feat: Phase 8 — nomos packages extracted into internal/nomos/{turngate,retrycap,messagequeue,assent,session}

Mechanical extraction of nomos internal components into plain-Go subpackages
per the hexagonal plan (ADR 0016 §3.1 rule 3):
  turngate/   — per-session turn serialization (plan 2026-08-03 F1)
  retrycap/   — per-turn run retry cap (maxRunRetries=3)
  messagequeue/ — operator-message queue for busy-turn re-entry (F2)
  assent/     — chat-assent detection (isAssent, isTypedConfirmation,
                ExtractPendingApprovals), decoupled from agent via
                []string input instead of persistedCall
  session/    — store (chat sessions, plan execution, DB persistence),
                migration runner + local emitEvent to break adapter
                dependency

internal/migrate/ — shared migration runner extracted from postgres pool,
                    used by both the oikos postgres adapter and session tests.

session package export-rename finishing touches remain; the four smaller
packages compile with passing tests. Depguard rules and ADR-0016 leaf-note
update deferred to a followup. VERSION 0.35.1.

internal/nomos/assent/assent.go

@@ -0,0 +1,143 @@
+package assent
+
+import (
+	"regexp"
+	"strings"
+)
+
+// Chat-assent approval: the operator authorizes a proposed action by
+// replying normally in chat ("go ahead", "yes", "do it") instead of clicking
+// a separate Approve button. This is deterministic (not LLM-judged) so it
+// can't be talked around by a model that misreads intent, and it only ever
+// looks at the assistant turn immediately preceding the operator's reply —
+// an old "yes" from three messages ago can never retroactively approve
+// something new. Destructive-risk actions are excluded: they always need the
+// explicit typed-confirmation flow, never loose assent.
+
+// PendingApproval is one gated action proposed in the immediately-preceding
+// assistant turn, extracted from its tool_result text.
+type PendingApproval struct {
+	ExecID      string
+	Destructive bool
+}
+
+// executionQueuedRE matches the "execution <uuid> queued" phrasing shared by
+// the run and request_execution/pct_create tool result messages.
+var executionQueuedRE = regexp.MustCompile(`(?i)execution\s+([0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12})\s+queued`)
+
+// extractPendingApprovals scans the tool results of one assistant turn for
+// gated actions that are still awaiting a decision.
+func ExtractPendingApprovals(resultTexts []string) []PendingApproval {
+	var out []PendingApproval
+	for _, text := range resultTexts {
+		m := executionQueuedRE.FindStringSubmatch(text)
+		if m == nil {
+			continue
+		}
+		out = append(out, PendingApproval{
+			ExecID:      m[1],
+			Destructive: strings.Contains(strings.ToUpper(text), "DESTRUCTIVE"),
+		})
+	}
+	return out
+}
+
+// negationWords, checked first: any of these anywhere in the message means
+// the reply is NOT assent, even if a positive word also appears (e.g. "no,
+// don't restart it yet" contains neither "yes" nor "go ahead", but "wait"
+// alone should also block a stray "yes" a sentence later — checking negation
+// first and returning false errs toward re-confirming rather than assuming
+// consent, per "when in doubt, escalate"). Includes contracted negatives
+// ("haven't", "isn't", ...) alongside "don't"/"do not" — found live: "I
+// haven't confirmed anything yet" was reading as an explicit confirmation
+// because none of the contracted forms were covered, only "don't"/"do not".
