planning

Plan multi-step or multi-session work with acceptance criteria, coherent units, verification and resumable checkpoints. Discuss consequential goal/product/ architecture forks; update ordinary implemen

SKILL.md

GitHub

Planning — enough structure to finish the work

Home branch in Memora: root/skill/planning. The live contract is AGENTS.md.

Choosing a part

Planning itself runs inline on this page; the parts cover the two ends.

Situation Part
The outcome itself is still vague parts/define-goal/
Intent is fuzzy and wants exploring before a plan parts/brainstorming/
Decomposition and handoff templates wanted parts/writing-plans/

Rules

  • Plan of record is write_todos plus one Memora node. A plan file appears only when the owner asks for one: no PLAN.md by default, and projects/<name>/docs/PLAN.md is not a required source. Memora is the source.
  • Questions only on forks with consequences: goal, architecture, cost, data risk, external effects. Newly discovered files, technical details and root causes already understood are updates to the plan, not owner decisions.
  • Checkpoint = a Memora node + 1–3 lines in chat: what is done, what was checked, which unit comes next.
  • After a compaction, read the plan node and write_todos first, then resume — never re-derive work that is already recorded.
  • Understand the outcome, the existing base and the exact delivery constraints before proposing anything; show the approach with real alternatives where they matter, and never manufacture options for routine commands.
  • Finish coherent units with evidence and exit conditions; an unresolved branch stays labeled in the node rather than silently decided later.

Related

grill-me for the full interrogation of a plan, brainstorming, relevance-scan, sessions, dev.