open-standards/wai/catalogue
Repository files (latest commit first)
Filename Latest commit message Latest commit date
Transaction Science 2fd26925d3 wai identify: what any file is, from its bytes alone
`wai inspect` reads envelopes; most files a toolchain meets are not envelopes, and the
first question about one is what it is. `wai identify <file>` (operation media.inspect,
extensions/media-identify.md) answers it for any file:

- the formats the bytes fit, each with the offsets of the bytes that say so, how far its
  structure was followed and where and why that stopped;
- a container's tracks as it states them: codec, size, channels, rate, protection;
- the registered capabilities they map to, and what a format probe reads of them;
- signatures found inside the file;
- where no format is recognised, measures that still hold: how the byte values are
  spread, uniform windows, and the lags at which the bytes repeat.

The verdict is identified, ambiguous, signature_only or unknown, and there is no
confidence figure: a candidate's strength says what was checked. Nothing is decoded, no
protected track is opened, and unknown stays unknown. Walks cover ISO media, Matroska,
transport streams, Ogg, FLAC, RIFF, MPEG audio, ADTS, PNG, JPEG, GIF, IVF, YUV4MPEG2,
FLV, AV1 OBU and Annex B streams, and HLS and DASH text; 35 more formats are known by
their signature. Every walk is bounded, reads a file where its lengths lead (a large file
is never held whole), and has been given hostile bytes; it builds for wasm32.

The agent door has it as media_inspect, its results marked untrusted_content: a file's
author chooses the strings in it.

identify-conformance/ holds files written by real muxers and encoders from test patterns
and tones, what an independent media prober reads of them (prober.json), and generated
cases sitting on each rule's edge. identify_verify.py is a second implementation written
from the text; it reads the extension's tables, gives every file its whole result, and
holds ext/ to the prober's reading. tests/identify.rs holds the reference's tables to
the extension's.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
2026-11-04 05:20:15 -05:00
..
ops.json wai identify: what any file is, from its bytes alone 2026-11-04 05:20:15 -05:00
README.md wai upscale and wai tonemap: an exact derivation from the command line, with its record and its check 2026-11-02 08:43:49 -05:00

The operation catalogue

Every operation of the WAI toolchain, once. ops.json is written by wai ops export from the table in wai-rs/src/catalogue.rs, and CI fails when the two differ (.github/workflows/wai-catalogue.yml). Do not edit it by hand: change the table, then

cd wai/wai-rs && cargo run -q --no-default-features --bin wai -- ops export --out ../catalogue/ops.json

An operation

An operation is named noun.verb: lowercase ASCII, a singular noun, and a verb from naming rule N3's closed list (create, list, show, update, delete, check, verify, run, build, export, import, start, stop, join, leave, grant, invite, revoke, rotate, tail, search, test, load, encode, decode, inspect, wrap, unwrap, fetch, generate, promote, rollback, lock, upgrade, setup, guide, stream, upscale, tonemap), never a near-synonym (get, describe and info are show; keygen, init and new are create). A command a person already types keeps its spelling as the operation's command line: wai hello is toolchain.test, wai stage refine is stage.encode. Its entry (format wai.ops/1) carries:

Member Meaning
id noun.verb
summary One line, for help, docs and the search
cli, aliases The command line today, and other spellings; every operation also answers to wai <noun> <verb>
usage The synopsis
placement local, client (on this machine, against a service it names), service or both
effect reads, runs, writes-project, writes-target or destructive
dry_run Whether it takes --dry-run: every operation whose effect is not reads
confirm When it waits for --yes: never, when-replacing-an-unnamed-file, when-replacing-a-file, when-changing-the-target-or-replacing-a-file or always
route Its HTTP method and path, for an operation that runs on a service
toolset core, dev, app, quantum or maintainer
stability stable, preview or internal
input, output, stream JSON Schemas: its arguments (each with the flag or position that gives it), the data of its --json envelope, and each line of a streaming operation
since, deprecated The toolchain version that first carries it; its replacement, once renamed
receipt Whether it writes a signed receipt or record that a verify operation checks
needs The build features it needs (all or any of them)
names Its name in each door

One name in every door

Door envelope.inspect Held by
CLI wai inspect, wai envelope inspect the wai binary's tests (tests/cli_ops.rs)
MCP envelope_inspect scripts/check-wai-ops-doors.py
HTTP API operationId: envelope.inspect scripts/check-wai-ops-doors.py
C wai_envelope_inspect scripts/check-wai-ops-doors.py
Rust wai::envelope::inspect
JavaScript, Swift, Kotlin wai.envelope.inspect (Swift and Kotlin: WaiEnvelope.inspect) scripts/check-wai-ops-doors.py (Swift, Kotlin)

A door that does not carry an operation yet is reported as such by the check; a door that names one differently fails it. In the C header and the native sinks, a name is an operation's when its noun is a catalogue noun and its verb an operation verb; the codec, envelope and sink functions that are not operations are library API.

Agent rules

Every command of the catalogue keeps them.

Exit Meaning
0 Done; every check held
1 Ran, and a check failed
2 Usage error or invalid input
3 Authentication or permission
4 Confirmation required; nothing changed
5 Not found (file, target, room, operation, or a build feature)
6 Target unavailable; safe to retry
7 Internal error

With --json, stdout holds one envelope and logs stay on stderr; without it the output is text, on a terminal or not. One module holds these rules for every command (wai-rs/src/bin/wai_cli/dx/envelope.rs).

{ "wai": "envelope/1", "op": "envelope.wrap", "ok": false, "exit": 4,
  "error": { "code": "confirmation-required", "message": "out.wai exists, and the command line did not name it as the output (-o); nothing was changed",
             "hint": "read the proposed changes, then run the confirming command", "retryable": false },
  "proposed": [ { "replace": "out.wai" } ],
  "confirm": "wai wrap wai.text.zstd p.bin --yes" }

A success carries data: the command's own JSON where it has one, {"text": ...} otherwise. A failure carries its exit; a check that failed (exit 1) carries both error and data. A dry run adds "dry_run": true and the plan of what it would write, and of each request it would send to a service it names (a fleet service), none of them sent. --yes is never implied; off a terminal, or with --no-input, a change that waits for it exits 4. --confirm is a deprecated spelling of --yes that does the same; the export lists the agent flags (agent_flags) with it as a deprecated alias.

wai sends nothing off your machine unless you name a remote service.