@devrouter/cli 0.0.72 → 0.0.74

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@devrouter/cli",
-  "version": "0.0.72",
+  "version": "0.0.74",
   "description": "Local dev routing CLI with shared Traefik reverse proxy",
   "author": "Roland Schlaefli",
   "homepage": "https://github.com/rschlaefli/devrouter#readme",

package/upgrade-prompts/0.0.73.md

@@ -0,0 +1,54 @@
+# Upgrade to devrouter 0.0.73
+
+This patch release adds opt-in bounded automatic recovery and fixes three
+capacity-managed defects found while dogfooding the eLearning canary.
+
+- Capacity-enrolled `exec` now returns the wrapped command's output. The
+  controller captured a supervised worker's stdout and stderr into its bounded
+  buffer and the client exposed it through `onOutput`, but the CLI never
+  forwarded it, so `devrouter exec <path> -- <command>` returned only the exit
+  status. The controller-supervised path now streams each bounded page to the
+  caller and prints one notice if the buffer dropped earlier bytes.
+- A capacity-enrolled operation keeps its effect authority while the controller
+  cannot observe capacity. Renewal previously wrote a zero deadline before each
+  attempt and again on failure, so one missing, stale, or unknown sample revoked
+  authority mid-start and later commands failed with `Capacity effect authority
+  is absent or stale.` Renewal now extends only when every reserved domain has a
+  usable sample. Ownership evidence also compares environment identity instead
+  of whole journal records, so the controller rewriting its own bookkeeping no
+  longer reports `unknown` for a healthy runtime.
+- The observation collector reports a required managed process that has
+  positively exited as a failed capability instead of `unknown`, so bounded
+  recovery can act on it. Ambiguous or unreadable process state stays `unknown`.
+- Optional bounded automatic recovery. A new `recovery` block in the capacity
+  policy enables a supervisor that reacts to a positively failed required
+  capability with one journal-admitted corrective `ensure`, bounded by
+  per-scope restart counts, an aggregate corrective-action cap and a rolling
+  observation window. The block is absent by default. Parking and controlled
+  resume remain model-only, and there is no OOM detection: a container-local OOM
+  still surfaces only as a failed capability.
+
+No `.devrouter.yml` schema, CLI flag, or default behavior changes.
+
+## Task
+
+Install the published release and update `.devrouter.yml` devrouter.version to
+0.0.73. No configuration change is required. Enabling bounded recovery is
+optional: add a `recovery` block to the operator capacity policy only when you
+want automatic corrective ensures.
+
+## Validation
+
+On a capacity-enrolled checkout, run
+`devrouter exec <path> -- bash -lc 'echo checked'` and confirm the command's
+output appears. Confirm a failing command still returns its non-zero exit
+status, and that `devrouter ensure <path>` reports the environment ready with
+its routes published. If you enabled recovery, confirm an environment whose
+required process is positively absent is corrected by one admitted `ensure`
+within the configured bound.
+
+## Report
+
+Report the installed CLI version, the enrolled repository, the observed `exec`
+output and exit code, the final `ensure` readiness with its route count, and
+whether the recovery policy was enabled.

package/upgrade-prompts/0.0.74.md

@@ -0,0 +1,35 @@
+# Upgrade to devrouter 0.0.74
+
+This patch fixes canonical stop after managed Devsy startup fails before
+registration, including Devsy-only installations without the legacy DevPod CLI.
+It preserves completed startup outcomes and refuses unknown ownership, live
+workers, surviving containers (including stopped containers), and routes.
+No configuration schema, command, or flag migration is required. Successful
+absence proof is reported as `runtimeAbsent: true` in stop JSON.
+
+## Task
+
+Use the published 0.0.74 CLI. Update `.devrouter.yml` devrouter.version to 0.0.74
+when consumer source changes are authorized; preserve another task's staged
+work. No DevPod installation or runtime-selection change is needed. Refresh
+generated agent guidance only when that repository's owner authorizes it.
+
+For an affected exact linked checkout, use canonical `devrouter stop <path>
+--json`. A completed journal alone is not evidence that resources are absent;
+do not edit lifecycle state or delete provider data to bypass a refusal. This
+fix does not recover retained-container identity drift or replace surviving
+resources.
+
+## Validation
+
+Verify the CLI version and exact checkout ownership. Run the supported stop
+under the checkout owner's recovery authority. Require successful absence proof
+and settled lifecycle state before retrying ensure. Revalidate the original port
+conflict before that retry, then require provider, application and route proof.
+Do not infer consumer recovery from package or regression tests.
+
+## Report
+
+Report the installed CLI version and release revision, exact checkout, stop
+result, lifecycle state, subsequent ensure and route/application evidence, or
+the precise fail-closed blocker. Confirm source/index and runtime data retention.