onetaskgraph
One interface over the ticketing systems your work actually lives in.
Every image in this README is a real capture of this CLI's output, produced by running the binary against a small fixture and refused by CI the moment its bytes stop matching what the binary prints — so the pictures cannot drift away from the tool.
Tasks, projects, labels and the dependencies between them are spread across Linear, GitHub
Projects and a folder of Markdown files, and every tool that wants to reach them ends up
reimplementing all three. onetaskgraph implements them once, behind a single query
surface: a command-line tool, the Rust engine crate's library API, and SDKs for Python and
TypeScript. Every consumer reaches the same engine, so their query semantics cannot drift.
Two properties make it different from a lowest-common-denominator wrapper:
-
A rich source is not reduced to a poor one's floor. Each source declares what it can do natively. The engine pushes those predicates down and compensates in memory for the rest — and every response carries the plan it ran, so
--explainshows you which source filtered server-side and which one the engine narrowed for. -
Nothing of your work is kept outside the system that owns it. No cache, no index, no local mirror. The engine holds at most one source page at a time and writes nothing down — enforced by a supply-chain gate that refuses every embedded store and cache crate, a sandboxed journey that fails if any file written during a run contains your data, and an assertion that the same query asked twice reaches the source twice.
copyis not an exception to that, and the difference is worth stating plainly. A destination write is at your explicit request, names its destination, goes through that source's own write interface into that source's own store, and is never read back to answer a query. A cache is a write nobody asked for that the engine reads back. Copying a task into a folder of Markdown puts it in the plugin that now owns it; nothing is kept anywhere else, and the sandboxed journey above drivescopyand fails, naming the path, if any file outside that destination's own store changes.
Status. The plugin contract, the workspace, the gate, the configuration layer and the query engine are in place, and the binary answers every verb below. The three sources this product ships for your work —
local-md,linearandgithub-projects— can all be read and written; none is read-only. A copy can still refuse a configured destination with no write side, such as a subprocess source whose capability handshake declares no writes, but that is a property of that configured source rather than of these plugins.
Using it
Each source says what it can do, and sources list is where you read that back: the
predicates it applies itself, how far it walks dependencies, and the largest page it will
serve.
# Without --apply this only reports what a GitHub Projects board lacks of the fields its
# source's configuration names: the Status options status_mapping resolves to, and the
# Priority field and its options when priority_mapping is set. --apply adds the missing
# options and creates a missing Priority field, sending every existing option back with its
# id, then verifies every pre-existing option and every item's values of both fields and
# prints recovery data if GitHub drifted; a status name is reported missing for each item kind
# that names it. For a Linear source it reports every name its status_mapping gives a task
# against the team's workflow states, and every name it gives a project against the
# workspace's project statuses, each present with its type or missing; --apply creates each
# missing one first, of the type its category derives, and renames, retypes or deletes nothing.
# See "Status mapping" below.
# The Status-only form of `sources fields`, which supersedes it; kept as it was.
# Where an item with these repositories, written to SOURCE, would land by its routes —
# read from configuration alone, never from a source. See "Routing" below.
|
|||||done||
||||
|
|
|
# flags of `task create`
# and asset flags of `task create`
|
|
An <ID> is qualified: work:ENG-142. Repeating --label narrows — a second one is a
second requirement — and --not-label excludes. --project takes a qualified id, which
narrows the query to that project's own source, or a bare native id, which is asked of
every selected source.
A document is what lives in a project and is not work — a design note, a runbook, a
page somebody has to read. It carries no status and takes part in no dependency graph, so
document has no --status filter and no deps verb. What it does carry is a
location: where a reader can actually open it, as a link (url …) or as an absolute
path on the machine its source runs on (path …). task show, project show and
document show all print it, and --json carries the contract type's own shape —
{"url": …} or {"path": …} — so a program branches on which key is present. Not every
source has documents; one that says it has none is reported as holding none rather than as
having failed, and a copy naming it is refused before anything is read.
A comment is added to a task after the task exists, without rewriting it — evidence
appended to an issue somebody else opened. task comment add and edit read the body from
--body-file, or from standard input when that flag is absent, and never from a word of the
command line, so a body quoting a command never passes through a shell; it is stored and
returned byte for byte, and an empty one is refused. --author is for a source that records
what it is given, such as a folder of Markdown; GitHub and Linear record the signed-in
account themselves and refuse it rather than drop it. task show prints a task's comments
after its body, and its --json carries them as a top-level comments list for a source
whose tasks have comments — absent, rather than empty, for one whose tasks have none. A
task copy, project copy or document copy never reads or writes a comment at either end.
task show-many <ID>... shows several tasks at once, whatever sources they are in, and its
--json is {"details": [...]}: one entry per id, in the order given, each exactly what
task show <ID> --json prints. An id that cannot be read — no such task, no such source, a
GitHub draft's comments — carries its failure in that entry's own errors and does not refuse
the others; the command exits non-zero exactly when some entry carries one. Each source is
asked for its ids together, so GitHub Projects reads 24 items, with their comments, per
request.
task list --commented-since <INSTANT> keeps the tasks one of whose comments was created,
or last edited, at or after the instant — an RFC 3339 time with its offset, such as
2026-09-28T12:00:00Z; one without an offset is refused, naming the flag. A task with no
comments never matches, and a comment deleted before the query is not a match. It combines
with every other filter, and the answer is exactly their intersection. A folder of Markdown,
an in-memory source and a GitHub Projects board apply it themselves. A board asks GitHub's
issue search, scoped by the board alone, for the issues updated since the instant, then reads
those candidates' comments and no others' — exact for new and for edited comments in every
repository the board's items live in, because GitHub moves an issue's updatedAt when one of
its comments is added or edited, which the board's credentialed journey re-takes on every
run. That search is an index that lags a write by a second or two, so when you ask again from
the time you last asked, overlap the two instants by more than that. Linear asks for the
issues with a comment created or last edited at or after the instant, by the same two times a
comment read reports. A source that does not apply it itself is narrowed by the engine, which
reads that source's comments task by task for every task the other filters kept: correct,
and as costly as that sounds on a large workspace.
task list --metadata <KEY>[/<SEGMENT>…]=<VALUE> keeps the tasks whose caller-defined
metadata holds the string VALUE at that location. / splits the top-level key — which may
itself contain dots, such as orchestrator.follow-up — from the nested object keys under it,
and the first = splits the location from the value, so
--metadata orchestrator.follow-up/root_cause=stale-cache keeps the tasks whose
orchestrator.follow-up object holds root_cause: "stale-cache". The value there must be a
string equal to VALUE, case-sensitively. Repeat the flag and a task is kept only when it
holds every one. task list --origin <SOURCE>:<ID> keeps the tasks copied from that item:
those whose onetaskgraph.origin is exactly that qualified id — never one it only begins or
ends with. A folder of Markdown and an in-memory source apply both themselves; Linear asks for
the issues whose description holds each value as its metadata slot spells it and confirms every
one over the parsed slot; a source that applies neither is narrowed by the engine over each
task's own metadata.
