Planning

What gets done in what order, and what is deliberately out of scope.

Issue: #950

What is already done?

Step	Result
Survey the six layers	design.md, written against the tree, not from memory
Give workflows a drift gate	sync-workflows.js --check, Workflow Sync in validate, proven to fail both directions. #951
Build project-creation	The skill, wired into agenda-creation Phase 6 and the project-manager dispatch rules
Revise the creation spec	Six flat files, skill not workflow, project to issue to launch plan
Skill tiers	tier: required on all 34 skills, enforced by Skill Sync, proven by mutation
Workflow portability	repoRoot required on all three, contextHome on the two that need it
Make the voice rule checkable	A stated mode line, plus check-transcript-voice.mjs reading the transcript after the fact. voice-checkability.md

What happens next, in order?

Reordered 2026-08-22 after a working session produced six operator rulings. The three original items are still here, pushed down: nothing above them existed as scope that morning.

1. Ask her before evaluating, at day start

• agenda-creation ranks in Phase 2 and grills her in Phase 5, so every ranking is built on inference and checked afterwards.
• The proposal inverts that: open the day with a short set of steering questions ("any specific areas of focus today?") and rank against her answers.
• It answers what the focus ruling left open. With active/focus.md retired, nothing yet says who fills in the agenda header's Focus: line. Her answers do.
• Not settled: which questions, how many, whether they are tappable, and whether a day can open without them.
• (Operator side quest, 2026-08-22.)

2. Restructure agenda-creation.js to backlog-first

• It mines all four launch-plan stage files and 177 open issues.
• The agenda-creation skill explicitly forbids that: "Do not generate candidates from the launch plan or the issue tracker."
• It also never opens the focus source at all.
• Together those explain the 40-minute run that ranked the wrong things.
• Ruled 2026-08-22: the backlog generates candidates, the launch plan and the tracker become filters.

3. Retire active/ and boards/ from the context home

• Coupled changes, not deletions.
• active/ is recreated by context-bootstrap.mjs on every bootstrap, and read by the agenda-creation skill and test-rank.js.
• boards/ is named by context-bootstrap.mjs, context-dated.mjs, a test, and the project-manager skill.
• audits/ and memory-curation/ were deleted on 2026-08-22 and needed none of this.

4. DONE 2026-08-22: the PM skill refactor, the agenda format, and agenda-status

• project-manager, agenda-documentation and agenda-status are live in .agents/, synced, gates green.
• The standalone agenda template was deleted: two sources for one format is drift by construction, and all three writers now point at the agenda-documentation skill.
• Proposal copies deleted rather than left to drift. The record is in implementation.md and logging.md.

5. The four document-type skills

• project-creation carries a README template inline as a stopgap.
• One skill per document type, matching test-plan: planning.md, design.md, implementation.md, logging.md.
• Each one that lands replaces the corresponding inline template.
• Blocked on nothing.

6. Make command required

• Nine skills omit the field entirely, and sync-skills.js reads omission as false.
• So nine deliberate-looking choices were never actually made.
• A required field turns that into a sync failure.
• Small, and it depends on nothing above.

7. Hooks into .agents/

• The only layer outside the source of truth.
• Moving it makes the pairing rule checkable: a hook with no matching rule could fail the sync.
• Costs a real move plus settings rewiring, so it goes last.

What is deliberately out of scope?

• Migrating docs/planning/, docs/worklog/, or the testing runs. No migration, per the 2026-08-21 ruling. This is the process going forward.
• Reducing the skill count. 33 skills may be too many, but the tiering has to land first or the answer is a guess.
• Testing whether a skill is any good. Named as the real gap in design.md. The only honest test is whether an agent given a realistic task reaches for it, and that is a research problem, not a sprint item.

How does this relate to the current focus?

• The live focus is the app, the admin portal and the merchant portal working end to end. This project is not that.
• It was asked for directly, which outranks the ranking filter.
• Items 4 through 7 are the kind of work that should wait behind the focus rather than be picked up because they are listed here.