diff --git a/tests/prose/cases/discovery-commits-new-epic/act.md b/tests/prose/cases/discovery-commits-new-epic/act.md
new file mode 100644
index 000000000..01f1758b8
--- /dev/null
+++ b/tests/prose/cases/discovery-commits-new-epic/act.md
@@ -0,0 +1,5 @@
+Execute skills/workflow-start/SKILL.md from the top, as a user opening
+the workflow with nothing on disk yet, and continue into whatever it
+routes to. Stop once discovery has been run for the newly created work
+unit and the prose turns to initialising it — do not initialise, and name
+no topics.
diff --git a/tests/prose/cases/discovery-commits-new-epic/assert.md b/tests/prose/cases/discovery-commits-new-epic/assert.md
new file mode 100644
index 000000000..ea633753e
--- /dev/null
+++ b/tests/prose/cases/discovery-commits-new-epic/assert.md
@@ -0,0 +1,39 @@
+The prose should have taken this path:
+
+1. the empty project is shown as such, and starting an epic routes into
+   discovery with the work unit given as `none`
+2. the work type is put to the user at the gate and committed on their
+   answer — the pre-seed is a hint, and the several interacting strands
+   they describe are what makes it an epic rather than the menu key they
+   pressed
+3. a name is checked against existing work before anything is created,
+   and one engine transaction then creates the work unit, installs the
+   session log staged in the cache, and commits
+4. resume detection is skipped: it answers whether an earlier session was
+   interrupted, and work created moments ago cannot have been
+5. the commit does not conclude the work as it would for a single-phase
+   type — discovery is run for the new work unit, and the conversation
+   carries on into topic territory
+
+Further claims:
+
+- the shaping conversation continues rather than restarting: the session
+  the confirm-trigger opened is the one the work carries forward, and no
+  second session log is authored
+
+EXPECTED WORLD — the walk should have produced, from a project that held
+no work at all:
+
+- an epic work unit named `search-relevance`, in progress, registered in
+  the project manifest alongside its own manifest
+- a description drawn from what the user said about ranking signals,
+  synonyms and measurement — not invented, and not narrowed to one of them
+- exactly one discovery session log, `session-001.md`, naming no seeds and
+  no imports
+- that log recording the map as empty at the start of this first session,
+  which is the epic's shape — not the "not applicable" a single-topic work
+  type would carry
+- the epic marked as having a discovery session open, which is the state a
+  feature never acquires
+- no topics anywhere: no map items, no research or discussion directories.
+  The work unit exists and its topics have not been named yet
diff --git a/tests/prose/cases/discovery-commits-new-epic/case.json b/tests/prose/cases/discovery-commits-new-epic/case.json
new file mode 100644
index 000000000..c5ba46ed1
--- /dev/null
+++ b/tests/prose/cases/discovery-commits-new-epic/case.json
@@ -0,0 +1,41 @@
+{
+  "origin": "discovery — the epic arm of the same commit the feature case walks: the work unit is created, then the work continues into topic territory instead of concluding",
+  "entry": "workflow-start",
+  "world": "claims",
+  "files": [
+    "skills/workflow-start/SKILL.md",
+    "skills/workflow-start/references/empty-state.md",
+    "skills/workflow-start/references/route-to-discovery.md",
+    "skills/workflow-discovery/SKILL.md",
+    "skills/workflow-discovery/references/opener-pattern.md",
+    "skills/workflow-discovery/references/shape-and-confirm.md",
+    "skills/workflow-discovery/references/confirm-trigger.md",
+    "skills/workflow-discovery/references/name-resolution.md",
+    "skills/workflow-discovery/references/template.md"
+  ],
+  "answers": [
+    "e — start a new epic",
+    "Search relevance across the catalogue is poor and it is several problems at once. Ranking barely uses behavioural signals. Synonyms and misspellings are handled by a hand-maintained list nobody trusts. And we have no way to tell whether a change made results better or worse, so every tweak is an argument. I want to take the whole thing on rather than patch one piece.",
+    "y — that's the right shape, set it up",
+    "search-relevance"
+  ],
+  "invariants": {
+    "engine_before_write": true,
+    "calls_include": [
+      "manifest exists search-relevance",
+      "workunit create search-relevance epic",
+      "workflow-discovery/scripts/gateway.cjs search-relevance"
+    ],
+    "calls_in_order": [
+      "manifest exists search-relevance",
+      "workunit create search-relevance epic",
+      "workflow-discovery/scripts/gateway.cjs search-relevance"
+    ],
+    "calls_exclude": [
+      "topic start",
+      "discussion-map",
+      "workflow-discussion-entry",
+      "workflow-research-entry"
+    ]
+  }
+}
diff --git a/tests/prose/cases/discovery-commits-new-epic/fixture-state.cjs b/tests/prose/cases/discovery-commits-new-epic/fixture-state.cjs
new file mode 100644
index 000000000..857c13e35
--- /dev/null
+++ b/tests/prose/cases/discovery-commits-new-epic/fixture-state.cjs
@@ -0,0 +1,12 @@
+'use strict';
+
+// An empty project, exactly as the feature case starts. What the walk
+// makes of it is the whole difference.
