Project Restructure
Principle expression
Primary: P07
Supporting: P12, P15, P04
Principle source
Use a host Sequence and interpretations when the host declares them. Otherwise
use this package's read-only fallback under references/sequence-snapshot.
Scope
Reconstitute an inherited project carrier: the repo still contains old shapes,
duplicate projections, stale commands, or pre-regeneration paths while the
accepted design, sequence, or target architecture has already moved on.
The objective is not a prettier tree. It is a carrier whose layout lets a later
actor recover the governing truth, retire what no longer owns a decision, and
take the smallest irreversible step that actually advances the new shape.
Do not use this skill when:
- The system's shape is still unsettled — route to
reframe, then graduate into
design-driven.
- The gap is only design-vs-code drift in an otherwise stable layout — use
design-driven audit.
- The work is a Sequence or principle admission question — use
principle-cultivation.
- The work is creating or testing one agent skill — use
skill-engineering.
- The user wants a one-off file delete without a governing plan — do it directly.
The code → carrier metaphor
| Code refactor |
Project carrier analogue |
| Monolith |
Repo with tangled legacy paths, duplicate mirrors, stale docs |
| Dead code |
Obsolete skills, abandoned blueprints, superseded commands |
| Extract |
Separate canonical source from projections (.claude/, .agents/) |
| Move module |
Relocate artifacts to the design-aligned home |
| Interface |
Accepted design/, sequence, regeneration target, adopted decisions |
| Implementation |
Concrete files, symlinks, generated copies |
| Rename |
Align naming to the new architecture without changing semantics |
| Delete / retire |
Remove a carrier only after its preserved function is recoverable elsewhere |
Use this vocabulary when classifying inventory rows.
Commands
/project-restructure inventory → Read and follow commands/inventory.md./project-restructure plan → Read and follow commands/plan.md./project-restructure execute → Read and follow commands/execute.md./project-restructure verify → Read and follow commands/verify.md.- No argument → Identify whether accepted design truth exists. If not, route to
design-driven or reframe. If yes, start with inventory unless a durable
restructure/INVENTORY.md already exists, then continue from the next missing
artifact.
Phases are sequential. Each phase produces a durable artifact under
restructure/ so later sessions can resume without replaying discovery.
Hard constraints
Never violate these while restructuring:
- Semantic source stays singular:
principles/SEQUENCE.md, adopted design
decisions, and declared host truth are not merged, duplicated, or silently
rewritten.
- Historical evidence in
principles/adopted/, principles/reviews/, and adopted
design/decisions/ is not deleted to reduce clutter.
- Destructive retirement, semantic moves, and authority-changing merges require
explicit human approval per wave.
- Packaged sequence snapshots are regenerated with
python3 scripts/sync-sequence-snapshot.py, not hand-edited.
- A projection (
.claude/, .agents/, installer output) is not canonical merely
because it exists; canonical source lives in the declared upstream tree unless
design says otherwise.
Handoffs
| Signal during restructure |
Route to |
| DESIGN.md or target architecture is missing or contradicted |
design-driven |
| A proposed cleanup changes sequence meaning |
principle-cultivation |
| A cleanup reveals a missing reusable skill gate |
skill-engineering |
| The initiative's destination, not its carrier, is unclear |
goal-driven |
| Build/acceptance evidence is needed before retiring runtime paths |
evidence-driven |
Completion standard
A restructure wave is complete only when it has:
- An inventory tied to source paths and governing design references.
- A plan that names hard constraints, disposition per row, and the smallest
valid wave boundary.
- Human approval for any destructive or authority-changing step in that wave.
- Verification evidence that canonical truth, snapshots, and required
inheritance still hold.
If any item is missing, leave the repo unchanged and report the gap instead of
improvising a broad cleanup.