A GitHub Projects board answers --search, --metadata and --origin without listing the
board, which is what keeps a large board affordable:
--searchis one board-scoped issue search for the text as a phrase, in the title, the body or both as--insays, and every candidate is confirmed by the same case-insensitive substring rule a folder of Markdown uses — so every task returned contains the text. GitHub matches whole words, so a task holding the text only inside a longer word (shipinsideShipment) is not returned: that is this source's declared behaviour, not a defect.--metadatais the same search with each value as a phrase in the body — GitHub indexes the metadata comment a board keeps at the end of each issue body — and every candidate is confirmed against that parsed comment.--originis the board's own field filter on itsonetaskgraph.originfield, which finds every carrier whichever release wrote it, together with the issue search for the id in the body, where a write mirrors it so a copy another process made seconds ago is found too. Every candidate is confirmed against the origin field, which is the one place a board's origin is read from.
Every answer is completed with what the same command has just written, so a task written a moment ago is found. A task another process wrote moments ago may be missing until GitHub indexes it — usually a second or two, and for a task whose origin is in the field alone, as long as the board's own item list is behind on it, which can be minutes. A board draft is not an issue and is never returned by any of the three. Labels, status, priority and project are confirmed over the candidates, and a query carrying none of the three reads the board as before.
project list --search, and document list --search without --project P — --no-project
included — are the same board-scoped search, each candidate confirmed by its kind and by the
substring rule, and narrow the same way: whole words only, so ship does not find a project
titled Shipment plan, and no board draft — a draft titled as a design document included — is
ever returned. A text with no letter or digit, such as --, is refused there before anything
is sent. document list --project P --search sends no search: it reads that project's
sub-issues and matches the text inside words too, as before.
A task's status is set on its own with task status set, which writes that one field —
title, body, labels, metadata, dependencies and comments stay exactly as they are — and
answers with the status as the source reads it back. queued is the category for work
claimed by something that will do it and not yet started, between todo (ready, and nothing
has claimed it) and in-progress. A folder of Markdown reads the word queued; a GitHub
Projects board sends it to its Queued column by default, status_mapping.queued naming
another; Linear writes it as the name its source's status_mapping.queued gives the item's
kind, and refuses it by name when that source gives none. See "Status mapping" below.
On GitHub Projects, terminal writes keep both GitHub representations aligned: done
selects the mapped Done option and closes the issue as completed, while cancelled
selects the mapped Cancelled option and closes it as not planned. Those are the shipped
option names. An open-category write reopens a closed issue before selecting its mapped
option; a draft item has no issue state to reopen. If any mapped option is absent, the
write is refused by that option's name before either representation changes —
sources fields (and its Status-only form, sources status-options) counts a terminal
category's mapped option as configured, so it names a missing Done or Cancelled and
--apply adds it. Reads keep
GitHub's issue decision authoritative for closed issues—completed reads done and not
planned reads cancelled, regardless of the displayed option—while the Status option
decides an open issue's category.
A task's priority is one of none, urgent, high, medium and low, and every task
carries one — none when none is set, and for every task of a source that cannot hold one.
task list --priority keeps the tasks at any of the priorities named, and task priority set
writes that one field and answers with the priority as the source reads it back; none
clears it. A copy carries a task's priority like its status. A folder of Markdown holds it as
a priority: key, Linear as its own issue priority, and a GitHub Projects board as the
option of a single-select Priority field that the source's priority_mapping names —
Urgent, High, Medium and Low unless it says otherwise. A board source configured
without priority_mapping holds no priority, and a copy or a task priority set that would
write one to it is refused by name before the board is asked anything. sources fields <SOURCE> --apply creates the board's Priority field, or adds the options it lacks, without
disturbing an option or an item's value that is already there.
A task's content is replaced on its own with task content set <ID> --file PATH: the
file's bytes become the task's body, and status, priority, metadata, labels, repositories and
dependencies stay exactly as they are — on a GitHub board, whose metadata lives in the issue
body beside the content, that block is kept as it was. There is no compare-and-set: what the
file holds replaces whatever the task held.
One metadata key of a task, a project or a document is set on its own with task,
project or document metadata set <ID> <KEY> <VALUE>, which adds the key or replaces what
it holds and changes nothing else about the record — every other key, and every other field,
stays exactly as it was, and a key already holding the value is not written at all. KEY is
your own dotted <namespace>.<name>: two or more non-empty segments, never in the
onetaskgraph. namespace this product keeps in step itself. VALUE is exactly one JSON
value, parsed as JSON and never as YAML, so a bare yes or 2026-01-01 is refused rather
than given a type — quote a string as '"text"'. An unqualified id, a key or a value of
those shapes is refused before any source is asked. The answer — MetadataSet under
--json — carries the id, the key, the value as the source reads it back after the write,
and the record's location. Metadata is not status, so no task it delivers is re-evaluated.
A folder of Markdown edits the one entry of its metadata: block and no other byte, and
refuses by name a block it cannot edit that narrowly; a GitHub Projects board sends one
update of the issue body that changes only its metadata slot; the in-memory source holds the
value for the life of its process; and Linear sends one update of the issue's or the
project's description, or the document's content, that changes only the metadata slot at its
end. A source with no
write side, a record the source does not hold, and a stdio plugin whose handshake does not
declare the write are each refused by name.
