fix(lifecycle): recover owned mux sessions

.codex/skills/hack-cli/SKILL.md

@@ -113,6 +113,8 @@ Use `hack` as the primary interface for local-first development.
 - For fixed-port host helpers such as SSM tunnels or local proxies, set `singleton.ports` and usually `onConflict: "adopt"` so Hack reuses a healthy existing listener instead of starting duplicate tunnel stacks.
 - `singleton` is a listener guard, not process ownership transfer; adopted external processes are left running on `hack down`.
 - Inspect lifecycle status via `hack projects --details` and stream via `hack logs <service-or-process>`.
+- Lifecycle session recovery is ownership-proven: Hack adopts healthy token-, definition-, and environment-matched sessions, replaces owned stale sessions, and refuses to kill same-name sessions without deterministic ownership proof.
+- `hack doctor --fix` reaps an orphan lifecycle session only when mux ownership is proven and its Compose instance is absent; unverified same-name sessions are never modified.
 
 ## Workspaces (mux-managed, tmux-first by default)
 

.cursor/rules/hack.mdc

@@ -61,6 +61,8 @@ Prefer `hack` when shell access is available. Use MCP only when shell access is
 - For fixed-port host helpers such as SSM tunnels or local proxies, set `singleton.ports` and usually `onConflict: "adopt"` so Hack reuses a healthy existing listener instead of starting duplicate tunnel stacks.
 - `singleton` is a listener guard, not process ownership transfer; adopted external processes are left running on `hack down`.
 - Inspect lifecycle status via `hack projects --details` and stream via `hack logs <service-or-process>`.
+- Lifecycle session recovery is ownership-proven: Hack adopts healthy token-, definition-, and environment-matched sessions, replaces owned stale sessions, and refuses to kill same-name sessions without deterministic ownership proof.
+- `hack doctor --fix` reaps an orphan lifecycle session only when mux ownership is proven and its Compose instance is absent; unverified same-name sessions are never modified.
 
 ## Host-side env helpers
 

.github/workflows/ci.yml

@@ -56,6 +56,26 @@ jobs:
       - name: Smoke slim runtime mounted-project env flow
         run: bash scripts/portable-container-smoke.sh hack-runtime-ci:slim linux/amd64
 
+  docker-e2e:
+    runs-on: blacksmith-4vcpu-ubuntu-2404
+    permissions:
+      contents: read
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v4
+      - name: Setup Bun
+        uses: oven-sh/setup-bun@v1
+        with:
+          bun-version: "1.3.9"
+      - name: Install tmux
+        run: sudo apt-get update && sudo apt-get install --yes tmux
+      - name: Install dependencies
+        run: bun install
+      - name: Create isolated Hack network
+        run: docker network create --subnet 172.30.0.0/16 hack-dev
+      - name: Run local and Docker E2E
+        run: HACK_E2E_REQUIRE_DOCKER=1 HACK_E2E_REQUIRE_TMUX=1 bun run test:e2e:local:docker
+
   test:
     runs-on: blacksmith-6vcpu-macos-15
     steps:

AGENTS.md

@@ -299,6 +299,8 @@ Lifecycle + startup:
 - For fixed-port host helpers such as SSM tunnels or local proxies, set `singleton.ports` and usually `onConflict: "adopt"` so Hack reuses a healthy existing listener instead of starting duplicate tunnel stacks.
 - `singleton` is a listener guard, not process ownership transfer; adopted external processes are left running on `hack down`.
 - Inspect lifecycle status via `hack projects --details` and stream via `hack logs <service-or-process>`.
+- Lifecycle session recovery is ownership-proven: Hack adopts healthy token-, definition-, and environment-matched sessions, replaces owned stale sessions, and refuses to kill same-name sessions without deterministic ownership proof.
+- `hack doctor --fix` reaps an orphan lifecycle session only when mux ownership is proven and its Compose instance is absent; unverified same-name sessions are never modified.
 
