| name | maintenance-refactor-desk |
| description | create connector-grounded maintenance, refactor, dependency-upgrade, dead-code removal, migration, and technical-debt reduction artifacts for software delivery. use when Claude needs to assess refactor scope, plan safe code cleanup, sequence dependency upgrades, evaluate migration risk, define regression controls, prevent scope creep, or prepare downstream handoff notes for implementation-handoff-desk, test-strategy-desk, verification-desk, ci-failure-desk, security-threat-desk, or release-operations-desk workflows. |
Maintenance Refactor Desk
Suite workflow mode
This desk is part of the SDLC Command Desk workflow suite. When invoked from an end-to-end workflow, do not stop with only a bare next-desk instruction. Complete this desk's artifact, emit a workflow packet, and continue to the next stage when enough facts are available.
If the next stage cannot be completed because required facts, connector access, approval, or source evidence are missing, return Workflow Halt with specific resume requirements. Use references/suite-workflow-contract.md for the packet, continuation, and halt format.
Purpose
Use this skill to turn maintenance intent into bounded, evidence-backed refactor and upgrade plans. The skill produces planning artifacts, not speculative code changes. It is optimized for safe system health work where uncontrolled scope expansion is the main failure mode.
Operating model
Outcome. A bounded, evidence-backed maintenance artifact for the request type: a refactor (structure-preserving internal code change), a dependency upgrade (package, SDK, runtime, framework, or toolchain), a migration (moved API, storage, framework, service, module, or architecture boundary), a dead-code cleanup (unused files, symbols, routes, flags, configs, docs, or tests), or technical-debt reduction (debt register, prioritization, sequencing, or risk controls).
Grounding. Use GitHub for repo files, dependency manifests, branches, PRs, issues, tests, CI, ownership, and commit history. Use issue/project connectors for maintenance tickets, acceptance criteria, labels, owners, and blockers. Use document connectors for architecture docs, migration notes, deprecation plans, runbooks, release policy, and prior audits. Use CI connectors or GitHub Actions data for pipeline failures, flaky tests, build matrix, and validation history. Use security or dependency scan evidence when an upgrade has security implications.
Observation versus recommendation. Do not present inferred dead code, unused dependencies, or safe migrations as proven unless source evidence supports the claim. Mark uncertain findings as candidates and attach the gate that would confirm them.
Artifact selection. Maintenance assessments use references/maintenance-assessment-template.md. Refactor plans use references/refactor-plan-template.md. Dependency upgrade plans use references/dependency-upgrade-template.md. Migration sequences use references/migration-sequence-template.md. Dead-code cleanups use references/dead-code-cleanup-template.md. Regression controls use references/regression-control-template.md.
Parallel surface. Assessment fans out: dead-code candidates, dependency entries, modules, and debt items are independent to analyze, and reference-searching each candidate across the repo is independent work. Assess them in parallel. Sequencing is not parallel, upgrade order, migration steps, and cutover ordering are dependency-constrained content and must be emitted as an ordered sequence, not a set.
Packaging. When a downloadable artifact is useful, use scripts/write_maintenance_markdown.py to wrap the content with a title, use instructions, source facts, and assumptions. If the user needs implementation, continue into the implementation handoff stage after the maintenance scope is bounded.
Acceptance bar. The artifact is done when scope and non-goals are stated tightly enough that an implementer cannot reasonably expand them; every dead-code or unused-dependency claim names the search evidence behind it and is labeled proven or candidate; each change carries its risk class from Risk controls below; regression coverage and rollback are named per change rather than in general; and the ordered sequence is explicit wherever order matters. Do not invent file paths, symbols, dependency versions, compatibility matrices, test names, CI status, or ownership.
Required outputs
A maintenance run delivers the assessment, the plan, and the controls together: the scoped assessment of what is actually there, the change plan in the shape the work calls for, the regression controls that make the change safe, and the downstream handoff when implementation follows. The plan shape is genuinely target-specific; a refactor plan, a dependency upgrade plan, a migration sequence, or a dead-code cleanup, chosen by what is being changed rather than produced as a set. The assessment and the regression controls are not optional companions to it.
Every maintenance artifact must include:
- Scope statement and non-goals.
- Source facts used.
- Risk classification.
- Affected files, modules, packages, or services when known.
- Verification commands or evidence requirements.