Several fields of one task are updated together with task update <ID>, which writes
every field it names and nothing else: --title, the content from --body-file, --status
(and --status-name, the status's own word where the source keeps one), --priority, any
number of --metadata KEY=JSON to set and --remove-metadata KEY to remove, and
--delivers or --depends-on lists that replace the task's own, their --no-… forms
replacing them with none. A field that already holds the value named is not written, and an
update in which nothing differs writes nothing at all; the answer — TaskUpdated under
--json — says which fields were written, carries the task as its source reads it back, and
under spent what the source's own meter says the call cost. Labels, repositories and the
project a task is filed under are not among the fields: an existing item is never moved. On a
GitHub Projects board that is at most one read of the issue, one updateIssue carrying the
title, the body — visible content and metadata block together — and any state change, one
field write each for Status and Priority, and only the blockedBy edges that differ;
Linear sends one issueUpdate of what differs, and a folder of Markdown replaces the file
once, every byte it was not asked to change kept. Naming a status or a delivers list keeps
every delivered task in step exactly as task status set does, and exits 4 when one could
not be; naming neither re-evaluates nothing. A key both set and removed is refused before
anything is written, and so is an update naming no field.
A task can name the tasks it delivers: finishing it finishes them. In a folder of
Markdown that is a delivers: list in the front matter — a bare id names a task of the same
folder and <source>:<id> a task anywhere — and on a GitHub board it is the
onetaskgraph.delivers key of the issue's metadata block. The delivered task's
delivered_by is onetaskgraph's to keep. Whenever it writes a task that delivers anything,
or delivered something before — a copy, task status set or task update — each delivered
task gains the deliverer's qualified id, a task it dropped loses it, and the delivered task's
status follows its deliverers while it is at todo, queued or in-progress:
in-progress while any runs, queued while any is queued, done once every one is done or
cancelled and at least one is done, and todo when they release it. A delivered task at draft, backlog, unknown,
done or cancelled is left alone, a deliverer its source no longer holds is dropped from
delivered_by, and a delivered task that cannot be read or written is reported failed while
the deliverer's own write stands.
A copy rewrites a delivers entry naming another item it copies to that item's new id,
counts those under delivers_rewritten, and never carries delivered_by. Every verb that
writes a deliverer reports each delivered task it evaluated under delivered. A --dry-run
copy writes nothing and so keeps no delivered task in step: its delivered list is empty,
which says nothing about whether a delivered task would have moved.
How a source spells a document is its own business. A GitHub Projects board has no
document type, so github-projects reads one as an ordinary issue whose title begins
DESIGN: — the title you see has that prefix taken off, and writing a document puts it
back, so a design note copied out of a board and back returns the title it started with.
docs/metadata.md records the whole of that rule.
Seeing which plan you got
--explain renders the plan the query ran, per source. Here one --label query reaches
two sources of differing capability — a folder of Markdown, and a source that declares it
cannot filter by label at all:
The folder filtered the rows itself; the other source returned the wider set and the engine
narrowed it. One query, two plans, and the same correct answer either way — which is what
the capability declaration above buys you and why it is worth reading. --json carries the
same plan as a field, so a script does not have to parse the prose.
Writing tasks: Markdown in, ticket out
Authoring against a ticketing API is not something a person or an agent does well, and a folder of Markdown files is. So that is how work is created here: write the files, read them back through the CLI to be sure they parse, then copy them where your team works.
With notes configured as a local-md source rooted at ./notes, create its task folder
and write notes/tasks/rate-limit.md:
title: Rate-limit the sync loop
status: Todo
labels: [{id: local-reliability, name: reliability}]
metadata: {onepipeline.turn_budget: 12}
repositories: [github.com/acme/sync]
Back off when the API asks us to slow down.
Read it through the CLI before writing to the permanent linear destination, then copy
the qualified id the read returned:
$ onetaskgraph task list --source notes
notes:rate-limit todo Rate-limit the sync loop
$ onetaskgraph task copy notes:rate-limit --to linear
The read is intentional: a malformed front matter block caught in a local file is cheaper
to fix than a parse failure first discovered while writing to somebody's ticketing system.
Projects use the same flow under notes/projects/, followed by project list and
project copy.
Editing is the same road in reverse — copy out, edit, copy back:
$ onetaskgraph task copy work:T-1 --to notes
$ grep -A2 '^metadata:' notes/tasks/T-1.md
metadata:
onetaskgraph.origin: work:T-1
$ $EDITOR notes/tasks/T-1.md
$ onetaskgraph task copy notes:T-1 --to work
The copy back updates rather than duplicating because the copied file carries the id
it came from, under the reserved metadata key onetaskgraph.origin. The item it was copied
from records where it landed too, under onetaskgraph.copies — an object naming the
counterpart at each destination it was copied into, such as
{"notes": "notes:T-1"}. Nothing anywhere holds a mapping: the correspondence lives on the
two items, inside the plugins that own them.
These rules find the counterpart, in this order:
- The link. If the item's
onetaskgraph.copiesnames an item at the destination, that item is read by id, and when its own origin names the item being copied it is the counterpart and the copy updates it — one read, and no search of the destination. If it names an item the destination no longer holds, the copy refuses with the failure kindstale-link, naming both ids, because creating there would duplicate work somebody deleted;--recreategoes on to the rules below instead, and then creates. If it names an item whose origin somebody re-pointed at something else, the link is ignored. - The origin. If the item's origin names the destination, that origin is the destination item and the copy updates it.
- The search. Otherwise the destination is asked for the task whose origin is the id
being copied — the same question
task list --originasks, which a source that can answers without listing everything it holds; found, it is updated, and not found, one is created carrying that origin.
Once the whole copy has landed, every item found by the link, the search or --match-by, or
created, has its link for that destination recorded or refreshed — one metadata write at its
own source, undone with everything else if the copy cannot finish. An item found by its own
origin records nothing: that correspondence is already written down, on the destination
item. A source that cannot hold the link — it has no write side, or will not make the narrow
metadata write — is copied from exactly as before, and the report says the link was not
recorded. So after one copy the pair is found from either side by one read by id: forward
by the link, back by the origin.