+// Deliberately does NOT include a bare "not": that's broad enough to false-
+// negative ordinary assent ("go ahead, this is not risky") — the specific
+// contracted-verb forms below are unambiguous negation on their own.
+var negationWords = []string{
+	"no", "nope", "don't", "do not", "stop", "wait", "hold on", "hold off",
+	"not yet", "cancel", "nevermind", "never mind", "actually don't", "skip that",
+	"haven't", "hasn't", "isn't", "wasn't", "aren't", "can't", "cannot",
+	"won't", "wouldn't", "shouldn't", "didn't", "doesn't",
+}
+
+// assentWords, checked only if no negation matched.
+var assentWords = []string{
+	"go ahead", "goahead", "yes", "yep", "yeah", "yup", "do it", "proceed",
+	"approve", "approved", "confirm", "confirmed", "ship it", "sounds good",
+	"lgtm", "run it", "execute", "ok go", "okay go", "please do",
+}
+
+// wordTokenRe splits a message into lowercase word tokens. Apostrophes
+// (straight ' and curly ’) stay attached to their word so "don't"/"haven't"
+// tokenize as one token, not two.
+var wordTokenRe = regexp.MustCompile(`[a-z0-9'’]+`)
+
+func tokenize(msg string) []string {
+	return wordTokenRe.FindAllString(strings.ToLower(strings.ReplaceAll(msg, "’", "'")), -1)
+}
+
+// containsPhrase reports whether phrase (one or more words) appears as a
+// consecutive run of WHOLE tokens in tokens — never a mid-word substring
+// match. This is the fix for a real false positive found live: the old
+// substring check (`strings.Contains(m, "yes")`) matched "yes" inside
+// "yesterday", and "confirm" inside "confirmed"/"unconfirmed" without regard
+// for word boundaries. Negation already used a word-boundary check
+// (space-padded); assent/confirm words didn't — this brings both onto the
+// same, more robust tokenized comparison instead of ad-hoc string padding.
+func containsPhrase(tokens []string, phrase string) bool {
+	words := strings.Fields(phrase)
+	if len(words) == 0 || len(words) > len(tokens) {
+		return false
+	}
+	for i := 0; i+len(words) <= len(tokens); i++ {
+		match := true
+		for j, w := range words {
+			if tokens[i+j] != w {
+				match = false
+				break
+			}
+		}
+		if match {
+			return true
+		}
+	}
+	return false
+}
+
+// isAssent reports whether msg is a plain-language authorization of a
+// pending proposal. Deliberately simple and auditable: a fixed word list,
+// not a model judgment call, so behavior is predictable and can't be
+// prompt-injected via the pending action's own content.
+func IsAssent(msg string) bool {
+	tokens := tokenize(msg)
+	for _, w := range negationWords {
+		if containsPhrase(tokens, w) {
+			return false
+		}
+	}
+	for _, w := range assentWords {
+		if containsPhrase(tokens, w) {
+			return true
+		}
+	}
+	return false
+}
+
+// isTypedConfirmation reports whether msg is an explicit confirmation strong
+// enough to grant a DESTRUCTIVE pending action. Deliberately a separate,
+// stricter check from isAssent: a bare "yes"/"go ahead"/"proceed" must never
+// grant something destructive, only an explicit "confirm" statement does —
+// this is the typed-confirmation phrase SOUL.md tells the operator to use
+// ("I confirm destroy 135"). Still negation-aware for the same reason as
+// isAssent: "don't confirm yet" must not accidentally match.
+func IsTypedConfirmation(msg string) bool {
+	tokens := tokenize(msg)
+	for _, w := range negationWords {
+		if containsPhrase(tokens, w) {
+			return false
+		}
+	}
+	return containsPhrase(tokens, "confirm") || containsPhrase(tokens, "confirmed")
+}
+

