A context pipeline built into the repo.
Work blurs — chats scatter, plans go stale, memory fades. Project Context keeps one sharp frame on what is true: small Markdown records that outlive any one person, agent, or chat.
One command, or one prompt
One command, nothing installed
The install command runs the CLI straight from the repository — no clone, no virtualenv. Swap --apply for --dry-run and it prints the exact file plan instead of writing it.
Or a command that stays
pipx install git+https://github.com/monomind-ai-lab/project-context keeps init, inspect, review, and doctor on your path for every repository after this one.
Or no tooling at all
Paste one prompt into any agent that can read and edit the folder. It asks the onboarding question, proposes the plan, and waits for your approval before it writes.
How the pipeline runs
Work produces evidence
Primary artifacts stay primary. Nothing is copied into a second knowledge base.
Milestones promote context
Current state to NOW.md, constraints to DECISIONS.md, verified lessons to LEARNINGS.md.
The next agent reads first
NOW.md, decisions, learnings, then only the relevant evidence links.
What the doctor actually checks
The documents hold up
Required files, installed scaffold version, freshness of NOW.md, duplicate decision and learning IDs, broken relative links. Read-only: it reports, it never rewrites your knowledge.
The evidence still says what it said
Evidence may pin the state it cites as path/to/file@<commit>. evidence-drift warns when that path changed since — the justification may no longer hold. evidence-unverifiable warns when the commit is unknown.
Something still delivers it
The managed instruction block, a harness pointer, a working hook — at least one route has to carry the protocol into a session. no-delivery-path is an error, because perfect files that nothing ever loads otherwise report healthy.
Optional add-ons, informed opt-in
Project Context records what the code cannot say — the decisions, the learnings, the current handoff. A generated wiki regenerates what the code does say. They compose: Project Context is the authority layer, a generated wiki an optional derived view.
Code symbols, relationships, impact, and execution flows.
Relationships across code, documents, research artifacts, and media.
Ongoing generated documentation and navigation for stable projects.
Independent open-source projects, recommended when they fit. Project Context works without any of them.
Setup notes live in the optional-tools reference.