Which rule found it decides what the copy records there. A copy that got its counterpart from the first rule is a copy back: the destination is the original, and the item being copied is the one that came out of it — so the destination keeps the origin it already holds, holding none included. Stamping it with the id of its own copy would destroy the original's provenance, and the next ordinary copy from the store it was authored in would match nothing and create a second item beside it. Every other copy records the id it was copied from, which is what makes the next copy of it an update.
| Flag | What it is for |
|---|---|
--dry-run |
Every read, no write, and the action each item would have got. |
--recreate |
An origin or a link naming an item the destination no longer holds refuses by default, because creating there would duplicate work somebody deleted. This says create instead. |
--create |
task copy only: you assert the destination holds no counterpart of any task named — typically because you have just asked it — so each is created without the correspondence lookup, and a GitHub board is sent no origin query. Sound because that query reads an index that lags a fresh write by the same amount whoever asks, so repeating yours could find nothing yours did not. Refused beside --match-by or --recreate, and for a task whose own link or origin already names an item at the destination, naming that item. |
--match-by KEY |
Delete or corrupt the origin key and neither rule can find the counterpart, so the next copy back creates a new item. This re-establishes the lost correspondence by matching on title, or on a metadata key of your choosing, without hand-editing ids. |
--no-tasks |
Copy a project on its own. By default project copy copies the project and every task in it, matching each task independently. |
--member TASK-ID |
Copy the project and exactly the tasks named, repeating the flag for each, when you know which of them changed. A task not named is not read at the destination, not written and not reported, so a one-task change costs what one task costs rather than a read of the whole project. A named task the destination does not hold yet is still created. An edge to a task not named is written to the destination id that task records at onetaskgraph.origin, and a copy whose edge names a task recording none is refused before anything is written, naming that task. A copy naming members was not told about the rest, so it reports nothing orphaned. |
--dry-run is how you read that table's first row before trusting it: every source is
read, nothing is written, and each item is reported with the action it would have got.
Every field a copy read is written — title, content, status, labels, project,
repositories, metadata and the edges — except url, location, key, created_at and
updated_at, which are the destination's own: the short handle a backend shows people is
issued by that backend, so the one the source wore is not the one the destination does. Nothing is silently dropped: a field the destination cannot
represent, or a metadata key it cannot carry, refuses the write and names it. A copy never
deletes work either, so a destination item the source no longer holds is left exactly as it
is and reported as orphaned.
A copy either completes or leaves the destination as it found it. A copy that cannot finish — a field the destination refuses, a credential that expires, a rate limiter — undoes what it has already written and takes back the items it created in that run, so the retry starts from the destination you started from. That is what stops a half-written project having to be re-run, and the re-run is the burst of writes that trips a hosted destination's rate limiter. When the destination will not take one of them back, the refusal says so and names what is still there rather than leaving you to find it.
--json gives one entry per item for a script to read:
action says which of the four things above happened to that item, and destination is
null only for a dry run that would have created something. The vocabulary itself is the
CopyAction root of onetaskgraph schema, which is what both SDKs are generated from and
what the journeys validate this output against.
via is one field with five words, on every created, updated and unchanged item. On
updated and unchanged it names the rule above that found the counterpart: link,
origin, scan, or match for --match-by. On created it is always created, because
none did. link says what became of the item's own link, on those same three actions of a
copy that writes: recorded, unchanged, or unrecorded where its source could not hold
one. A dry run reports via and leaves link out, and an orphaned item carries neither.
In the generated SDKs the four rule words are the type CopyVia and created is
NoCounterpart, and the link words are CopyLink.
A copy that reaches a source which counts its own requests — github-projects does — also
says what it spent: the requests those sources sent for the command, and what that cost
each budget they draw on, with a flag on any amount that is partly a modelled lower bound
rather than a figure the backend reported. It is the source's own account, summed over the
command, and a copy whose sources count nothing carries no spent at all rather than a
zero. Its shape is the spent property of the CopyReport root of onetaskgraph schema.
Task templates
A task template is a minijinja document whose variables are declared in YAML front matter, so the shape of a task is a file you own rather than prose several tools restate:
---
onetaskgraph_template: 1 # required when front matter is present
description: <string> # optional
variables: # optional; name -> declaration
<name>: # ^[a-z][a-z0-9_]*$
description: <string> # required, non-empty: what a prompt shows
type: string # string | text | integer | boolean | list | object; default string
items: string # list only: string | object; default string
required: true # default: true without `default`, false with one
default: <value of `type`> # optional; `required: true` beside a default is refused
---
# {{ title }}
{% for criterion in criteria %}
- {{ criterion }}
{% endfor %}
A file without front matter declares nothing; front matter with an unknown key is refused
by that key. string is one line and text any number. The body is minijinja, whole —
filters, loops, conditionals, macros, a variable used many times or not at all.
{% extends %}, {% include %} and {% import %} find the files they name in the
--search-path directories, in order, and never in the working directory unless you name
it. The declared set is every chain file's front matter together: a redeclaration may change
description, default and required — the file nearer the rendered one wins — but never
type or items, which is refused naming both files. A template named by an expression —
{% include kind ~ ".md" %} — is part of the chain too, for the answers that name it, and
joins it at the tag that names it, exactly as a file a literal names would: its front matter
joins the declared set at its own distance from the rendered file, and its bytes join the
digest where rendering first reads it. Rendering is strict — a name that is
neither declared nor set fails, naming it and its file — with no auto-escaping, trailing
newlines kept, and trim_blocks and lstrip_blocks on. An optional variable with no
answer and no default is none.
template variables lists the declared set and the template's digest: sha256: over
every chain file's name and bytes, which moves whenever any file of the chain does. Each
variable template render needs takes the first of:
--var NAME=VALUE— literal text for astringortextvariable, YAML for any other;- the answers file,
--answers FILE(-for standard input): a YAML mapping of name to value; - a prompt, when the
interactivesetting is on — one variable at a time, showing its description, type and default, asking again when a value is not of its type; - its declared
default.
An answer to no declared variable, an answer of the wrong type, and — when not interactive
— every required variable left unanswered, listed in one refusal, each exit 2. So does an
interactive run with something left to ask whose standard input is not a terminal: it never
waits for an answer nobody can type. Automation passes --no-interactive, which both SDKs
always do.
