Maintaining and sustaining research software

Use this skill when the goal is to keep software usable over time rather than to ship a first version: setting up maintenance habits, managing dependencies and technical debt, deciding whether to keep, deprecate, or archive a project, and keeping its sustainability story honest. Unmaintained software degrades even with no code changes - dependencies age, environments shift, and the knowledge to run it erodes - so treat maintenance as a recurring cost, not a one-off.

Establish maintenance habits early

Calibrate effort to the user base, but build the habits before the software is widely used:

• Write a test suite and check coverage with a language-appropriate tool (pytest-cov for Python, covr for R). Without tests, every dependency or environment update carries unknown regression risk.
• Set up a CI pipeline that runs tests on every commit and on a schedule (e.g. weekly). Scheduled runs catch breakage from external changes even when nobody is actively developing. See the rseng-ci-cd and rseng-testing skills for the mechanics.
• Keep documentation current as part of maintenance, not after it. If you cannot install and run the software from scratch using only its README and install guide, the docs need updating. See rseng-documentation.
• Maintain a visible issue tracker (GitHub/GitLab Issues) so the maintenance backlog is shared, not held in one person's head.

Manage dependencies deliberately

Every dependency is a liability as well as an asset - it can change, deprecate, or introduce a security issue:

• Prefer a small, well-understood dependency tree drawn from already well-maintained projects over a large one.
• Pin exact versions with a management tool for the language (pip + venv, pip-tools, uv, or Poetry for Python; renv for R) so updates are deliberate and auditable. See the rseng-reproducible-environments skill.
• Automate update pull requests with Dependabot or Renovate so security and version bumps surface as reviewable changes rather than silent drift.

Communicate change clearly

• Use Semantic Versioning (MAJOR.MINOR.PATCH) so users can tell a breaking change from a feature from a bug fix and decide when to upgrade. See the rseng-publishing-releasing skill.
• Keep a CHANGELOG and update it with each release: a record of what changed, when, and why serves both users and your future self.

Reduce the bus factor

If only one person understands the software, it becomes unmaintainable the moment they are unavailable:

• Document key decisions, architecture, and operational knowledge in the repository itself, not just in someone's head.
• Share ownership with at least one other person who can act if the primary maintainer is away.
• Use regular code review to spread understanding of the codebase (see the rseng-version-control-review skill).
• Recruit community help: label low-barrier issues (good first issue), run maintenance sprints, and make contributing easy.
• Apply for maintenance-specific funding where it exists (funders increasingly recognise maintenance as a distinct cost).

Deprecate or archive honestly

When you can no longer sustain a project, say so. A prominent README notice, a repository archive, or an explicit deprecation statement is more helpful to users than silent abandonment. See the rseng-publishing-releasing skill for archiving mechanics.

Track and pay down technical debt

Technical debt is often unavoidable in research code written quickly to test a hypothesis; the danger is that it compounds until change becomes slow and risky:

• Distinguish intentional debt (a known, documented workaround) from unintentional debt (unclear code, missing tests, hardcoded values). The latter is more dangerous because it is harder to reason about.
• Record debt where it is visible: create issues or a tech-debt label, and add TODO/FIXME comments at the point of the problem with enough context for a future reader. Debt held only in memory is forgotten.
• Allocate protected time for maintenance and refactoring - a maintenance sprint, a fixed fraction of each cycle, or scheduled calendar time. Debt does not reduce itself.
• Refactor incrementally, not with a big rewrite: small, individually reviewable improvements (rename for clarity, extract a function, add a missing test). Ensure tests cover current behaviour before restructuring, or you cannot tell whether a refactor introduced a regression.
• Use static analysis (Ruff or lintr, SonarQube) to surface quality issues automatically and track metrics over time. See the rseng-code-quality skill for readability and structure guidance.

Reduce environmental impact

Environmental sustainability of computing is its own practice with its own skill: rseng-green-computing covers measuring energy and carbon (CodeCarbon, the SCI metric), reducing footprint in payoff order and carbon-aware scheduling. From the maintenance perspective, two habits matter here: include the compute footprint in the project's sustainability story (long-running services and repeated pipelines dominate), and revisit it at the same cadence as dependency and debt reviews - then follow rseng-green-computing for the how.

Working with this skill

The generated references.md beside this file lists the source material and pointers:

• references.md - verified Learn more pointers

Learn more (verified):

• https://opensource.guide/best-practices/ - best practices for open source maintainers
• https://www.software.ac.uk - Software Sustainability Institute
• https://chaoss.community/kb-metrics-and-metrics-models/ - CHAOSS community health metrics
• https://endoflife.date - end-of-life dates for dependencies

Related skills

Check whether any of these applies before moving on:

• rseng-archiving - retiring software needs archival deposit
• rseng-ci-cd - scheduled runs catch external breakage
• rseng-code-quality - incremental refactoring and static analysis
• rseng-contributor-onboarding - recruiting community maintenance help
• rseng-dependency-management - dependency update and audit mechanics
• rseng-green-computing - footprint review at maintenance cadence