Next
Claude Code plugins for spec-first development: every agent claim cites a source, every code change traces to a requirement, and ceremony scales to the weight of the task.
npx -y skills add NVZver/claude-marketplace --skill nextAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its author says it does
Copied from the file, not written here
Recommend what to work on next. Input: none (or a plain 'what's next' / 'recommend an order' question). Output: a plain 'what's next' gets a fast-path answer — the highest-priority backlog/not-started roadmap item quoted with a file:line citation, no agent dispatch; 'what should I pick' / 'sequence the backlog' dispatches the project-manager agent for dependency/risk/value sequencing and runs its returned pick gate. Reads ${specs_root}/roadmap.yaml on demand via scripts/roadmap-row.sh (whole-file read is fallback only).
SKILL.md
6.0 KB, as published. Nobody here has run it
Trace. On load, print first:
=============== [manager/skills/next/SKILL.md] [manager] ===============
Next
Recommend the next backlog item to work on. A plain "what's next" returns the highest-priority backlog/not started roadmap row directly (fast-path, no dispatch); a "recommend an order" / "what should I pick" question dispatches the project-manager agent for dependency/risk/value sequencing and runs its returned pick gate. Does not contain recommendation logic — that lives in the agent and ../../knowledge/sequencing-heuristics.md. The dispatch → gate → re-render loop is the shared contract at ../../knowledge/roadmap-orchestration.md.
Goal
Tell the user what to build next — instantly for a plain "what's next", with dependency/risk/value reasoning when they ask for an ordering or selection — without manually reading roadmap state or invoking the agent directly.
Input
- None required. The roadmap is the entry point; the agent reads ambient state (roadmap, pitches, branches, specs) internally.
- The fast-path contract at
../../../core/knowledge/fast-path-source-of-truth.md— governs the Step 0 branch that answers a plain "what's next" directly from the roadmap before dispatch.
Steps
-
Fast-path branch: "what's next" → cited roadmap item (before dispatch). When the question shape is a plain "what's next" / "what's the next backlog item" (per
../../../core/knowledge/fast-path-source-of-truth.md§"Question-shape detection"), do NOT dispatch the agent. Get the highest-prioritybacklog/not_starteditem of the${specs_root}/roadmap.yamlledger and quote it inline with itsfile:linecitation per the shared knowledge file's §"Citation format", then exit cleanly. Run the roadmap-row extractor (this repo:bash scripts/roadmap-row.sh, which prints that item + itspath:linedeterministically from the YAML — Pro-safe, zero model tokens) and quote its output — do not whole-file-read the ledger on this happy path. Only if the extractor exits non-zero (no ledger / no backlog item — its fallback contract) do you fall through to a model-sideReadof${specs_root}/roadmap.yaml. Reserve the full agent dispatch in Step 1 for explicit "recommend an order" / "what should I pick" / "sequence the backlog" questions — those need the agent's dependency/risk/value reasoning. Fall through to Step 1 — with an observable note — if the extractor reports no backlog item, the ledger is empty, or the question carries the ordering/sequencing intent above. Observable result: either the highest-priority backlog item is quoted with itsfile:linecitation and the skill exits, or an observable fall-through note and Step 1 dispatches the agent. -
Dispatch the project-manager and run its gate — per the shared contract. Dispatch the
project-manageragent with an explicit intent (intent: recommend-nextfor "what should I pick",intent: sequence-backlogfor "sequence the backlog") and run the returned pick gate per../../knowledge/roadmap-orchestration.md. The agent returns the sequenced candidate list as a pending gate (each candidate + re-sequence / exit, top-ranked as the recommended default); this skill presents it viaAskUserQuestion. Nolsa:discoverhandoff — selecting a pick to decompose ismanager:decompose's job. Observable result: sequenced recommendation delivered, pick gate resolved by the user.
Output
Either the highest-priority backlog row is quoted inline with a file:line citation (fast-path), or the agent's sequenced recommendation is delivered and the user's pick gate is resolved. No file writes — recommendation is read-only.
Example Output
[illustrative]
> what's next
.lsa/roadmap.yaml:21 — onboarding-checklist | Onboarding checklist | Should | backlog
(fast-path: highest-priority backlog item via scripts/roadmap-row.sh, no agent dispatch)
> what should I pick and why
Dispatching project-manager (intent: recommend-next)...
Sequenced (2 candidates): 1. onboarding-checklist (no deps, unblocks scaffold) / 2. plugin-scaffold (blocked).
Gate — pick next item: [1] onboarding-checklist (recommended) / [2] plugin-scaffold / re-sequence / exit
Constraints
- Orchestrator only — per
../../knowledge/roadmap-orchestration.md. Dispatch with an explicit intent, run the gate, re-render. No recommendation logic here. - Recommend-only. This verb never writes the roadmap or hands off to
lsa:discover. Usemanager:decomposeto break a pick into epics;manager:checkfor hygiene writes. Once a pick is decomposed into epics,manager:implementis the parallel build-execution entry point — it runs the epics in parallel (wave plan → isolated worktrees → gated PRs). Surfacing it here keeps the parallel engine findable from the recommend step (observation log C1 — entry-point discoverability,../../../.lsa/observations/2026-06-17-tripanchor-manager-implement.md:83"C1 — entry-point discoverability.manager:implementhard to find."). - Outputs follow
core/output— citation by link, never restated.
/manager:next — manual invocation.