)
• Parked ideas — features/*/ with status: parked; .engineer/discussions.log
• Tracker captures (triage queue) — tasks a human added directly to the onboarded tracker: rows with no Slug yet (the reconcile() orphan-as-intake case, see references/tracker.md), or .engineer/inbox.md lines in local mode. These are quick-captured bugs/ideas/tasks awaiting triage — not yet a feature or fix.
• Pending human actions — recent handoffs/*.md (per-feature and .engineer/handoffs/) flagged human_action_needed: yes
• Session-log next-tasks — the latest session-log.md entry per active feature ("Next tasks")
• Dispatched to cloud — features whose latest handoff carries cloud_session_url (running on a cloud agent; PR pending). Surface as DISPATCHED — not actionable until the PR lands.
• CHARTER.md + manifest — autonomy levels and path overrides, for execution-mode advice
• Open fixes — .engineer/fixes/*.md via ${CLAUDE_PLUGIN_ROOT}/scripts/dae_fix.py list_open_fixes (status != closed)
• Stale merged branch — if the current branch is not main/master, run git fetch origin --quiet then check git merge-base --is-ancestor HEAD origin/HEAD (fallback origin/main). If true, the branch is merged and lingering — surface it as a STALE BRANCH item.

Step 1.5 — Offer branch cleanup (if stale)

If a stale merged branch was detected, defer to the post-merge skill — it owns the full cleanup flow (checkout, pull, branch -d, prune, tracker update). At autonomy high/medium, auto-invoke /engineer.post-merge before continuing to Step 2. At low, surface the finding and stop until the user invokes it themselves. Do not inline the cleanup commands here — keep next advisory and let the dedicated skill do the work, so the post-merge handoff lands in the audit trail.

Step 2 — Triage into seven buckets

NEEDS YOUR DECISION   — blocked features; handoffs flagged human_action_needed.
                        These stall progress — surface them first.
OPEN FIXES            — bug-fix artifacts not yet closed. Sub-priority within the bucket:
                         1. blocks_user: true AND workaround: none      ← top (user is hit right now)
                         2. blocks_user: true AND workaround: <text>    ← middle (user has a workaround)
                         3. blocks_user: false                          ← bottom
                         Within each tier, order by severity (critical, high, medium, low).
READY TO ADVANCE      — in-flight features sitting at a checkpoint that can proceed.
READY TO DISPATCH     — features / consolidation tasks that can go to a cloud agent
                        right now (bounded, automatable verification, no mid-stream
                        human input needed). At medium/high autonomy the dispatch
                        router auto-sends these (gated by dae_delegable.py).
DISPATCHED            — already running on a cloud agent (handoff carries
                        cloud_session_url). Show the session / PR link; review when
                        it lands. Not actionable until then.
TRIAGE                — tasks added directly to the tracker (a row with no Slug
                        yet) or to .engineer/inbox.md — bugs/ideas/tasks dropped
                        in for DAE to pick up, not yet a feature or fix. For each:
                        promote (→ /fix, /discuss, /feature-init) or defer. A
                        triage bug that blocks users ranks with OPEN FIXES top tier.
COULD START           — the next consolidation-backlog item; parked ideas worth
                        promoting; fresh work.

Step 3 — Rank and recommend

Priority order:

1. Unblocking first — a blocked feature or a pending human-action stalls everything downstream; clearing it usually beats starting something new.

1b. User-blocking defects — an OPEN FIX with blocks_user: true and no workaround ranks above starting fresh feature work; a real defect hitting users now is more urgent than forward progress on new scope. A TRIAGE capture that reads as a user-blocking bug ranks here too — promote it to a /fix first.

1. Triage priority / dates — consolidation-backlog order, feature target dates.
2. Dispatchability — if the human has limited time, favour surfacing what can be dispatched (freeing the human) over what needs them.

Present the seven buckets, then one top recommendation — the single best next action — with a one-line rationale and a suggested execution mode (you, local subagent, agent team, or cloud agent — for the last, the dispatch router gates via dae_delegable.py).

Promoting a TRIAGE item is the human's call — next only recommends. Work it now → promote to the right unit and start: bug→/fix, idea→/discuss, task→/feature-init (each reads the tracker row as intake, reuses its tracker_ref, and writes the assigned Slug back — no duplicate row). Later → leave it, optionally raise a priority/Status. Drop → set Status: wontfix.

Step 4 — Stop

next recommends; it does not act. The human picks, then invokes the relevant skill themselves (prime-context, discuss, feature-init, a consolidation task, …). No handoff is emitted.

Example shape

NEEDS YOUR DECISION (1)
  • 042-customer-export — plan.md has a charter deviation; approve amendment ADR-012 or revise
 
OPEN FIXES (2)
  • 2026-05-20-login-crash [critical] blocks_user=true workaround=none — login fails on iOS 17
  • 2026-05-18-export-timeout [high] blocks_user=true workaround="retry after 30 s"
 
READY TO ADVANCE (1)
  • 015-image-formats — at Checkpoint 2; ACs done, ready for /atdd:atdd
 
READY TO DISPATCH (2)
  • consolidation #1 core-image-generation — full ATDD coverage; remote one-shot
  • consolidation #3 mcp-connector — remote one-shot
 
TRIAGE (2)
  • bug: CSV export drops the UTF-8 BOM | blocks_user=yes — added to tracker, no slug yet
  • idea: dark mode for the settings page
 
COULD START (1)
  • parked: bulk-admin-export — discussed 3 weeks ago; promote if priorities allow
 
→ TOP PICK: fix 2026-05-20-login-crash. Users are hitting a critical crash on iOS 17
  with no workaround — this outranks new feature work. Run /fix to continue the
  fix workflow. Then approve 042-customer-export's plan to unblock its pipeline.

When NOT to use this skill

• You already know the feature and just need to start → prime-context
• You want full project state validated, not a recommendation → consistency-check --project
• You want to wrap up a session → session-summary

References

• Foundation Design — feature.md / progress.md / agentic summary schemas
• The DAE methodology page — the three-layer visibility model; the parallelism model (one human driving several pipelines)
• Sister skill: session-summary — the session-end bookend.
• Sister skill: fix — produces the OPEN FIXES artifacts surfaced here.