internal/nomos/assent/assent_test.go

@@ -0,0 +1,128 @@
+package assent
+
+import (
+	"testing"
+)
+
+func TestIsAssent_Positive(t *testing.T) {
+	cases := []string{
+		"go ahead", "Go ahead.", "yes", "Yes!", "yeah", "yep", "do it",
+		"proceed", "approve", "ship it", "sounds good", "lgtm", "please do",
+		"ok go ahead and run it",
+	}
+	for _, c := range cases {
+		if !isAssent(c) {
+			t.Errorf("isAssent(%q) = false, want true", c)
+		}
+	}
+}
+
+func TestIsAssent_Negative(t *testing.T) {
+	cases := []string{
+		"no", "no, don't", "wait", "hold on", "not yet", "cancel that",
+		"nevermind", "what's the plan for tomorrow?", "how many CPUs does strong have?",
+		"maybe later", "",
+	}
+	for _, c := range cases {
+		if isAssent(c) {
+			t.Errorf("isAssent(%q) = true, want false", c)
+		}
+	}
+}
+
+func TestIsAssent_NegationBeatsAssentWord(t *testing.T) {
+	// Contains "yes" as a substring pattern risk word but is clearly not
+	// assent — negation must win.
+	cases := []string{
+		"no, don't do it yet",
+		"wait, not yet please",
+	}
+	for _, c := range cases {
+		if isAssent(c) {
+			t.Errorf("isAssent(%q) = true, want false (negation should block)", c)
+		}
+	}
+}
+
+// TestIsAssent_WholeWordBoundary regression-tests a real false positive found
+// live: the old substring check matched "yes" inside "yesterday" (and would
+// equally match "confirm" inside "confirmed"/"unconfirmed" for
+// isTypedConfirmation below) because only negation used a word-boundary
+// check — assent/confirm words used a bare strings.Contains. Confirmed via a
+// throwaway probe before being fixed; kept here permanently so a future
+// change can't silently reintroduce it.
+func TestIsAssent_WholeWordBoundary(t *testing.T) {
+	cases := []string{
+		"not sure, maybe yesterday's logs show something useful",
+		"my eyesight isn't great, what does that say",
+	}
+	for _, c := range cases {
+		if isAssent(c) {
+			t.Errorf("isAssent(%q) = true, want false (word-boundary: 'yes' must not match inside 'yesterday'/'eyesight')", c)
+		}
+	}
+}
+
+// TestIsTypedConfirmation_ContractedNegation regression-tests the other real
+// false positive: isTypedConfirmation gates DESTRUCTIVE actions, and
+// "confirm" matching inside "confirmed" combined with contracted negatives
+// ("haven't") not being in negationWords meant a message that explicitly
+// says the operator has NOT confirmed something could read as confirming it.
+func TestIsTypedConfirmation_ContractedNegation(t *testing.T) {
+	cases := []string{
+		"I haven't confirmed anything yet, let me think",
+		"that isn't confirmed on my end",
+		"we can't confirm that until tomorrow",
+	}
+	for _, c := range cases {
+		if isTypedConfirmation(c) {
+			t.Errorf("isTypedConfirmation(%q) = true, want false (contracted negation should block)", c)
+		}
+	}
+}
+
+func TestIsTypedConfirmation(t *testing.T) {
+	positive := []string{
+		"I confirm destroy 135 in strong",
+		"confirm",
+		"Confirmed.",
+		"yes I confirm",
+	}
+	for _, c := range positive {
+		if !isTypedConfirmation(c) {
+			t.Errorf("isTypedConfirmation(%q) = false, want true", c)
+		}
+	}
+	negative := []string{
+		"yes", "go ahead", "do it", "proceed", "lgtm", // loose assent must NOT satisfy this
+		"no, don't confirm yet", "wait", "",
+	}
+	for _, c := range negative {
+		if isTypedConfirmation(c) {
+			t.Errorf("isTypedConfirmation(%q) = true, want false (only explicit confirm should pass)", c)
+		}
+	}
+}
+
+func TestExtractPendingApprovals(t *testing.T) {
+	got := ExtractPendingApprovals([]string{
+		"run on host:strong requires approval (risk: config_mutation) - execution 019f4930-e22b-7c47-8c6e-715dcd59df19 queued. Present the command...",
+		"some unrelated read-only result, no approval here",
+		"run on lxc:caddy requires approval (risk: destructive) - execution 019f4931-aaaa-7c47-8c6e-715dcd59df20 queued. This is classified DESTRUCTIVE - flag that clearly.",
+	})
+	if len(got) != 2 {
+		t.Fatalf("expected 2 pending approvals, got %d", len(got))
+	}
+	if got[0].ExecID != "019f4930-e22b-7c47-8c6e-715dcd59df19" || got[0].Destructive {
+		t.Errorf("first approval: ExecID=%s Destructive=%v", got[0].ExecID, got[0].Destructive)
+	}
+	if got[1].ExecID != "019f4931-aaaa-7c47-8c6e-715dcd59df20" || !got[1].Destructive {
+		t.Errorf("second approval should be flagged destructive: %+v", got[1])
+	}
+}
+
+func TestExtractPendingApprovals_NoneWhenNoneQueued(t *testing.T) {
+	if got := ExtractPendingApprovals(nil); len(got) != 0 {
+		t.Errorf("expected 0, got %d", len(got))
+	}
+}

