The vocabulary that runs through the rest of the docs. Each entry is a short
definition; follow the link for the page that covers the term in depth. Every
term has its own anchor, so you can deep-link a single definition (for example
glossary/#output-reference).
The magus root directory that owns a set of projects and shared config; the unit magus operates over. See workspace.
A directory magus recognizes as a unit of work (it has a magusfile); the unit of caching, scheduling, and dependency tracking. See workspace.
The magusfile.buzz that declares a project's targets (as export funs) and
binds its spells. See targets.
A named operation (build, test, ...) you invoke with magus run <target>; it
may compose a spell's tool-native operations and depend on other targets. See
targets.
A single tool-native command a target composes (long form: operation); the middle of the work hierarchy (Spell to Op to Target). See operations.
A language/runtime adapter (e.g. go, md) that maps generic targets onto a
toolchain's real commands. See spells.
An execution modifier attached with : (lint:rw) that changes how a target
runs, not which one; the built-in rw flips a check-only target to mutate in
place, and ci always strips it. See charms.
A coded diagnostic that inspects a resolved op and nudges or blocks an anti-pattern before it runs. See wards.
A magus stdlib namespace a magusfile imports for host capabilities: filesystem, exec, vcs, and more. See the module reference.
The language magusfiles are written in (the .buzz engine). See
engines.
The interpreter a magusfile runs on; magus embeds the Buzz engine. See engines.
The content-addressed store magus consults before running a target, so unchanged work is skipped. See cache.
The set of projects touched by a change; magus affected <target> runs a target
only over them. See affected.
The restricted filesystem and environment a target runs in, so builds stay reproducible and side-effect-free. See sandbox.
A long-running or shared process magus manages across runs, distinct from a one-shot target. See services.
The background magus host that owns shared state such as services and the warm knowledge graph. See daemon.
An ordinary magusfile-defined target you compose yourself with magus\needs -
magus does not hardcode its stages. Magus.RunCI treats it specially only in
that it strips the rw charm, it is the anchor magus affected ci keys off,
and a selected scope with no project declaring it is a load error rather than
a silent no-op. See targets.
A short, shareable id (out1a2b3c, "ref" for short) for one target execution's
captured output; it appears on each target's line, and
magus query output out1a2b3c prints those exact bytes. In OpenTelemetry terms
it corresponds to a span (one target execution) within its trace (the
whole magus invocation). See output-refs.
OpenTelemetry's name for one whole magus invocation; every target it runs is a
span beneath it. See telemetry.
OpenTelemetry's name for one unit of work under a trace - a target execution, whose sub-operations are child spans. An output reference points at a span's captured output. See telemetry.
The concurrency pool: the shared set of slots that caps how many targets run in
parallel on one machine. Its capacity defaults to MAGUS_CONCURRENCY, then 4 on
GitHub-hosted runners, then min(NumCPU, 8); magus status and the
dashboard report it live. See daemon.
One unit of the pool's capacity. A target acquires the slots it needs to run (most take one) and releases them when it finishes; the pool tracks capacity (total slots), running (acquired), and queued (blocked). See daemon.
How many targets run at once. It is bounded by the pool's capacity and set with
--concurrency, MAGUS_CONCURRENCY, or the concurrency config key. See
daemon.
A target that wants a slot while the pool is full; it blocks first-in-first-out until a slot frees. The dashboard colors a sample with queued > 0 accordingly. See daemon.
Which pool a run uses: daemon (one shared pool the background daemon owns across every workspace and client) or proc (a per-process pool for a single one-off invocation). See daemon.
A single magus invocation that runs a target and exits, using a per-process
pool; the opposite of the long-lived daemon or a service. See
daemon.
A CI-only backend that shares content-addressed artifacts across runners: a cold machine replays a build another runner already did instead of rebuilding. Every remote artifact must be signed by a trusted key. See remote.
A point-in-time view of live state - the pool's occupancy or a tick of exported metrics - as opposed to accumulated history. See daemon.
The recent history the daemon replays to a dashboard on connect, so its charts start populated instead of empty. It is served from a bounded ring buffer of the last few hundred samples. See daemon.
How long an operation takes. magus records latency as OpenTelemetry histograms per family - target execution, cache op, pool wait, and graph query - and reports each as a count, sum, and percentiles. See telemetry.
A latency value at a given rank, interpolated from a histogram's buckets: p50 is the median, p95 and p99 are the tail that most latency budgets care about. See telemetry.
The at-a-glance daemon state derived from the pool: healthy when the pool is reporting, degraded when it reports an error, down when there is no pool. The dashboard color-codes each state. See daemon.
A target that fails once and passes on rerun is volatile, as opposed to a regression that started failing and stays failing. magus keeps per-target pass/fail history and a Wilson-score volatility rate to tell them apart and auto-retry the noise. See volatility.
The queryable graph of a workspace's spells, targets, docs, and code
relationships; query it with magus query/explain/path. See
knowledge.
