Plan Workflow
Turn a goal into a milestone-based, dependency-aware, scheduled, trackable plan.
Start
- Route-confirm (see
CONVENTIONS.md): one line stating this is the plan workflow and the suggested mode; wait for the go-ahead. - Confirm mode: default ⚙️ Engineer; the user may switch to 🧠 Philosopher.
- Establish the baseline: use any context the user has provided (notes, prior plans, constraints) as the starting point; fine-tune from current info. When a key input is missing, ask the user before scheduling — don't assume.
⚙️ Engineer mode (default)
Goal: a plan you can execute against.
- Define the goal — goal + measurable success criterion + deadline.
- Take stock — assemble the starting point from context + known constraints/resources; state your assumptions out loud ("I'm assuming X — correct?") and ask for any missing key facts.
- Decompose — milestones → tasks → mark dependencies.
- Schedule — timeline, critical path, buffers.
- Risks + Plan B — what's most likely to slip, and the fallback.
- Checkpoints — how to track progress and when to review.
- Capture — Save to
outputs/plans/plan_<english-slug>.md.
🧠 Philosopher mode
For thinking before scheduling: should this goal be pursued, what's the real objective? Then switch to Engineer mode to schedule.
Wrap-up conventions
- In the plan, separate: known facts / your assumptions / awaiting-confirmation.
- Live-search principle: verify external facts (deadlines, policies, requirements) online — don't rely on training memory.
- Keep a "current progress + next step" line so work resumes across sessions.
- Promote: if it's directional/reusable, offer to save it to long-term memory; one-off outputs stay put.
- Filenames use English slugs.