planning-pipeline

name	planning-pipeline
kstack	true
description	Planning pipeline: write-plan, then hand off to review-plan-pipeline for the full review pass (plan-review, product-review). Spike mode skips the review pass and ends with decision capture. Only invoked by the kstack subagent.
disable-model-invocation	true

Planning Pipeline

You are producing an implementation plan and reviewing it to production-ready quality. Do not implement code.

Mode selection: If the request is a timeboxed question to de-risk ("Spike X", "figure out whether Y works") rather than a feature to scope, run in spike mode. Otherwise run in full mode (default).

Start by creating a todolist — full mode: Phase 1 (write plan), Phase 2 (review plan), Phase 3 (sync). Spike mode: Phase 1 (write spike plan), Phase 2 (sync + handoff). Mark Phase 1 in_progress after reading the write-plan skill.

Phase 1: Write the plan

Read .agents/skills/write-plan/SKILL.md in full and follow its workflow, using the matching plan mode (full, spike, or project). Author the plan under docs/plans/. Stay within the planning directory — do not implement anything.

Pause after the plan is written and present a summary to the user. Wait for approval before proceeding.

Phase 2: Review pass (full mode only)

Read .agents/skills/review-plan-pipeline/SKILL.md in full and follow its workflow.

Spike mode skips this phase — a timeboxed question doesn't need a product review pass.

Phase 3: Task tracker sync

Sync the plan to the task tracker (see context.md for project-specific sync instructions):

• If the plan references an existing task, update its status and post a progress comment
• If no task exists yet, the write-plan skill already offers to create one — this phase is a catchall
• Spike mode: create a new task from the prototype context if none exists

Skip if no task is linked and the user doesn't specify one.

Spike tail (spike mode only)

The pipeline ends after Phase 3 — the engineer runs the investigation. Tell the user the closing steps so they aren't lost:

1. Investigate within the timebox (read code, check constraints — if it turns into real building, that's prototyping)
2. When concluded, invoke .agents/skills/decision-capture/SKILL.md to record the outcome
3. On proceed, .agents/skills/handoff-to-implementation/SKILL.md creates the follow-up task

Final Output

After all phases, present a final summary:

• Plan location (file path)
• Key requirements validation findings (full mode)
• Review issues raised and resolutions (full mode)
• Product completeness assessment (full mode)
• Kestral task synced (or skipped)
• Recommended next step — full mode: hand off to kstack with an implement request; spike mode: the spike tail steps above (investigate → decision-capture → handoff)

Rules

• Follow the project's coding rules in .cursor/rules/ (read engineering-principles.mdc and meta-workflow.mdc).
• Use the AskQuestion tool to resolve ambiguity rather than guessing.

---

Project Context

Read context.md and apply it as additional project-specific constraints layered on top of this workflow. If it does not exist, skip this section.