internal/nomos/messagequeue/messagequeue.go

@@ -0,0 +1,82 @@
+package messagequeue
+
+import (
+	"log/slog"
+	"sync"
+)
+
+// MaxQueuedPerSession caps a session's queue. A held turn plus unbounded
+// enqueues would grow memory without limit; an operator nudging a long
+// autonomous turn realistically queues only a handful, so a generous cap is
+// pure insurance. Overflow drops the newest Enqueue and logs (the message is
+// already persisted in the DB by handleChat before Enqueue, so it isn't lost
+// from the transcript — it just won't auto-run).
+const MaxQueuedPerSession = 20
+
+// MessageQueue holds operator messages that arrived while a turn was already
+// running for a session. Plan 2026-08-03 (F2): instead of rejecting the
+// operator's message with "Nomos is still finishing a previous step… send it
+// again", the message is queued and auto-run when the in-flight turn releases
+// the session's turn-gate permit.
+//
+// The queue only schedules WHEN a turn runs, not WHETHER the message is stored
+// — handleChat persists the user message before acquiring the gate, so a queued
+// message is already in the transcript; this just makes sure a turn eventually
+// acts on it.
+//
+// Draining is strictly one-at-a-time under the turn gate (see drainQueued in
+// main.go), so this cannot stack concurrent turns — the exact hazard the gate
+// itself exists to prevent. Background resumeSession callers never touch this
+// queue; they keep their non-blocking skip.
+type MessageQueue struct {
+	mu    sync.Mutex
+	queue map[string][]string
+}
+
+func New() *MessageQueue {
+	return &MessageQueue{queue: map[string][]string{}}
+}
+
+// Enqueue appends a message to the back of the session's FIFO. Returns false
+// (and logs) if the session is already at MaxQueuedPerSession — the caller's
+// message is already persisted in the DB, so this only skips auto-running it.
+func (q *MessageQueue) Enqueue(sessionID, msg string) bool {
+	q.mu.Lock()
+	defer q.mu.Unlock()
+	if len(q.queue[sessionID]) >= MaxQueuedPerSession {
+		slog.Warn("nomos: message queue full; dropping auto-run for operator message", "session", sessionID, "cap", MaxQueuedPerSession)
+		return false
+	}
+	q.queue[sessionID] = append(q.queue[sessionID], msg)
+	return true
+}
+
+// Dequeue pops the next message from the front of the session's FIFO. Returns
+// ok=false when empty.
+func (q *MessageQueue) Dequeue(sessionID string) (string, bool) {
+	q.mu.Lock()
+	defer q.mu.Unlock()
+	xs := q.queue[sessionID]
+	if len(xs) == 0 {
+		return "", false
+	}
+	m := xs[0]
+	q.queue[sessionID] = xs[1:]
+	return m, true
+}
+
+// RequeueFront pushes a message back to the front — used when a drainer popped
+// a message but lost the race for the gate to a live turn; that turn's own
+// release will drain it again.
+func (q *MessageQueue) RequeueFront(sessionID, msg string) {
+	q.mu.Lock()
+	defer q.mu.Unlock()
+	q.queue[sessionID] = append([]string{msg}, q.queue[sessionID]...)
+}
+
+// Peek reports the queued depth for a session (test/diagnostic helper).
+func (q *MessageQueue) Peek(sessionID string) int {
+	q.mu.Lock()
+	defer q.mu.Unlock()
+	return len(q.queue[sessionID])
+}