Creating and regenerating from a template
task create, project create and document create make an item in one source, its body
(a project's description) rendered from a template — --template FILE, or --template-loader FILE for a template a caller states (below) — with its answers taken exactly as template render takes them, or given as it is with --body-file PATH or on standard input. task create prints the new item's qualified
id, and under --json the item exactly as task show --json prints it; document create
answers as document show does, and with --id DOC naming a document the source holds it
replaces that document rather than adding a second. project create names the project's own
id with --id and answers as project show does; over a project the source holds it
replaces the content, the answers and the provenance whole and keeps everything else — its
status, labels and repositories unless --status, --label or --repository names them,
every metadata key --metadata does not set, and its dependencies. A key of the reserved onetaskgraph.
namespace given with --metadata, and a source that cannot be written, are refused by
name before anything is written.
An item rendered from a template records where it came from under the reserved metadata key
onetaskgraph.template: the template's reference — a template file's absolute path, or a
loader document's reference verbatim — the chain digest it rendered with, and the
SHA-256 of its content (body_digest) and of its resolved answers as canonical JSON
(answers_digest). An item made from a plain body records none. From that entry alone a
hand edit (the content's hash is not body_digest) and a changed template (the chain's
digest is not digest) are both visible. It proves nothing about who wrote it: a check
reading only these hashes trusts them, so provenance forged by hand passes it. The entry is
four strings, whatever the template's reference — nothing caps its length but what the
destination caps a whole item at.
A copy carries the entry like any other metadata, with one exception. A document copy
points the references in a document's content at the destination's own records, and when it rewrote at least one of them in a rendering that
still hashes to its recorded body_digest, the copy records as body_digest the digest of
the rendering as the copy rewrote its references — so the content that lands matches
its entry — and carries template, digest and answers_digest as they were. The answers
kept where the item was authored, and the answers_digest naming them, still hold the
pre-copy locations, so the copy is not a fresh rendering of those answers, and a regenerate
from them reproduces the pre-copy content. A rendering edited by hand, an entry this product
did not write, no entry at all, and a copy that rewrote nothing are carried verbatim, so a
hand edit stays visible wherever it is copied.
The answers themselves are kept in one place only: beside the item in a local-md
folder's own file, which is where an item is authored — never in its content or its
metadata, so they cost a hosted item nothing, and nothing of a task is written twice into a
GitHub issue. task answers, project answers and document answers print them as YAML (--json: one JSON
object), and refuse, naming the item, when none are stored — which is every item of a
source that keeps none. A copy carries content and metadata alone, never the answers,
at either end.
task render, project render and document render regenerate an item in place, and write its content, its
provenance and its stored answers in one write and nothing else — its id, title, status,
labels, project, repositories, dependencies and every other metadata key stay as they were.
The answers start from the stored ones when they hash to the recorded answers_digest;
otherwise every required variable has to be answered again, and a render that leaves one
unanswered is refused as supply every required answer, naming each and why the stored
answers were not used (exit 2). --var, --answers and --unset NAME — which drops an
answer so its default applies — are laid over that base. The template is the one given;
else the recorded reference, re-read when it is a readable file — its extends, include
and import resolved over --search-path again, because the provenance records none;
else the render is refused naming that reference, because a reference is never turned into
a location. --dry-run renders and writes nothing; --json prints {id, digest, body_digest, changed, body}, and a render that would change nothing reports changed: false and writes nothing.
A caller that layers templates of its own states the result with a template loader
document — --template-loader FILE, - for standard input, but never together with
--answers - — and this product renders exactly that, never resolving a caller's layers
nor running a caller's command:
reference (required, non-empty) is what the item records; entry (required) is loaded over
the search_path directories, then the inline templates; digest, when present, must be
the digest that chain computes, or the render is refused naming both. Every other key is
ignored. template variables and template render take one in place of their <FILE>.
docs/local-md.md describes the answers block.
Image assets
A task or a document carries pictures — a before-and-after screenshot in a design document,
say — as image assets stored with the record. Its content references each one as a
Markdown image whose target is ./<name>:
<name> is a bare file name — no /, no \, no .. — ending, in any case, in .png,
.jpg, .jpeg, .gif or .webp. That is the only shape that is an asset reference: an
image at an absolute URL, one whose target has a directory (./img/before.png) or starts
../, one with no ./ (before.png), and a plain link to ./before.png are content like
any other, stored, shown and copied exactly as written. So is an image outside the
convention's reach because it is not an image at all: one whose ! is escaped
(\), and image syntax inside an inline code span, a fenced code block
or an indented code block, which is how a document shows the convention without using it. A
record whose only image syntax is of those kinds holds no asset.
task create, task render, document create and document render store an asset from
--asset PATH, repeatable, under the file's base name:
Each is refused by name before anything is written: a file whose extension is not one of the
five, two --asset with one base name, a reference in the content naming an asset the record
would not hold, and an --asset the content does not reference. A render keeps every asset
the record holds that the regenerated content still references, an --asset replaces a stored
one of the same name, and an asset the new content no longer references is dropped; document create --id over a document the source holds leaves it holding exactly the assets that call
gave. A source whose plugin stores no assets refuses the record, naming the source and the
asset.
task show --json and document show --json list them as assets, in the order the content
first references them — [] for a record that holds none:
"assets":
path is where the bytes are on this machine — a local-md folder keeps a record's assets in
a directory beside its file (docs/local-md.md) — and null for a
hosted source. content_type is image/png, image/jpeg, image/gif or image/webp,
decided by the extension.
A copy carries every asset a record's content references. task copy, document copy and
every project copy that carries tasks or documents hand each asset's bytes to the destination
with the record:
- into a
local-mdfolder, the bytes are stored as the copied record's own assets and the content is unchanged; - into a source whose plugin stores them at a URL, the content's
./<name>references are rewritten to where that destination serves each image, and the destination records what it uploaded under the reserved metadata keyonetaskgraph.assets— each asset'ssha256andurl— so a later copy whose bytes have not changed reuses the URL and uploads nothing again; - into a source whose plugin stores no assets — any plugin that declares nothing — the record is refused, naming the source, the record and the asset, before anything is written for it, because a design document whose pictures silently vanished on the way would be approved without them.
