research-project-maintenance

Research Project Maintenance

Keep the repository usable as both software project and scholarly record.

Read First

references/repository-contract.md
references/output-contracts.md
references/external-skill-recommendations.md

Maintenance Tasks

create missing research folders
update AGENTS.md
update README layout
update configs/agent-stack.yaml
update configs/capabilities.yaml
update docs/agent/ workflow and capability records
update docs/agent/capability-profile.md
update docs/agent/mcp-setup.md
update docs/agent/project-quality.md
update wiki templates
update artifacts/badge-evidence-ledger.csv
add structural tests
separate raw sources from derived outputs
move notebook logic into src/
add reproducibility commands
keep maintenance changes scoped to the repository contract the user selected

Structure Check

Validate at least:

source and SOTA directories exist
wiki index/log exist
data policy is documented
output directories are separated by trust level
experiments have registry and record templates
the Project Quality Contract is present and matches the repository layout
badge evidence has an append-only ledger when the project targets artifact review
MCP and skill recommendations are documented
agent onboarding and capability profile are documented when the repo supports them
generated or large files are ignored appropriately

Project Quality Gate

Maintenance is responsible for keeping the project easy to continue after many agent turns. Before finishing, confirm that new files live in the right work zone, trusted outputs have provenance and validation, scratch or debug material is not mixed with final outputs, and badge-related evidence is linked from artifacts/badge-evidence-ledger.csv and artifacts/artifact-checklist.md.

Do Not

Flatten the repo into notebooks.
Hide important workflow rules inside chat history.
Put secrets, raw sensitive datasets, or model weights in git.
Move user data without preserving provenance.