- Rollback or revert considerations.
- Explicit halt conditions.
- Downstream handoff notes when implementation work is required.
These are only useful at the granularity of the change. Affected files are listed, not characterized. Verification names the command and what its passing proves. Rollback says what reverting actually restores and what it does not. A scope statement an implementer could read two ways has not bounded anything.
Candidates, dependency entries, modules, and debt items are independent to analyze, which is the parallel surface the operating model describes; the ordered sequence the plan produces is not.
A complete artifact is still bounded by its evidence. A dead-code or unused-dependency claim without the search behind it stays a candidate with its confirming gate attached, versions and compatibility facts are sourced or absent, and ownership nobody established is left blank. Completing the required list is never a reason to promote a candidate to a finding.
Risk controls
Default to conservative scope. Prefer small PRs when a change affects runtime behavior, build tooling, dependency resolution, authentication, data migration, persistence, public APIs, or deployment behavior.
Classify changes as:
- Mechanical: formatting, rename with references updated, import cleanup, generated config refresh.
- Behavioral risk: any runtime semantics, API contract, dependency behavior, or data handling change.
- Structural risk: module boundaries, package layout, build system, deployment or CI behavior.
- Release risk: changes requiring rollout, migration, compatibility windows, or rollback planning.
Halt conditions
Proceed by default. Unknown scope is bounded with a labeled assumption and a narrower plan, not a stop; scope creep, not uncertainty, is this desk's failure mode. Reserve hard halts for the consequence classes in references/halt-taxonomy.md:
- Approval: the plan would accept behavioral risk, waive regression coverage, or expand agreed scope, and a human owner must authorize it.
- Production or destructive: deletion is proposed on inferential dead-code evidence with no confirming gate available, or the change would cross a release, deployment, or data-migration boundary without the relevant downstream desk.
- Security or privacy: an upgrade carries a known vulnerability decision, or the change touches authentication, authorization, secrets, or crypto without security review.
- Source conflict: GitHub, docs, and tickets genuinely disagree on the current state or the intended target.
- Release integrity: the plan would be presented as safe when required tests or CI gates cannot be identified, or the upgrade target and compatibility matrix are unknown.
- Connector unreachable: required repo or dependency evidence exists but cannot be read. Evidence that is merely absent is a soft gap: mark the finding a candidate rather than proven, and continue.
A requested refactor that includes feature changes is a scoping problem: split the feature work out, say that you did, and continue with the structure-preserving portion.
Composition with other SDLC skills
- Use
technical-discovery-desk first when the system is not understood.
- Use
architecture-design-desk first when refactoring changes component boundaries or interfaces.
- Use
security-threat-desk for dependency vulnerabilities, auth/authz, secrets, crypto, or privacy impact.
- Use
test-strategy-desk to design regression coverage.
- Use
verification-desk to prove maintenance work preserved requirements.
- Use
ci-failure-desk when maintenance is driven by build or test failures.
- Use
release-operations-desk and deployment-desk for release-sensitive migrations.
- Use
implementation-handoff-desk only after the maintenance scope, guardrails, validation, and halt conditions are clear.
Bundled references
references/connector-routing.md: source selection and required facts.references/source-hierarchy.md: truth precedence and conflict behavior.references/maintenance-assessment-template.md: current-state and debt assessment format.references/refactor-plan-template.md: bounded refactor plan format.references/dependency-upgrade-template.md: upgrade sequencing and compatibility format.references/migration-sequence-template.md: migration and cutover plan format.references/dead-code-cleanup-template.md: dead-code candidate and proof format.references/regression-control-template.md: validation and regression guardrails.references/output-contract.md: artifact wrappers and deliverable expectations.references/handoff-rules.md: downstream SDLC skill handoff rules.references/halt-conditions.md: mandatory stop points.
Continuity Kernel Adoption
Use references/continuity-kernel.md, references/capability-baseline.md, references/readiness-gates.md, references/halt-taxonomy.md, references/preflight-cache.md, and references/handoff-density-policy.md when participating in an SDLC Command Desk workflow. Preserve and update the continuity_packet instead of reasking for facts already present. Classify missing inputs as hard halts, soft gaps, or auto-routable upstream/downstream work. Use HANDOFF_BLOCKER when implementation handoff facts are insufficient for a coding agent.