 Workspaces (mux-managed, tmux-first by default):
 - Picker: `hack session` for persistent project workspaces.

CLAUDE.md

@@ -169,6 +169,8 @@ Lifecycle + startup:
 - For fixed-port host helpers such as SSM tunnels or local proxies, set `singleton.ports` and usually `onConflict: "adopt"` so Hack reuses a healthy existing listener instead of starting duplicate tunnel stacks.
 - `singleton` is a listener guard, not process ownership transfer; adopted external processes are left running on `hack down`.
 - Inspect lifecycle status via `hack projects --details` and stream via `hack logs <service-or-process>`.
+- Lifecycle session recovery is ownership-proven: Hack adopts healthy token-, definition-, and environment-matched sessions, replaces owned stale sessions, and refuses to kill same-name sessions without deterministic ownership proof.
+- `hack doctor --fix` reaps an orphan lifecycle session only when mux ownership is proven and its Compose instance is absent; unverified same-name sessions are never modified.
 
 Workspaces (mux-managed, tmux-first by default):
 - Picker: `hack session` for persistent project workspaces.

docs/architecture.md

@@ -125,13 +125,20 @@ stable home and can be torn down on `hack down`.
 Lifecycle state is checkout-local at `.hack/.internal/lifecycle/state.json`. For tmux processes, the
 saved process group belongs to the wrapped user command rather than its mux pane shell. Each compose
 instance has its own entry, so starting a branch instance does not replace the base instance's entry
-in the same checkout. Cleanup reconciles the saved process group with the live process table; if the group
+in the same checkout. A random token mirrored in mux metadata and lifecycle state proves current
+session ownership; deterministic session names never authorize teardown. A versioned SHA-256 digest
+of the effective overlay and sorted lifecycle environment participates in session freshness without
+persisting environment names, keys, or values. Cleanup reconciles the saved process group with the live process table; if the group
 leader exited while members remain, `hack doctor` reports the orphan. `hack down` terminates the
 persisted leaderless group and descendant process groups only while a matching lifecycle mux session
 still proves ownership; without that session, cleanup stays non-destructive because the numeric PGID
 may have been reused. A live group leader without its saved pane PID is likewise not trusted as
 lifecycle ownership.
 
+Doctor also compares ownership-proven lifecycle sessions with live Compose instances. `doctor --fix`
+can remove a proven session only when its corresponding instance is absent and a second liveness and
+ownership check still agrees. Foreign or ambiguous same-name sessions remain non-destructive findings.
+
 For fixed-port helpers such as SSM/database/search tunnels, lifecycle config can also declare a
 `singleton` listener set. This lets Hack reuse an already-running equivalent helper or fail fast on
 partial conflicts instead of launching a competing duplicate supervisor. The intent is to reduce local

docs/lifecycle.md

@@ -179,16 +179,35 @@ Lifecycle output is now surfaced across CLI/runtime views:
 5. Run `docker compose up` (or `up -d` when `--detach`).
 6. Run `lifecycle.up.after` hooks.
 
+Before starting persistent work, Hack reconciles the expected lifecycle session with
+`.hack/.internal/lifecycle/state.json`:
+
+- A session whose mux ownership token, saved definition hash, and live process windows all match is
+  adopted without starting duplicate processes.
+- A token-owned stale session is replaced. Pre-ownership tmux sessions are replaced only when saved
+  state, checkout path, creation time, and window count provide deterministic legacy ownership proof.
+- A same-name session without matching ownership proof is left untouched and `hack up` fails with an
+  actionable collision error.
+
+If the operation later fails in Compose or `up.after`, Hack removes only the lifecycle session created
+by that operation. The same exact-token cleanup runs for supported `SIGINT` and `SIGTERM` paths, including
+signals received while the mux session is still being initialized.
+
 ### `hack down`
 
 1. Run `lifecycle.down.before` hooks.
 2. Run `docker compose down`.
 3. Stop lifecycle processes by killing the lifecycle session.
 4. Run `lifecycle.down.after` hooks.
 
