@outerlayer/cli 0.34.0-next.20261024225538 → 0.34.0-next.20261025132734

package/dist/build-info.json

@@ -1 +1 @@
-{"id":"2026-10-24T22:57:19.596Z+5cf285c","root":"/home/runner/_work/outerlayer/outerlayer","pkg":"outerlayer"}
+{"id":"2026-10-25T13:28:45.990Z+09b988d","root":"/home/runner/_work/outerlayer/outerlayer","pkg":"outerlayer"}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@outerlayer/cli",
-  "version": "0.34.0-next.20261024225538",
+  "version": "0.34.0-next.20261025132734",
   "description": "OuterLayer CLI — commands: init, claude, connect, doctor, login, logout, sync, daemon (alias: watch), emit, findings, work, spec, runner, hooks, mcp, context. Capture and sync coding-agent sessions, and record proof of factory work.",
   "license": "Apache-2.0",
   "author": "Magu Studios, Inc.",

package/skill-pack/maintained/outerlayer-build/SKILL.md

@@ -235,14 +235,17 @@ Run them in order.
 
    Set `proof` to an artifact kind (`video`, `screenshot`, `report`, `log`,
    `test` or `file`). When the issue states a proof, as in
-   `(proof: screenshot)`, use it. A criterion that names a test seam, as in
-   `(seam: component)`, keeps `null`: a test proves it. Infer the proof of
-   every other criterion from its Then. A Then that a person sees in the
-   product's user interface takes `video` when its When has more than one
-   step, and `screenshot` when it shows one state. A label, a disabled button
-   and a closed box count. A Then that no screen shows, such as a stored row,
-   keeps `null`. Infer a screen proof only when the build can run and record
-   that interface. The final report lists each proof the build inferred.
+   `(proof: screenshot)`, use it. Infer the proof of every other criterion
+   from its Then, even when it names a test seam such as `(seam: component)`.
+   A seam says where the criterion's test binds, not whether a person needs to
+   see the result. A Then that a person sees in the product's user interface
+   takes `video` when its When has more than one step, and `screenshot` when
+   it shows one state. A label, a disabled button and a closed box count. A
+   Then that a person reads in the CLI's output, such as a doctor line or an
+   error message, takes `log`. A Then that nothing shows, such as a stored
+   row, keeps `null`, and its test proves it. Infer a screen proof only when
+   the build can run and record that interface. The final report lists each
+   proof the build inferred.
    Then run:
 
        node .outerlayer/skills/outerlayer-build/scripts/record-criteria.mjs --item "$OUTERLAYER_WORK" --criteria-file <file>

package/skill-pack/maintained/outerlayer-build/references/agent-briefs.md

@@ -286,16 +286,19 @@ tests at HEAD that hold the id of each proof that is not current, and runs them.
 - A proof that no test can make stays without a test. Give the reason in
   `build-reports/evidence.md`.
 
-Commit the setup file and the tests. Then run
+Commit the setup file and every test. Then run
 `outerlayer work proofs --record` from the committed checkout. It stops when a
 tracked file differs from HEAD. It runs only what is committed at HEAD, and
-emits each proof at HEAD. Never emit by hand a proof that a test can make.
-Always pass `--pr <number>` when the item links more than one pull
-request in the repository. Record again after every later commit. It skips a
-proof that is current.
+emits each proof at HEAD. It skips a proof that is current. Never emit by hand
+a proof that a test can make. Always pass `--pr <number>` when the item links
+more than one pull request in the repository.
+
+The output names each proof that is still not current, with the reason. Fix
+that test, commit it and record again. Do not record again only because the
+head moved. Each run starts the app again. The orchestrator records once, at
+the run's last head.
 
 Run `outerlayer sync`, then `outerlayer work proofs --item "$OUTERLAYER_WORK"`.
-For each proof it lists as out of date or missing, fix its test and record again.
 
 Write `build-reports/evidence.md` naming every criterion with its artifact, or
 with the reason none could be produced. A criterion you could not evidence is

package/skill-pack/maintained/outerlayer-build/references/evidence.md

@@ -27,20 +27,22 @@ finds those tests at HEAD, runs them through `.outerlayer-proofs.yaml`, and emit
 each proof. A capture script outside the tests is not allowed, because no later
 session can run it again.
 