The committed routing index at a workspace root, regenerated from the knowledge graph: it lists every node and points at the exact query for a given question, so it is the entry point an agent reads first. See knowledge.
The reports magus derives over the graph and history (hotspots, affinity, ownership, trend, volatility, unreferenced). See insight.
An insight lens: edit frequency times complexity, the prime refactoring targets.
The project view heat-colors the dependency graph by churn; --files ranks
individual files. See insight.
An insight lens: projects that change together (temporal coupling). A pair that co-changes without either declaring a dependency on the other is a candidate architectural smell. See insight.
An insight lens: author concentration - the primary author and their share, the distinct-author count (the bus factor), and abandonment. See insight.
An insight lens: the recent half of the window against the earlier half. A positive delta is a rising hotspot; a negative one is cooling. See insight.
A stable MGSxxxx identifier attached to a magus warning or error, so it can be
referenced and looked up; some are guardrails (see wards), others hard
errors.
Two rules named often enough elsewhere to need a definition of their own.
The question every proposed capability has to answer: does it read the model magus already had to build, or does it make magus learn something new about the world? Reads stay small; acquisitions are where a tool loses its shape. See scope.
Each concept gets one name, used everywhere: target, spell, charm, op. A second word for the same thing is a house dialect, and it costs every reader (and every agent) a lookup that never ends. See doctrine.
The vocabulary of magus watching work happen: who ran what, what an agent is blocked on, and which agent owns which paths. The policy behind these terms lives in doctrine.
One magus process's recorded facts - the targets it finished, their outcomes,
and the lease it acted as - kept in a repo-scoped store every worktree shares.
magus session lists them; the store prunes itself by last-fact age.
A durable "an agent is blocked" record, opened when a magus session notify
event carries the waiting or permission outcome and held until a person disposes
it. magus session attention lists what is open. Nothing closes one on its own - see
doctrine.
The human act of closing an attention request: a judgment rendered, recorded with who and why. Distinct from resolving a review thread or a merge conflict - a disposition answers a request; it does not merge anything.
One row of the lease ledger: a piece of work an orchestrating agent handed out, with its goal, the checkpoint it was cut against, and the paths it owns or must not touch. The ledger records; the agent guard is what reads those facts back when grading a write. See doctrine.
The short identifier a worker carries (the --lease flag, or the
magus.lease member of the W3C BAGGAGE environment channel) so its runs,
journal facts, and guard verdicts attribute to its lease. Letters, digits
and -_./: only.
What a spawning tool said about itself in the environment: TRACEPARENT (the
W3C trace and the parent span this process runs under) and the
magus.spawner baggage member (a label for whoever spawned it). magus records
each verbatim beside the session's own minted span id, and no verdict reads
any of them - the ancestry is a relation between recorded sessions, the way a
process tree is a relation between pids.
One read-only check from the advice suite: it reads the changeset through magus
and writes one titled section of findings. The same advisors run as a pull
request comment in CI and inside magus diff --impact locally.
The vocabulary of the browser app. These terms name things you only meet in the console's UI, so they are defined here rather than left to be inferred from it.
The browser app that reads a magus workspace: a tabbed, tiling page hosting the log viewer, graph explorer, dashboard, and activity trail. It is a separate static app, not something the daemon serves - the daemon exposes a loopback API it calls: read-only views plus one bearer-gated job-control service for maintenance jobs. See reference/console.
One of the console's applications (Runs, Log Viewer, Graph Explorer, Dashboard, Activity Trail, Settings). "App" rather than "page" because one is never a document you navigate to: it is mounted into a tab, or into a pane beside another one. Each is single-instance - opening one you already have focuses it instead of duplicating it.
The glossary term is two words on purpose. As a bare "App" the auto-linker matched every unrelated "app" in the corpus - a ChatGPT desktop app, a Postgres app - and pointed each at this definition. See reference/console.
A split within a tab. Splitting divides the focused pane along its longer side, so the same action tiles side-by-side on a desktop and stacks on a phone; a tab with no split is a single pane. Drag the divider to re-weight the split. See reference/console.
A key combination bound to a console command, written mod+k - where mod is
Cmd on macOS and Ctrl elsewhere, so one binding fits both. Every chord is
rebindable (Settings > Keybindings), and a command remains reachable from the
command bar whether or not it has one. See reference/console.
The console's runner: one searchable list of every command and its chord,
opened with mod+k. It is the discoverable route to any action - the menus and
chords dispatch the same commands it does. See reference/console.
The URL that points an app at a running daemon. The daemon serves the console
from its own loopback origin, so the link is that origin plus the app path and
a bearer token in the fragment (http://127.0.0.1:7391/console/graph/#token=...).
The daemon prints it; the console consumes the token, stores it, and strips it from
the URL, so the secret never lingers in history or a copied link. The origin must be
literal loopback - localhost and hostnames are rejected before any request.
Without one, an app
reads only what rides in the link itself. See reference/console.
This page uses none of the site's convention markers. The full set is on the conventions page.