| name | kit-plan-next |
| description | Forward planning from the current state: run `cckit plan-next` to inventory what the kit can already do (skills, verbs, rules, docs) and propose what to build next — grounded in that inventory — then hand a chosen item off into `/kit-effort-new`. The proposed plan always carries a mandatory docs + README step.
|
| when_to_use | When you ask "given where we are, what should we build next?" — at the start of a chapter of work, after closing an effort, or when triaging a backlog against the kit's current capabilities. Not for ordering issues that already exist (that is `/kit-next` / `cckit plan`), and not for scoping an effort you have already chosen (go straight to `/kit-effort-new`).
|
| metadata | {"version":"1.0.0"} |
kit-plan-next — what to build next, grounded in current state
Answer one question: given everything the kit can already do, what is the highest-value next
effort? This wraps the cckit plan-next verb (the engine — never re-author its logic here;
kit-engine-boundary) and turns its proposals into a scoped effort via /kit-effort-new.
The verb introspects the CURRENT capability surface at runtime, so the plan is always consistent
with what ships today — planning pre-grounded in current docs/skills/rules means the executing agent
burns fewer tokens rediscovering context.
CRITICAL — context-anxiety rule
The plan cckit plan-next emits is human / orchestrator-facing only. NEVER paste it (or the
inventory) into a monitored model's prompt or context, and never wire it into an agent's context
automatically. Read it yourself, decide with the human, and hand only the CHOSEN item to
/kit-effort-new. See rules/plan-next.md.
1. Get the forward plan
cckit plan-next
Prints the current capabilities (skills / commands / rules / docs counts + samples) and a ranked,
grounded set of proposals. Rank 1 is always the mandatory docs + README step (docs-always rule —
it is never dropped, even for a tiny change). Append --llm for TOON rows {rank,area,proposal,next}
if you need to reason over the proposals structurally — but keep that output out of any monitored
context.
Run it from the cckit repo for a full inventory, or from a host project where the kit is installed —
there it degrades gracefully to whatever capability dirs exist (host docs only) and never crashes.
2. Choose with the human
Surface the proposals as a short, visual summary — lead with the human-meaningful outcome of each
item (its proposal), not the verb noise. Confirm which one to pursue. Do not invent scope beyond
what the inventory supports.
3. Hand off to /kit-effort-new
Once an item is chosen, scope it into an effort:
/kit-effort-new
Carry the chosen proposal into the effort's Goal, decompose it into sub-issues in Scope, and
always include the mandatory docs + README sub-issue — the planner surfaces it on every plan and
/kit-effort-new must preserve it (docs are part of "done," never a follow-up).
Output
- The current-capability inventory (counts + samples per area).
- A ranked, grounded proposal table — rank 1 is always docs + README.
- A chosen item handed to
/kit-effort-new as a new effort (Goal / Scope / For agents / Verification).
Rules
- Engine boundary — the planning logic lives in
scripts/lib/plan-next.sh (plan_next); this
skill only runs the verb and routes the result. Never re-implement the inventory or proposals here.
- Docs are mandatory — every plan carries a docs + README step; the effort you scope must keep it.
- Context-anxiety — planner output never enters a monitored model's context; it informs the
human/orchestrator's decision only.
- Don't plan what exists — for ordering already-created issues use
/kit-next or cckit plan;
for scoping an item you've already chosen go straight to /kit-effort-new.
- Scrub secrets from anything pasted into the resulting effort body (trace hygiene).