This page describes the pinned development build. There is no public release to install today.
The four entrances of the command ring move a work session from plan to close. /cleanup and
/maintenance act on the harness around that work. Cleanup reports accumulated problems and proposes responses;
maintenance rebuilds replaceable views from their authored sources and re-measures named harness
conditions. Neither is a general application-code cleanup command.
After several export sessions
The export code has been renamed twice, one historical-ID migration remains queued and a generated map still points to an old file. Without dedicated upkeep, you can search for leftovers, edit the map and prune the list by hand. That is sensible for a small repository. As the chores repeat, two risks grow: judgment-heavy deletion gets automated too eagerly, while mechanical generated work is maintained by hand and drifts again.
Coldstart separates those jobs because they have different owners.
Cleanup: sweep, group and propose
Cleanup runs the strict check roster and groups findings by the decision they need. Strict mode promotes selected warnings to failures; it does not expand the roster into universal diagnosis.
| Finding class | Export example | Cleanup's role | Owner of the response |
|---|---|---|---|
| Near-duplicate | Two helpers now appear to format the same CSV field | Show both and propose consolidation | Human confirms they truly have one job |
| Over-budget file | The export reference has grown beyond its recorded size band | Propose a split | Human decides the new ownership boundary |
| Stale generated view | A map still names the pre-rename file | Report the mismatch | Maintenance performs the mechanical rebuild |
| Deletion residue | A rejected background-job adapter appears unused | Present evidence and a proposed action | Human owns uncertain subtraction |
| Open queue item | Historical-ID migration remains open | Keep, re-date, escalate or propose removal | Human decides whether the work still matters |
Structural changes are proposals, not silent edits. Small formatting fixes already inside current scope may be applied when they are named. Work that this session did not complete is not discarded merely because it is old.
Maintenance: rebuild, compare and measure
A generated map is a projection: a replaceable view rendered from an authored source. The maintenance entrance regenerates machine-owned projections and compares the fresh result with what was stored. If the bytes differ, it has established one narrow fact: the stored view did not equal a fresh render from the current source and generator.
That difference does not prove that the source is current, the generator is correct or the content is wise. Regeneration detects represented disagreement; it does not make disagreement impossible.
The pass then runs the strict roster and compares internal measurements with their recorded envelopes. A moved score is an internal signal. It is not evidence of productivity, code quality or a better user outcome, and the command does not redefine the acceptable band to force green.
Finally it reports old open items and reference sets that have passed their review age. Reporting staleness is not permission to prune authored work.
When to use each entrance
Run cleanup after fast-moving work, before a release or when duplication and leftovers are hard to locate. Run maintenance after structural moves and renames, when a generated view looks wrong or on a deliberate review cadence. Keep both outside an ordinary work session unless the active plan explicitly owns harness upkeep; otherwise they introduce a different class of decisions into the session's boundary.
For a small project, manual search plus one regenerated file may be cheaper. These entrances earn their cost when the same classes of upkeep recur and need an inspectable result.
Limits
- Cleanup reports and proposes. An ignored report does not keep a project tidy.
- Strict green means the applicable represented checks passed in that mode and environment. It does not mean the project is healthy.
- Maintenance measurements compare the harness with its own recorded history, not an external benchmark.
- Neither entrance verifies the order app, its tests or its data. The application can be broken while the harness upkeep is green.
Next
That completes upkeep: two receipts and the two entrances that produce them. The next section is what has to be in place before any of it runs: setup.