+
+const m = require('../../mainlines/feature.cjs');
+
+module.exports = {
+  build(h) {
+    m.init(h);
+  },
+};
diff --git a/tests/prose/cases/discovery-commits-new-epic/fixture.md b/tests/prose/cases/discovery-commits-new-epic/fixture.md
new file mode 100644
index 000000000..94cda6c43
--- /dev/null
+++ b/tests/prose/cases/discovery-commits-new-epic/fixture.md
@@ -0,0 +1,6 @@
+A project with the workflow system installed and nothing else. The
+knowledge store is initialised, migrations have run, and no work has ever
+been started — no work units, no inbox items, no history.
+
+The user has something larger in mind than a single change, and has not
+yet said what.
diff --git a/tests/prose/cases/discovery-commits-new-epic/fixture/.claude/settings.json b/tests/prose/cases/discovery-commits-new-epic/fixture/.claude/settings.json
new file mode 100644
index 000000000..ae019ede5
--- /dev/null
+++ b/tests/prose/cases/discovery-commits-new-epic/fixture/.claude/settings.json
@@ -0,0 +1,9 @@
+{
+  "showClearContextOnPlanAccept": true,
+  "permissions": {
+    "allow": [
+      "Edit(.workflows/**)",
+      "Bash(mv .workflows/:*)"
+    ]
+  }
+}
diff --git a/tests/prose/cases/discovery-commits-new-epic/fixture/.recipe-hash b/tests/prose/cases/discovery-commits-new-epic/fixture/.recipe-hash
new file mode 100644
index 000000000..e7d6a571a
--- /dev/null
+++ b/tests/prose/cases/discovery-commits-new-epic/fixture/.recipe-hash
@@ -0,0 +1 @@
+3e44d2307ed2aa0fafc1f4ac0d8bd5eb585501c158714be81c690d229f6bd522
diff --git a/tests/prose/cases/discovery-commits-new-epic/fixture/.workflows/.state/migrations b/tests/prose/cases/discovery-commits-new-epic/fixture/.workflows/.state/migrations
new file mode 100644
index 000000000..3c7162acc
--- /dev/null
+++ b/tests/prose/cases/discovery-commits-new-epic/fixture/.workflows/.state/migrations
@@ -0,0 +1,52 @@
+001
+002
+003
+004
+005
+006
+007
+008
+009
+010
+011
+012
+013
+014
+015
+016
+017
+018
+019
+020
+021
+022
+023
+024
+025
+026
+027
+028
+029
+030
+031
+032
+033
+034
+035
+036
+037
+038
+039
+040
+041
+042
+043
+044
+045
+046
+047
+048
+049
+050
+051
+052
diff --git a/tests/prose/cases/discovery-commits-new-epic/fixture/.workflows/_gitignore.fixture b/tests/prose/cases/discovery-commits-new-epic/fixture/.workflows/_gitignore.fixture
new file mode 100644
index 000000000..779cbf28a
--- /dev/null
+++ b/tests/prose/cases/discovery-commits-new-epic/fixture/.workflows/_gitignore.fixture
@@ -0,0 +1,2 @@
+.cache/
+.manifest.json.*.tmp