A record whose content references an asset it does not hold is refused, naming both. Rewriting
references follows the rule above for a copy that rewrites references: a rendering that still
hashes to its recorded body_digest is re-recorded with the digest of the content as
rewritten, and template, digest and answers_digest are carried as they were. A record
whose content references no asset reads, renders, copies and digests exactly as it does without
this feature, on every source, and a plugin that has never heard of assets is never sent
anything about them. docs/plugin-protocol.md §4.9a states what a
plugin is sent.
Exit codes
| Code | Meaning |
|---|---|
0 |
Success — every source asked, every source answered. |
1 |
The command failed while running: an id that names nothing, a configuration it will not run on, a source name nothing configures. |
2 |
The invocation itself was wrong — an unknown flag, a value out of range, or answers a template refuses. |
4 |
The query ran and at least one source did not answer. The others' results still stand and the failure is named on standard error. A write — a copy, task status set or task update — also exits 4 when it landed and a task it delivers could not be kept in step; its delivered list says which. |
--allow-partial says a partial answer is acceptable and turns 4 into 0. Nothing else
does: a run that lost a source never exits 0 unless you asked for that.
With machine output selected — --json, --output json, or output: json set in a
document, by --set or by ONETASKGRAPH_OUTPUT — a command that exits 1 also writes
exactly one document to standard output, and nothing else goes there. Its shape is
published rather than restated here: it is the FailureDocument root of onetaskgraph schema, which describes every member and is what both SDKs are generated from and what
the journeys validate this output against.
Its class is the member to branch on: refused means asking again unchanged gets the
same answer, and transient — a rate limit, or a source that could not be reached — means
it may not. The standard error line and the exit code are the same with
or without it, and exit 2 writes no document. A partial answer at exit 4 carries the
same class on each entry of its errors.
Paging
--limit N gives you a page and, when there is more, the token for the next one:
# ... rows ...
# next page: --page 5b7b22736f7572...
Rows are interleaved across the selected sources — one from each in configured-name order,
then the next from each — and within a source they keep that source's own order. The turns
carry on across page boundaries, so a walk returns the same rows in the same order whatever
page size you choose: --limit 3 and --limit 50 differ in how many round trips they cost
and in nothing else. The token is the engine's own; a source's cursor travels inside it
untouched, and a walk returns every row exactly once.
A token belongs to the query that produced it. Resuming it from a query that asks for
something else — another --label, another --search, another --source, the other
--direction — is refused rather than answered, because every cursor inside it is a
position in the result set the original query returned, and picking up there under new
filters returns real rows from a walk you are not doing. Change the query, drop --page.
Install
The command-line tool ships as a self-contained binary. Once a release is cut, install it whichever way suits your machine:
For Rust, the SDK surface is the engine crate itself rather than a separate wrapper package. Add it when the application should link the engine, or add either subprocess SDK when it should drive the installed binary:
The engine links its local sources — in-memory, local-md and subprocess — in every
build. The two that reach a network are features no default enables, so an application that
wants only a local store does not compile an HTTP and TLS stack it never runs: add
--features github-projects,linear (or either one) to the first line for GitHub Projects or
Linear. A configuration naming one a build left out is refused with the feature that
enables it.
This complete example constructs an engine over two in-memory sources, copies a task through
Engine::copy, inspects the outcome, and reads the destination back through the engine:
use ;
use SourceName;
use json;
async
Unlike the Python and TypeScript SDKs, which spawn the compiled binary, a Rust consumer
links onetaskgraph-core and calls Engine in process. The engine and its copy semantics
remain the single implementation in either case.
A caller that holds one Engine for many units of work — a worker writing one settlement
back after another over a run that lasts hours — must call engine.end_command().await
between every two of them, after a unit fails as well as after one lands. Within a command a
source may reuse what it read: a GitHub Projects source holds the board, its search answers
and the item records it resolved, on the grounds that nothing but itself writes the board
while one command runs. Across units that stops being true, and without the call a person's
edit to an issue's body, or a card they moved, between two units would be overwritten or
mis-moved by the next write. After the call no source answers a read or bases a write on
item content, statuses, board contents or search results it held before; each keeps only
identifiers and vocabulary that stay valid — Linear keeps its team's workflow states and its
workspace's project statuses, so a warm status write is still one request. The caller names no
plugin: every source decides what it may keep under the one rule. A one-shot use — the binary,
or an engine built for a single unit and dropped — needs no call, because dropping the engine
drops everything it held.
use ;
async
A failure reaches a linking caller as a typed value rather than as a document to parse: a
write that landed while a task it delivers could not be kept in step reports that entry as
DeliveryOutcome::Failed, whose Failure reads each member of the failure document —
class() among them — through the read-only getters its own API documentation lists.
To work on the repository instead, clone it and run just bootstrap; just --list shows
the rest.
Configure
One YAML document, onetaskgraph.yaml, discovered upward from the working directory and
layered over a user-level file at $XDG_CONFIG_HOME/onetaskgraph/config.yaml:
default_sources: # omitted means every configured source
page_size: 50
output: text # text | json
interactive: true # prompt for what a command was not given
sources:
work:
plugin: linear
config:
notes:
plugin: local-md
config:
A linear source's project, the id of one Linear project of the team, scopes the source to
that project: every read and listing is narrowed to it, a task written with no project is
placed in it, and a write to any project but that one is refused naming the scope. A status-only
write — task status set, or task update naming a status alone — is one mutation to the item
it names, wherever that item is filed, with no read before it.
sources:
example-team:
plugin: linear
config:
team: ENG
project: 986a467e-775a-4f8f-80dd-aca405063cf4 # optional
status_mapping:
backlog:
draft:
todo:
queued:
in-progress: In Progress
unknown:
done:
cancelled: Canceled
Status mapping
A github-projects or linear source names its statuses with one status_mapping, the same
grammar on both. Its keys are status categories — draft, backlog, todo, queued,
in-progress, done, cancelled, unknown — and each value is one of:
- a name, which is the category's name for a task and for a project alike;
null, which disables the category for every kind;- an object with the keys
taskandproject, each optional and each a name, which names the category per item kind; a kind the object leaves out leaves the category unmapped for that kind.
status_mapping:
todo: Todo # every kind
draft: null # disabled for every kind
done: # per kind
Anything else is refused when the configuration loads, naming the source, the category and the
part: a key that is not a category, an object naming no kind ({} — write null to disable a
category), naming another key, holding a null, or a blank name. So are two categories mapped
to one name of one kind, compared ignoring case — that name could read back as only one of them
— naming the source, the kind, both categories and the name; a bare name counts for both kinds,
and one name used by different categories in different kinds is fine.
