Docs
A collaborator returning after three weeks should not rebuild the project from stale chats. They should open one small file and be able to answer:
Current Project State
Snapshot
| Area | Current state | Evidence |
|---|---|---|
| Evidence scope | The report uses public studies published through 2025-12-31. | D-001 |
Active work
| Initiative | Status | Next action | Evidence |
|---|---|---|---|
| Findings synthesis | in-progress | Resolve three claims that lack page-level citations. | L-001 |
Blockers
None.
what the next session reads first
This is not a mock-up. It is the complete NOW.md from the shipped sample project, unedited — a research report with an evidence cutoff. D-001 and L-001 point into the decision and learning registries beside it.
Not only software, and not only GitHub — but also not everything. The last row is the honest one: a project nobody will return to has nothing to hand off, and installing a handoff record into it is busywork.
| Your project | Fits? | Start here |
|---|---|---|
| Code repository, several agents touch it | Yes — the core case | Install |
| Brand-new repository, nothing written yet | Yes | Install |
| Research corpus — papers, datasets, findings | Yes | Use cases |
| Writing or documents project | Yes | Use cases |
| A shared folder with no Git | Yes — Git adds history, it is not required | Install |
| A project you will not return to | No — nothing to hand off |
Preview first. --dry-run prints the exact file plan and writes nothing; run --apply only after you have read it.
uvx --from git+https://github.com/monomind-ai-lab/project-context project-context init --target . --install-skills --dry-run
uvx --from git+https://github.com/monomind-ai-lab/project-context project-context init --target . --install-skills --apply
pipx install git+https://github.com/monomind-ai-lab/project-context && project-context init --target . --install-skills --dry-run
Installing is not adopting. The record earns its keep through five moments in every working session — and the agents handle all five.
The agent runs project-context context --task "<one line>" --files <paths> and reads the packet it returns — the owner's constraints, the current state, then the records anchored to the files it named. Reading NOW.md and PLAN.md and searching the registries is the fallback where the CLI is not available. Automatic where the harness discovers the installed skill; the managed block in AGENTS.md / CLAUDE.md covers the rest.
Anything worth keeping that is not yet a registry entry goes into project-context/inbox/ with project-context capture — one short, dated, attributed note. Deciding what it becomes is deferred to promotion, where it is cheap.
Trigger detection flags moments that produce a durable record — an accepted trade-off, a reproduced bug, a verified benchmark. The agent proposes the entry, and updating the record is what closes the window. When nothing fired, an ack records that honest outcome instead.
NOW.md is updated to what is now true. Decisions and learnings are appended only if they were earned — durable constraints and verified lessons, not a diary.
The doctor runs. Read-only, always: it flags stale state, broken links and dead delivery paths, and it never rewrites your judgement.
The full day-2 ritual, trigger examples and the first-week checklist → operate
The six safety promises, both install paths, per-harness setup, verifying, and uninstalling.
recordsWhat each file answers, what does not belong in it, and a good entry next to a bad one.
operateThe session loop in detail, trigger detection with ack, and the first-week checklist.
The layer under all of this: the authority ladder, frontmatter, one lifecycle per kind, the reference grammar, capture and retrieval.
scaleProject Hub is the optional other half — one private repository where an owner authors what applies everywhere and pushes it down.
use casesThe repository many agents work in — and every collaborative project that is not code.
Clarity comes with context.