-After the evidence agent returns, and after every later commit, run
-`outerlayer sync`, then `outerlayer work proofs --item "$OUTERLAYER_WORK"`. It
-lists each declared proof as current, out of date or missing at the checkout's
-head. Commit the work, then run `outerlayer work proofs --record`: it records
-each proof that is not current and skips the rest. Run the list command once
-more before the run ends. A criterion still listed goes in
-`build-reports/evidence.md` with the reason. The command leaves out criteria
-that declare a test proof.
-
-When a commit lands on the branch after the evidence agent captured, run
-`outerlayer work proofs --record` again. It records every proof that is not
-current at the new head. The final report names the commit and the proofs it
-recorded again. A commit that lands after this call reopens it. Never drive the
-browser yourself.
+After the evidence agent returns, run `outerlayer sync`, then
+`outerlayer work proofs --item "$OUTERLAYER_WORK"`. It lists each declared proof
+as current, out of date or missing at the checkout's head. The command leaves
+out criteria that declare a test proof.
+
+A commit makes every proof out of date, and each `--record` run starts the app
+again. So record only at the run's last head. First make every commit the run
+still needs, such as a CI fix or a merge of the base branch. Push the last
+commit, then run `outerlayer work proofs --record` once. It records each
+proof that is not current and skips the rest. If a commit lands after that,
+record once more after it. The final report names that head and the proofs
+recorded there.
+
+Run the list command once more before the run ends. A criterion still listed
+goes in `build-reports/evidence.md` with the reason. Never drive the browser
+yourself.
 
 Re-read the pull request's evidence comment. Resolve or escalate every row that
 needs attention. A row is resolved when a re-read shows it, not when the fixing

package/skill-pack/shipped.json

@@ -87,10 +87,12 @@
       "82f75a53c591f94615dc27c099e94fcf95e2a09b95aceb38b9022d5f7f054cb0",
       "8d3a5131d0c82302279d319c51bcde74f7677e8d2e3f6cd68aa7f3ea84fa1051",
       "9039ae30d4e836b5c5cfba4a14344e46fd13694ae9e4d88e750abce34dfcd9e6",
+      "9ef1644ab58a06dbfb9dc6e6344413014cd1ac436dcebd3c2cab632c2a3063bd",
       "ba26711f3ac400f38e2bb78d4e410cde940e666317cd9aeb3fa949d89bb39f15",
       "fc2bec8d43764c4e4b40b0ee761b7f9ee8646950dd4e52392f652e68679f79ea"
     ],
     ".outerlayer/skills/writing-specs/SKILL.md": [
+      "78ffced3b8f3e34306968c69c467292adbc2d609fac20a6778f8bd4a62fbe117",
       "a956aac4249118a058c4327c5751472883588720950d4a96f80c7e91f454bef9",
       "d9423f48add241b2cf23bf49b467858355882727a08e068937a07571399555d8",
       "dbadb39cba5dae936e7c4b97cb852c59072e4a60e44caec01b128a045c1be090",

package/skill-pack/template/spec/SKILL.md

@@ -42,7 +42,9 @@ Run them in order. Do not skip the approval.
    a stable id the team chooses. Every criterion whose outcome a user can see
    carries a proof form beside its id: `(proof: video)` for a flow of more than
    one step, and `(proof: screenshot)` for a single state. A label, a disabled
-   button, a closed box and a new line are visible outcomes too. The build
+   button, a closed box and a new line are visible outcomes too. Output a
+   person reads from the CLI, such as an error message, carries
+   `(proof: log)`. The build
    records each one as a chapter of the item's demo. A test alongside it does
    not replace the proof form. A test is the proof only where nothing visible
    changes, such as a stored row or a job that does not start. A criterion

package/skill-pack/template/writing-specs/SKILL.md

@@ -100,9 +100,10 @@ implementing. Never write them by reading the finished implementation back.
 - **Every promised error path gets its own criterion.** An unstated edge case
   is not left open. It is guessed at during implementation.
 - **`(proof: ...)` between the id and Given** on every criterion whose outcome
-  a user can see. Use `video` for a flow of more than one step and `screenshot`
-  for a single state. A test is the proof only where nothing visible changes. The proof shows the running page, never
-  a component render. The outerlayer-evidence skill covers capture.
+  a user can see. Use `video` for a multi-step flow, `screenshot` for one state
+  and `log` for CLI output. A test is the proof only where nothing visible
+  changes. The proof shows the running
+  page, never a component render. The outerlayer-evidence skill covers capture.
 
 ## The two verdicts