ctx

Bring an EXISTING project (brownfield) under ctx with minimal disruption. Decides where the /ctx folder lives (in-repo directory by default, or a gitignored symlink to an external sibling store named project-name-ctx when the context must not ship with public code), which pointer file announces it (respect the project's existing AGENTS.md/CLAUDE.md; else the running agent's native one), and ROUTES the actual doc-writing to the domain skills (ctx-merge / ctx-spec / ctx-progress / ctx-report) rather than writing shallow versions itself. Use when a user says "apply ctx to this existing repo / bring this repo under ctx", when deciding whether ctx must stay separable from published code, or when onboarding a messy existing repo.

Sweeps everything decided in the current session into its correct source-of-truth home — progress, decisions, spec — and verifies the knowledge base is current and internally consistent before a context reset, so a /compact or a new session loses nothing. Use before running /compact, before ending a session, or whenever the user says they are about to reset or hand off and wants a safety checkpoint. A thin orchestrator over ctx-progress, ctx-merge, ctx-spec, and the consistency gate — it introduces no new rules, it runs the existing ones as a checklist. Not for routine mid-session saves — that is sink-same-turn in ctx-progress.

Converges many scattered sources — dated research notes, subagent outputs, audit reports, revision cycles — into one living source of truth without silently dropping or distorting anything, routing each conclusion to exactly one home via a visible disposition ledger and surfacing conflicts as choices for a human. Use when merging or consolidating notes/reports into a ctx source of truth, integrating subagent research, synthesizing multiple audit reports, or closing a decision cycle where alternatives existed. Not for writing a single fresh doc from scratch — use ctx-spec.

Tracks the WORK truth of a project — where we are, what's next, and what to read first — in a living progress file a fresh session can resume from, including cross-session handoff. Use when updating work or project state in a ctx knowledge base, recording what's done or what's next, opening a progress subtree as it grows past a flat file, handing off to a new session, or auditing a project for progress drift.

Sets up and orchestrates a ctx knowledge base — a lean, living single source of truth for multi-session AI-agent work — and classifies any doc or finding by its lifetime, routing the work to the right companion skill. Use when starting or scaffolding ctx on a new project, when unsure where a finding belongs (spec vs decision vs progress vs scratch), when unsure which ctx skill handles a task, or when docs have sprawled into redundant, contradictory piles that need converging. Entry point for the collection; routes to ctx-adopt (existing repos), ctx-merge (converge sources), ctx-spec (specs and ADRs), ctx-progress (work tracking), ctx-report (reports for a human), ctx-serve (host processes across sessions), ctx-compact (pre-reset checkpoint). For an existing/brownfield repo, use ctx-adopt instead.

Writes specs, ADRs, architecture and design docs that an AI coding agent can build directly from — the right granularity (constraints, not hand-written implementation), testable EARS acceptance criteria, and ADR discipline (record only decided choices, supersede in place). Use when writing or restructuring a spec, a decision record (ADR), an architecture doc, or a design doc in a ctx knowledge base, or when deciding how detailed a spec needs to be.

Produces a disposable HTML report laid out for a human to read, comment on, and decide from — then distills the keep-worthy conclusions into the source of truth and archives the report. Use when writing a report, a review doc, or an options-for-a-decision document for a human to weigh in on in a ctx knowledge base, when a human has commented on a report and it needs distilling, or when un-merged reports are piling up. Not for converging a pile of raw notes or subagent outputs straight into the source of truth — use ctx-merge for that.

Hosts a project's long-running processes — dev servers, watchers, build daemons — in tmux so they survive session exit and are shared across parallel agent sessions on one repo, recording their durable topology in a committed service manifest and detecting live status rather than storing it. Use when a task needs a dev server or watcher running, when parallel sessions must share one running service, when a process must outlive a /compact or session reset, or when reattaching to or restarting a service a prior session started. Records topology in ctx/services.md; a narrow operational companion in a ctx knowledge base. Not for one-shot foreground commands.