1---2name: project-restructure3description: Restructure a repository or workspace carrier when accepted design has moved ahead of legacy layout — inventory concrete state, derive a functional map, plan minimal retire/merge/regenerate waves, and verify hard constraints after each slice. Triggers on "整理项目", "清理旧内容", "restructure the repo", "retire legacy paths", "repo cleanup", "carrier refactor", "保留新设计清理旧的", "project housekeeping after regeneration". Use after target architecture or design truth exists; not for unsettled paradigm exploration or Sequence change. Args — inventory, plan, execute, verify.4---56# Project Restructure78## Principle expression910**Primary:** P0711**Supporting:** P12, P15, P041213## Principle source1415Use a host Sequence and interpretations when the host declares them. Otherwise16use this package's read-only fallback under references/sequence-snapshot.1718## Scope1920Reconstitute an inherited project carrier: the repo still contains old shapes,21duplicate projections, stale commands, or pre-regeneration paths while the22accepted design, sequence, or target architecture has already moved on.2324The objective is not a prettier tree. It is a carrier whose layout lets a later25actor recover the governing truth, retire what no longer owns a decision, and26take the smallest irreversible step that actually advances the new shape.2728Do not use this skill when:2930- The system's shape is still unsettled — route to `reframe`, then graduate into31 `design-driven`.32- The gap is only design-vs-code drift in an otherwise stable layout — use33 `design-driven audit`.34- The work is a Sequence or principle admission question — use35 `principle-cultivation`.36- The work is creating or testing one agent skill — use `skill-engineering`.37- The user wants a one-off file delete without a governing plan — do it directly.3839## The code → carrier metaphor4041| Code refactor | Project carrier analogue |42|---|---|43| Monolith | Repo with tangled legacy paths, duplicate mirrors, stale docs |44| Dead code | Obsolete skills, abandoned blueprints, superseded commands |45| Extract | Separate canonical source from projections (`.claude/`, `.agents/`) |46| Move module | Relocate artifacts to the design-aligned home |47| Interface | Accepted `design/`, sequence, regeneration target, adopted decisions |48| Implementation | Concrete files, symlinks, generated copies |49| Rename | Align naming to the new architecture without changing semantics |50| Delete / retire | Remove a carrier only after its preserved function is recoverable elsewhere |5152Use this vocabulary when classifying inventory rows.5354## Commands5556- `/project-restructure inventory` → Read and follow `commands/inventory.md`.57- `/project-restructure plan` → Read and follow `commands/plan.md`.58- `/project-restructure execute` → Read and follow `commands/execute.md`.59- `/project-restructure verify` → Read and follow `commands/verify.md`.60- No argument → Identify whether accepted design truth exists. If not, route to61 `design-driven` or `reframe`. If yes, start with `inventory` unless a durable62 `restructure/INVENTORY.md` already exists, then continue from the next missing63 artifact.6465Phases are sequential. Each phase produces a durable artifact under66`restructure/` so later sessions can resume without replaying discovery.6768## Hard constraints6970Never violate these while restructuring:7172- Semantic source stays singular: `principles/SEQUENCE.md`, adopted design73 decisions, and declared host truth are not merged, duplicated, or silently74 rewritten.75- Historical evidence in `principles/adopted/`, `principles/reviews/`, and adopted76 `design/decisions/` is not deleted to reduce clutter.77- Destructive retirement, semantic moves, and authority-changing merges require78 explicit human approval per wave.79- Packaged sequence snapshots are regenerated with80 `python3 scripts/sync-sequence-snapshot.py`, not hand-edited.81- A projection (`.claude/`, `.agents/`, installer output) is not canonical merely82 because it exists; canonical source lives in the declared upstream tree unless83 design says otherwise.8485## Handoffs8687| Signal during restructure | Route to |88|---|---|89| DESIGN.md or target architecture is missing or contradicted | `design-driven` |90| A proposed cleanup changes sequence meaning | `principle-cultivation` |91| A cleanup reveals a missing reusable skill gate | `skill-engineering` |92| The initiative's destination, not its carrier, is unclear | `goal-driven` |93| Build/acceptance evidence is needed before retiring runtime paths | `evidence-driven` |9495## Completion standard9697A restructure wave is complete only when it has:98991. An inventory tied to source paths and governing design references.1002. A plan that names hard constraints, disposition per row, and the smallest101 valid wave boundary.1023. Human approval for any destructive or authority-changing step in that wave.1034. Verification evidence that canonical truth, snapshots, and required104 inheritance still hold.105106If any item is missing, leave the repo unchanged and report the gap instead of107improvising a broad cleanup.
lidessen