Writes. A status is written as the name its category maps to for the kind of the item being
written. A write the source has no name for — a category it does not map, one set to null, or
one its per-kind object leaves out — is refused before any mutation, naming the source, the kind
(task or project), the category and the key to set, status_mapping.<category>.<kind>. So
is a mapped name that kind's vocabulary does not hold, naming the name. Nothing falls back by a
status's type, or by the name it was called where it came from.
Reads. An item at a name its kind maps reads as that category, under the name; every other
name reads as unknown, under its own name. --status returns exactly the items that read as
each category it names, unknown among them.
On a GitHub Projects board both kinds' names are options of its one Status field. A
category the source does not mention keeps its shipped default for both kinds — backlog →
Backlog, todo → Todo, queued → Queued, in-progress → In Progress, done → Done,
cancelled → Cancelled; draft and unknown have none — and a category it does mention uses
only what it configures. done and cancelled close the issue as completed or not planned,
for both kinds, and a closed issue reads by its state whatever its option.
On Linear a task's names are the configured team's workflow states and a project's are the
workspace's project statuses — two vocabularies. Linear has no built-in names, so a source
without a status_mapping reads every item as unknown and refuses every status write. A
source resolves its team, the team's workflow states and the workspace's project statuses in
one request, the first time a write or sources fields needs them, and holds them for its own
lifetime — never longer, and never shared with another source. A name it does not find is
looked for once more in a fresh read, so one added in Linear since is found; nothing a failed
call answered is held. A status write sends its mutation alone, with no read of the issue
first — one request once the resolution is held. sources fields <SOURCE> --apply creates every
mapped name a vocabulary lacks — a workflow state on the team for a task's, a project status in
the workspace for a project's, both for a bare name — of the type its category derives, in
Linear's neutral grey, a project status after the workspace's last; it never renames, retypes
or deletes a name that is there, and reports one present under another type as it is:
| Category | Workflow state | Project status |
|---|---|---|
backlog, draft |
backlog |
backlog |
todo, queued |
unstarted |
planned |
in-progress, unknown |
started |
started |
done |
completed |
completed |
cancelled |
canceled |
canceled |
A create Linear refuses stops the run, naming the name and its kind, and the report names every
name created before it; run --apply again once the refusal is lifted, and it creates only what
is still missing.
Migrating to 0.3. This grammar changes what an existing configuration does:
- On Linear, a write no longer falls back to the team's first state of the category's type, or resolves a category by the status's own name: a category with no mapped name is refused.
- On Linear, a project write resolves its status through the project side of the mapping, and no longer through the item's own status name.
- On Linear, a name the mapping does not give its kind reads as
unknown, and no longer by itsWorkflowState.typeorProjectStatusType. - A Linear source without a
status_mappingrefuses every status write. - On GitHub Projects, a category configured as a per-kind object no longer gets the shipped default for the kind it leaves out.
- On Linear, a status write no longer reads the issue first. Setting the category an issue
already reads as sends a same-state update; setting
unknownon an issue at a name the mapping does not give a task moves it to the mappedunknownname; and an issue Linear does not hold is reported from the mutation's own not-found refusal. - On Linear, a status-only write through a source scoped to one project (
projectset) goes to the item it names wherever that item is filed, while reads, listings and creation stay scoped. - On Linear,
sources fields --applycreates every missing mapped name, as workflow states and project statuses, where it used to be refused.
Every setting is reachable at three layers, lowest precedence first: the file, then the environment, then a command-line flag.
An environment variable is ONETASKGRAPH_ followed by the config path, each segment
upper-cased with - replaced by _ and segments joined by a double underscore; a list is
comma-separated:
| Variable | Sets |
|---|---|
ONETASKGRAPH_PAGE_SIZE=100 |
top-level page_size |
ONETASKGRAPH_DEFAULT_SOURCES=work,notes |
top-level default_sources |
ONETASKGRAPH_INTERACTIVE=false |
top-level interactive |
ONETASKGRAPH_SOURCES__WORK__CONFIG__ROOT=/tmp/tasks |
the root of the source named work |
ONETASKGRAPH_SOURCES__GH_MAIN__PLUGIN=github-projects |
the plugin of the source named gh-main |
The mapping is unambiguous because a source name may not contain an underscore.
On the command line the same dotted path is --set, and a few common settings have named
flags of their own:
onetaskgraph config show is what makes precedence something you can see rather than
something you have to reason about: it prints every setting, its value, and the layer it
came from — which file, which environment variable, or which flag — and --json renders
the same thing for a script.
That third column is the whole point: default_sources above came from an environment
variable, page_size from a flag that outranks the document's own value, and every
sources.* entry from the document, named by path.
Relative paths in a configuration document
A relative filesystem path a configuration document supplies is resolved against the
directory holding that document. onetaskgraph.yaml is discovered by walking upward
from the working directory, so the document is very often not in the directory you ran the
command in — a local-md root: of plans in a checkout's own document means that
checkout's plans, whether you run from the checkout, from a crate three levels inside it,
or from a worktree beside it.
A relative path the environment layer or a flag supplies keeps resolving against the
process working directory, because there is no document to rebase it on:
ONETASKGRAPH_SOURCES__PLANS__CONFIG__ROOT=plans and --set sources.plans.config.root=plans both mean plans under wherever the command is running.
Naming the directory outright is still how a launcher points a run at a store that is
neither beside its own document nor beside its working directory:
ONETASKGRAPH_SOURCES__PLANS__CONFIG__ROOT=/srv/plans
onetaskgraph config show reports the resolved path beside the document that supplied it,
so what a run will really read is something you can see rather than something you have to
work out.
The rule reaches only the configuration fields a plugin itself declares as paths, and each
plugin's own page says which of its fields those are. It holds across the subprocess seam
too, so a local-md root relative to its document names the same directory in process and
behind the seam: the engine rewrites nothing in a settings: block, and the child learns
which document's directory to measure its declared fields from through the handshake.
docs/plugin-protocol.md §3.8 is the one statement of that
member, including when it is absent and what a plugin written before it does.