+If `lifecycle.down.before` fails, shutdown is aborted before Compose or lifecycle processes are
+stopped. `hack restart` preserves the same guard semantics during its down phase.
+
 ### `hack restart`
 
 `hack restart` performs the same lifecycle steps as `hack down` followed by `hack up`.
+From the primary checkout, it targets only the base Compose/lifecycle instance. A linked worktree uses
+its isolated derived branch instance, and `--branch <name>` targets only that explicit branch.
 
 ### `--json`
 
@@ -212,10 +231,19 @@ Lifecycle session name:
 Notes:
 - If no mux backend is available, lifecycle process startup fails with an actionable error.
 - Teardown is implemented by killing the lifecycle session; anything running inside that session will be stopped.
+- Current lifecycle sessions carry the same random ownership token in mux metadata and persisted state.
+  Cleanup requires an exact token match; deterministic names alone never authorize session teardown.
+- Healthy-session adoption also matches a stable SHA-256 fingerprint of the effective overlay and
+  sorted lifecycle environment. Environment names, keys, and values are never persisted; changing an
+  overlay or value replaces the owned session so host helpers cannot retain stale configuration.
 - For tmux-backed lifecycle sessions, Hack also persists the pane PID and the wrapped command's actual
   process-group metadata to `.hack/.internal/lifecycle/state.json`, with per-hook/process output logged to
   `.hack/.internal/lifecycle/*.log`. If tmux pane state disappears before teardown, `hack down` still uses that persisted metadata to clean up any live lifecycle process groups instead of leaving orphaned host processes behind.
-- `hack doctor` reports stale lifecycle state when the persisted lifecycle entry no longer has a live mux session and points operators to `hack down` so cleanup and state removal happen through the supported path. `hack doctor --fix` does not tear down lifecycle sessions itself — use `hack down` for that.
+- `hack doctor` reports stale lifecycle state, leaderless process groups, ownership collisions, and
+  ownership-proven sessions with no running Compose instance. Recently updated sessions are treated
+  as possible in-flight startups for five minutes and stay untouched. `hack doctor --fix` reaps only
+  established orphans after rechecking runtime liveness and mux ownership; unverified same-name
+  sessions stay untouched.
 
 ## Tips
 

src/agents/instruction-source.ts

@@ -185,6 +185,8 @@ export const INSTRUCTION_SECTIONS: readonly InstructionSection[] = [
       'For fixed-port host helpers such as SSM tunnels or local proxies, set `singleton.ports` and usually `onConflict: "adopt"` so Hack reuses a healthy existing listener instead of starting duplicate tunnel stacks.',
       "`singleton` is a listener guard, not process ownership transfer; adopted external processes are left running on `hack down`.",
       "Inspect lifecycle status via `hack projects --details` and stream via `hack logs <service-or-process>`.",
+      "Lifecycle session recovery is ownership-proven: Hack adopts healthy token-, definition-, and environment-matched sessions, replaces owned stale sessions, and refuses to kill same-name sessions without deterministic ownership proof.",
+      "`hack doctor --fix` reaps an orphan lifecycle session only when mux ownership is proven and its Compose instance is absent; unverified same-name sessions are never modified.",
     ],
   },
   {

src/commands/doctor.ts

@@ -90,7 +90,10 @@ import {
   repairLegacyComposeEnvFileReferences,
   resolveProjectEnvConfig,
 } from "../lib/project-env-config.ts";
-import { inspectProjectLifecycleHygiene } from "../lib/project-lifecycle-hygiene.ts";
+import {
+  inspectProjectLifecycleHygiene,
+  repairProjectLifecycleSessions,
+} from "../lib/project-lifecycle-hygiene.ts";
 import {
   findMissingRegistryEntries,
   findOrphanRuntimeProjects,
@@ -1844,10 +1847,13 @@ async function checkProjectLifecycleHygiene({
 
