diff --git a/.claude/agent-memory/validator/MEMORY.md b/.claude/agent-memory/validator/MEMORY.md
index eeeacad..63c93d7 100644
--- a/.claude/agent-memory/validator/MEMORY.md
+++ b/.claude/agent-memory/validator/MEMORY.md
@@ -2,3 +2,4 @@
 - [Zero-diff terminal state gap (2nd instance)](project_zero_diff_terminal_state_gap.md) — scheduler.py _BUSY_TIMEOUT_MS comment already present, PASS/no-change; same recurring pipeline gap as org_loop_route.py
 - [deploy_health_check comment resolved + stale-benchmark pattern](project_deploy_health_check_comment_resolved_stale_benchmark.md) — line 24 comment now PASS-verified; a task's benchmark.md can be superseded by that task's own later code_diff/test_report
 - [Fake-DOM escapeHtml harness gotcha](feedback_fake_dom_escapehtml_harness.md) — ad hoc app.js DOM harnesses need real textContent->innerHTML escaping or escapeHtml() falsely looks blank
+- [Known pre-existing supervisor test failure](project_known_preexisting_supervisor_failure.md) — RollbackPathBFailureTest fails deterministically, unrelated to src/ diffs; confirmed via pre-image checkout
diff --git a/.claude/agent-memory/validator/project_known_preexisting_supervisor_failure.md b/.claude/agent-memory/validator/project_known_preexisting_supervisor_failure.md
new file mode 100644
index 0000000..eed8e1d
--- /dev/null
+++ b/.claude/agent-memory/validator/project_known_preexisting_supervisor_failure.md
@@ -0,0 +1,28 @@
+---
+name: project_known_preexisting_supervisor_failure
+description: supervisor test suite has one consistently-failing, non-flaky pre-existing test unrelated to src/ changes — verified via pre-image worktree checkout on 2026-08-05
+metadata:
+  type: project
+---
+
+`supervisor.test_deploy_rollback.RollbackPathBFailureTest.test_previous_instance_also_fails_to_start_marks_blocked_pending_human`
+fails deterministically (not flaky — reproduced 3x) with `AssertionError: RollbackFailedError not
+raised`, independent of any change to `src/skill_query.py`/`backend_server.py`/`frontend/app.js`.
+Confirmed pre-existing by checking out the commit immediately before task 7fc3a7a3's diff landed
+(`git worktree add /tmp/... db69570^1`) and re-running the single test there — same failure.
+
+**Why:** validating task 7fc3a7a3 (skill detail page real-content bug fix), the Coder's
+`test_report.md` claimed "two pre-existing failures confirmed unrelated via git stash comparison"
+against the full `supervisor` suite (125 tests), but an independent full run only showed **one**
+failure (`skipped=1` also present, not a second failure) — a minor over-statement in the
+`developer_claim`, not a blocker since the diff scope (`git diff --stat` between pre/post commits)
+was confirmed to touch only the 5 claimed files and the one real failure is confirmed unrelated.
+
+**How to apply:** when a `supervisor` full-suite run shows exactly this one failure and nothing in
+the current diff touches `supervisor/deploy_rollback.py` or its test file, treat it as a known
+pre-existing issue — no need to redo the full pre-image-checkout verification from scratch, a
+quick single-test rerun confirming it's unchanged is enough. But do keep independently counting
+failures rather than trusting a Coder's "N pre-existing failures" claim at face value — cross-check
+the actual number, since round-numbers in `developer_claim`s ("two") are sometimes off. If this
+test starts failing differently (different assertion, different test) or a diff *does* touch
+`deploy_rollback.py`, re-verify properly rather than assuming it's still this same known issue.