{;; Exact pins — the Closure compiler version rides on these two, and the
 ;; committed dist/ must rebuild byte-identical (deploy/check-dist.sh). Same
 ;; known-good pair as the rest of the suite's CLJS repos.
 ;;
 ;; Relative paths only — never :local/root, never an absolute path: the idpat
 ;; publish gate scans every decompressed git object for local path fragments.
 ;;
 ;; THE RULE: our OWN kits are ClojureScript libraries and are consumed as
 ;; SOURCE off this classpath — one compiled copy per build, one cljs.core, one
 ;; `Identity` class in the build. Foreign JS/native packages (@orbitdb/core,
 ;; the libp2p family, helia, ws, node:*) stay string requires: that is ordinary
 ;; interop, not the hybrid. ardegazu-id-kit is still reached through its npm
 ;; install, so the sha pin in package.json remains the ONE cross-repo
 ;; dependency mechanism (dev/docs/CLJS.md); this path only points the compiler
 ;; at what that pin unpacked, and this file needs no second pin.
 ;;
 ;; The comment that used to sit here argued the opposite — that id-kit had to
 ;; stay a dist import because this kit RE-EXPORTS `Identity`, and compiling
 ;; id-kit in here would make that a different class object from
 ;; `ardegazu-id-kit`'s. That was true only of the MIXED regime it described,
 ;; and it is now false in both directions:
 ;;   - Nothing in the suite does `instanceof` against id-kit's Identity. Every
 ;;     consumer duck-types it (`.fromSeed`, `.signRaw`, `.assert`,
 ;;     `.publicKeyB64`, `.fingerprint`), and id-kit attaches that whole surface
 ;;     with STRING names precisely so :advanced cannot rename it — which is
 ;;     what makes a source-compiled class and a dist-compiled one
 ;;     interchangeable across the boundary rather than dangerous.
 ;;   - Keeping the dist import is what MANUFACTURED a second copy, now that
 ;;     every other kit compiles id-kit from source: the bot's worker would hold
 ;;     rooms-kit's dist-imported id-kit while its own graph compiled another.
 ;; Our `Identity` is therefore ardegazu.id.identity/Identity, compiled here.
 ;; Same seed in, same publicKeyB64 out, same signatures on the wire — the
 ;; interop that matters is cryptographic, not object identity, and
 ;; test/api.test.mjs now proves exactly that against the installed id-kit dist.
 ;;
 ;; No new npm dependency comes with this: ardegazu.id.{identity,crypto} are
 ;; pure WebCrypto + base64 and require nothing outside the suite. @noble/curves
 ;; lives only in ardegazu.id.xkey, which this kit does not use.
 ;;
 ;; PROMESA IS DELIBERATELY ABSENT, and this is the note that says so before
 ;; someone adds it to match chat. dev/docs/CLJS.md names `p/let` as the suite's
 ;; one async idiom; it was measured here and rejected for three reasons that
 ;; are all specific to this kit, not to the idiom:
 ;;
 ;;  1. SIZE. chat pays ~1.8 KB gz because it builds :advanced and Closure DCEs
 ;;     everything promesa.core does not reach. This build is :simple (see
 ;;     shadow-cljs.edn — :simple removes the externs-inference risk class, and
 ;;     almost every line here is interop with untyped ESM). Nothing is DCE'd, so
 ;;     one `p/let` in one function measured at +115,703 bytes raw / +12,272 gz
 ;;     on dist/shared.js: +10.4% raw, +8.2% gz, for the whole runtime.
 ;;  2. MICROTASK HOPS. promesa's ClojureScript runtime is NOT js/Promise — it
 ;;     ships promesa/impl/promise.js and schedules each link through its own
 ;;     nextTick. This kit's recorded transcripts settle with a FIXED number of
 ;;     microtask turns (`tickMicro(8)` in test/helpers/fakes.mjs), so a chain
 ;;     with a different hop count moves a golden vector — and the vectors are
 ;;     the cross-implementation proof against the retired TypeScript. They may
 ;;     be added to, never regenerated.
 ;;  3. THE LIBRARY BOUNDARY. Every promise this kit returns crosses into a
 ;;     consumer (the bot, its per-room workers, @orbitdb/core callbacks). A
 ;;     `p/let` result is thenable but not `instanceof Promise`, and an
 ;;     unhandled rejection on one does not reach the global handler. chat owns
 ;;     both ends of its chains; a kit does not.
 ;;
 ;; What replaces it: `later` / `attempt` / `each-in-order!` / `every-in-order!`
 ;; in ardegazu.rooms.js, four native-promise combinators that name the four
 ;; chain shapes this tree actually repeats. Flattening came from decomposition
 ;; instead — lib/log's `open` and lib/net's `start` are the two that needed it.
 :paths ["src" "node_modules/ardegazu-id-kit/src"]
 :deps {org.clojure/clojurescript {:mvn/version "1.12.134"}
        thheller/shadow-cljs {:mvn/version "3.3.6"}}
 :aliases
 {:dev {:extra-deps {cider/cider-nrepl {:mvn/version "0.59.0"}}}
  ;; cljfmt is a TOOL, not a dependency. :replace-deps keeps it off the build
  ;; classpath (neither shadow-cljs nor clojurescript is loaded to run it) and
  ;; :replace-paths keeps the sources off it too — cljfmt only ever reads the
  ;; files named on the command line. Exact pin, same rule as the two above.
  ;; The config is committed beside this file; both paths are relative, because
  ;; the idpat publish gate aborts on a local filesystem path in any object.
  ;; Wired into `npm test` as the FIRST link of the test script itself, not as
  ;; a `pretest` hook: an npm lifecycle hook is skipped outright under
  ;; `ignore-scripts=true`, which is a gate that passes by not running.
  :cljfmt {:replace-deps {dev.weavejester/cljfmt {:mvn/version "0.13.1"}}
           :replace-paths []
           :main-opts ["-m" "cljfmt.main" "--config" ".cljfmt.edn"]}
  ;; clj-kondo is a TOOL too, same shape and same reasons as :cljfmt above:
  ;; :replace-deps keeps it off the build classpath and :replace-paths keeps the
  ;; sources off its own, since it reads only the paths named on the command
  ;; line. Running it as a pinned JVM dep rather than the native binary is
  ;; deliberate — the gate must not depend on what happens to be installed, and
  ;; three versions (2025.06.05, 2026.01.19 and this pin) were measured to give
  ;; identical findings on this tree, as did the native binary.
  ;; Config and the defclass hook are committed at .clj-kondo/ beside this file;
  ;; every path here is relative, because the idpat publish gate aborts on a
  ;; local filesystem path in any object.
  ;; Wired into `npm test` as the second link of the test script itself, right
  ;; after cljfmt and before anything compiles — never as a `pretest` hook, which
  ;; `ignore-scripts=true` skips outright (a gate that passes by not running).
  ;; --fail-level warning is the default and is what makes this a gate; the
  ;; config promotes the one :info linter this tree can trip so it cannot print
  ;; a finding and still exit 0.
  :clj-kondo {:replace-deps {clj-kondo/clj-kondo {:mvn/version "2026.08.04"}}
              :replace-paths []
              :main-opts ["-m" "clj-kondo.main" "--lint" "src" "--fail-level" "warning"]}}}