Credentials
A configuration document never holds a credential — it names the environment variable that
does (api_key_env: LINEAR_API_KEY). Before resolving sources the CLI reads
$XDG_CONFIG_HOME/onetaskgraph/secrets.env (override with ONETASKGRAPH_SECRETS_FILE) as
KEY=VALUE lines with # comments, and applies each value only where the process
environment does not already define that name — so anything you exported wins. A missing
file is not an error. A credential is never printed, never in debug output, and never in a
log line.
There is exactly one name per credential everywhere — in that file, in the configuration,
in the documentation and in CI: LINEAR_API_KEY and GH_PROJECTS_TOKEN. Nothing anywhere
translates between spellings.
Routing, and plans that span sources
A source may send what is written to it somewhere else, chosen by the repositories the item
concerns. routes sits beside plugin and config — it is the engine's, not the plugin's:
sources:
plans:
plugin: github-projects
config:
routes:
- repositories:
to: example-team
example-team:
plugin: linear
config:
An entry matches an item when every one of the item's repositories matches at least one of
its patterns. A pattern is a normalized origin, host/owner/name, where * matches exactly
one whole segment. An item naming no repository matches nothing. The first entry that
matches wins, and an item no entry matches stays in the source itself. A configuration with
no routes writes exactly where it always did.
Each of these is refused when the configuration loads, naming the source and the entry: a
to naming no configured source, a to naming the source itself, a to naming a source
that has routes of its own (a route never chains), an entry whose repositories is empty,
and a malformed pattern.
routes is set at every layer. A document writes it as a list; --set and the environment
address one entry by its index, because neither can spell a list of objects:
ONETASKGRAPH_SOURCES__PLANS__ROUTES__0__REPOSITORIES='github.com/example-org/*' \
ONETASKGRAPH_SOURCES__PLANS__ROUTES__0__TO=example-team
A comma in a repositories value makes it a list of patterns, as a comma does everywhere
else. Entries are
numbered from 0 with no gaps, and an entry one layer sets replaces the whole routes a lower
layer set rather than merging into it. config show prints routes with the layer it came
from — the document's list as one row, a flag's or a variable's entries one row per field.
onetaskgraph sources route <SOURCE> --repository <R>... answers where an item would go —
{"source": "plans", "destination": "example-team", "route": 0}, with route null when no
entry matched — from configuration alone.
Which writes route, and where each item lands
task create <SOURCE>, task copy --to <SOURCE>, project copy --to <SOURCE> (with or
without its tasks, and with --member) and document copy --to <SOURCE> all place each item
by the named source's routes. So does the engine's own member copy, which is what a running
engine's write-back uses.
- A task goes where its own repositories route it.
- A project's home goes to a routed source only when every task the copy carries routes
there, and so do the project's own repositories when it names any. Any other project's home
is the named destination. With
--no-tasks, the project's own repositories alone decide. A home that already exists stays where it is: a task added later that routes back to the named destination goes into a member project there. - A task routed away from its home's source goes into the home's member project in
that source, found through the home's
onetaskgraph.members— or created, with the home's title and status andonetaskgraph.member_ofnaming the home, and recorded on the home in the same copy. A home has at most one member per source.task createfiles a routed task the same way, under the member of the project it names. - A document goes with its project's home.
A task's project is still a native id of its own source: a plan spanning two sources is a home plus a member project in the other, never a project holding another source's tasks. An edge between two items that land in different sources is recorded as a cross-source edge on its near end, either way round.
Every outcome of a routed copy's report carries placed: the source it landed in and the
index of the route entry that placed it, or null — a dry run's included, so a would-be
create says where it would go. A copy into a source with no routes reports no placed, and
its output is what it always was. A copy across several sources is still complete or never
happened: its undo covers every source it wrote. The onetaskgraph.copies link a copy
records is keyed by the source the item really landed in.
An item whose counterpart already sits in another source than the one it routes to now is refused before anything is written, naming both: a repository changed after a copy is a person's decision about where the work lives, not a silent move. What finds a counterpart there is its link or its origin; a counterpart neither names is looked for only in the source the item routes to.
Reading a home with its members
reads the home's tasks and the tasks of every project its onetaskgraph.members names, each
under its own source, paged across all of them as one walk. The members are read off the home
on every request — nothing about them is kept — and a member that cannot be read is reported
in the response's errors, as any source failure is, never left out in silence. project show
of a home prints its members, and of a member its home. The SDKs spell it
task_list(project=..., members=True) and taskList({ project, members: true }).
Addressing
Every item is qualified by the source it came from and rendered <source>:<native> —
work:ENG-142, notes:2026-08-inbox. Parsing splits on the first colon, so a native
id may contain colons freely.
Custom metadata and repositories
A task and a project each carry a caller-defined metadata map and the repositories
they concern, and a dependency edge names both its ends by kind and qualified id — so an
edge may cross projects, cross the task and project levels, and cross sources.
docs/metadata.md says which keys are reserved and where each source
keeps them.
The far end of the second row is in another source. It is reported by qualified id and kind and never followed: opening it is a command of your own.
Writing a source in another language
A source that is not a Rust crate speaks a line-oriented JSON protocol over stdio:
docs/plugin-protocol.md specifies it completely — the
framing, the capability handshake, one method per trait method, and the error envelope.
Configure one with the subprocess plugin, which names the program to run and hands it
its own settings verbatim:
sources:
notes:
plugin: subprocess
config:
command: /usr/local/bin/my-source
args:
secrets: # forwarded in the handshake; nothing else is
settings: # this source's own `config:` block
A source behind that seam is a source like any other: it declares its own capabilities,
so a plan says pushed down for what it applies itself, and the engine compensates for
the rest exactly as it does in process.
onetaskgraph-source is the reference implementation of that plugin side, and the test
host this repository drives its own journeys against: it hosts any built-in plugin over the
same protocol, so every journey runs a second time over a real pipe to a real second
process, and so you can read a working peer beside the specification.
It is not part of the command-line interface. Nothing onetaskgraph does needs it at
run time, and nothing may depend on finding it beside an installed CLI or resolve it off a
search path — a downstream test chain that did picked up an unrelated stale build and
treated a reference host as a runtime dependency. Read it, or build it, from a checkout:
Licence
MIT. See LICENSE.