   const inspection = await inspectProjectLifecycleHygiene({
     projectDir: ctx.projectDir,
+    projectRoot: ctx.projectRoot,
   });
   if (
     inspection.staleEntries.length === 0 &&
-    inspection.orphanedProcessGroups.length === 0
+    inspection.orphanedProcessGroups.length === 0 &&
+    inspection.orphanedSessions.length === 0 &&
+    inspection.unverifiedSessions.length === 0
   ) {
     return {
       name: "lifecycle hygiene",
@@ -1875,12 +1881,29 @@ async function checkProjectLifecycleHygiene({
           `${orphanedGroups.size} orphaned lifecycle process group${orphanedGroups.size === 1 ? "" : "s"}`,
         ]
       : []),
+    ...(inspection.orphanedSessions.length > 0
+      ? [
+          `${inspection.orphanedSessions.length} owned lifecycle session${inspection.orphanedSessions.length === 1 ? "" : "s"} without a running instance`,
+        ]
+      : []),
+    ...(inspection.unverifiedSessions.length > 0
+      ? [
+          `${inspection.unverifiedSessions.length} same-name lifecycle session collision${inspection.unverifiedSessions.length === 1 ? "" : "s"} with unverified ownership`,
+        ]
+      : []),
   ];
 
+  let nextStep = "run: hack down";
+  if (inspection.orphanedSessions.length > 0) {
+    nextStep = "run: hack doctor --fix";
+  } else if (inspection.unverifiedSessions.length > 0) {
+    nextStep = "manual review required; no session was modified";
+  }
+
   return {
     name: "lifecycle hygiene",
     status: "warn",
-    message: `${details.join("; ")} (run: hack down)`,
+    message: `${details.join("; ")} (${nextStep})`,
   };
 }
 
@@ -2403,6 +2426,7 @@ async function runDoctorFix(opts: {
     return;
   }
 
+  await maybeRepairProjectLifecycleSessions({ startDir: opts.startDir });
   await maybeRepairHackd();
 
   const paths = getGlobalPaths();
@@ -2680,6 +2704,11 @@ export async function buildDoctorRemediationPlanLines(opts: {
   ];
 
   const project = await findProjectContext(opts.startDir);
+  if (project) {
+    steps.push(
+      "Reconcile lifecycle sessions and remove only ownership-proven sessions whose Compose instance is absent."
+    );
+  }
   const trackedGenerated = project
     ? await inspectTrackedGeneratedFiles({
         projectRoot: project.projectRoot,
@@ -2897,6 +2926,37 @@ function describeLegacyRepairStatus(input: {
   return input.didPruneLegacy ? "pruned" : "prune failed";
 }
 
+async function maybeRepairProjectLifecycleSessions(opts: {
+  readonly startDir: string;
+}): Promise<void> {
+  const project = await findProjectContext(opts.startDir);
+  if (!project) {
+    return;
+  }
+  const repaired = await repairProjectLifecycleSessions({
+    projectDir: project.projectDir,
+    projectRoot: project.projectRoot,
+  });
+  if (repaired.repairedSessions.length > 0) {
+    note(
+      [
+        `Removed ${repaired.repairedSessions.length} ownership-proven orphan lifecycle session${repaired.repairedSessions.length === 1 ? "" : "s"}:`,
+        ...repaired.repairedSessions.map((sessionName) => `- ${sessionName}`),
+      ].join("\n"),
+      "lifecycle repair"
+    );
+  }
+  if (repaired.failures.length > 0) {
+    note(
+      [
+        "Lifecycle sessions left unchanged because safe repair could not be re-proven:",
+        ...repaired.failures.map((failure) => `- ${failure}`),
+      ].join("\n"),
+      "lifecycle repair"
+    );
+  }
+}
+
 async function maybeInstallMutagenForDoctorFix(): Promise<void> {
   if (getMutagenPath()) {
     return;