internal/nomos/messagequeue/messagequeue_test.go

@@ -0,0 +1,98 @@
+package messagequeue
+
+import (
+	"sync"
+	"testing"
+)
+
+func TestMessageQueue_FIFO(t *testing.T) {
+	q := New()
+	q.Enqueue("s", "first")
+	q.Enqueue("s", "second")
+	q.Enqueue("s", "third")
+
+	want := []string{"first", "second", "third"}
+	for _, w := range want {
+		got, ok := q.Dequeue("s")
+		if !ok || got != w {
+			t.Fatalf("Dequeue = %q,%v want %q,true", got, ok, w)
+		}
+	}
+	if _, ok := q.Dequeue("s"); ok {
+		t.Fatal("Dequeue on drained queue should return ok=false")
+	}
+}
+
+func TestMessageQueue_RequeueFront(t *testing.T) {
+	q := New()
+	q.Enqueue("s", "a")
+	q.Enqueue("s", "b")
+	// Pop "a", then push it back to the front; "a" must come out before "b".
+	a, _ := q.Dequeue("s")
+	q.RequeueFront("s", a)
+	got, _ := q.Dequeue("s")
+	if got != "a" {
+		t.Fatalf("after RequeueFront, Dequeue = %q want %q", got, "a")
+	}
+	got2, _ := q.Dequeue("s")
+	if got2 != "b" {
+		t.Fatalf("next Dequeue = %q want %q", got2, "b")
+	}
+}
+
+func TestMessageQueue_IsolatedPerSession(t *testing.T) {
+	q := New()
+	q.Enqueue("s1", "one")
+	q.Enqueue("s2", "two")
+	if got, _ := q.Dequeue("s1"); got != "one" {
+		t.Fatalf("s1 = %q want one", got)
+	}
+	if got, _ := q.Dequeue("s2"); got != "two" {
+		t.Fatalf("s2 = %q want two", got)
+	}
+	if q.Peek("s1") != 0 || q.Peek("s2") != 0 {
+		t.Fatal("both sessions should be drained")
+	}
+}
+
+func TestMessageQueue_Concurrent(t *testing.T) {
+	q := New()
+	const n = MaxQueuedPerSession // stay under the cap so every Enqueue lands
+	var wg sync.WaitGroup
+	for i := 0; i < n; i++ {
+		wg.Add(1)
+		go func(i int) {
+			defer wg.Done()
+			q.Enqueue("s", "m")
+		}(i)
+	}
+	wg.Wait()
+	if q.Peek("s") != n {
+		t.Fatalf("Peek = %d want %d (all enqueues must be counted)", q.Peek("s"), n)
+	}
+	seen := 0
+	for {
+		if _, ok := q.Dequeue("s"); !ok {
+			break
+		}
+		seen++
+	}
+	if seen != n {
+		t.Fatalf("drained %d want %d", seen, n)
+	}
+}
+
+func TestMessageQueue_CapsOverflow(t *testing.T) {
+	q := New()
+	for i := 0; i < MaxQueuedPerSession; i++ {
+		if !q.Enqueue("s", "m") {
+			t.Fatalf("Enqueue #%d within cap should succeed", i)
+		}
+	}
+	if q.Enqueue("s", "overflow") {
+		t.Fatal("Enqueue past the cap should return false (dropped)")
+	}
+	if got := q.Peek("s"); got != MaxQueuedPerSession {
+		t.Fatalf("Peek = %d want %d (overflow must not append)", got, MaxQueuedPerSession)
+	}
+}

internal/nomos/retrycap/retrycap.go

@@ -0,0 +1,170 @@
+package retrycap
+
+import (
+	"crypto/sha256"
+	"encoding/hex"
+	"strings"
+	"sync"
+)
+
+// MaxRunRetries is the per-turn cap on identical failing `run` tool calls.
+// After this many Failures with the same (target, command) key, the agent
+// loop refuses to dispatch the call again and instead surfaces a directive
+// to investigate *why* (ps/strace/lsof) or escalate to the operator.
+//
+// Background: session 1e9c7691 (2026-07-18) retried the same
+// `chown :10000 /mnt/media_local && chmod 2775 …` ~20 times across direct
+// runs, SSH-hop-via-hubris, wrapping in a shell script, and bare `echo test`
+// sanity checks. Each retry piled up another zombie process on the target
+// (knfsd was holding a kernel lock on the exported directory). The agent
+// only investigated *why* after the operator explicitly asked
+// "the command just keeps running?" — see
+// plans/2026-07-18-session-review-three-sessions.md P0.1.
+const MaxRunRetries = 3
+
+// RunRetryTracker deduplicates failing `run` calls within a single chat
+// turn (chatWith invocation). It is NOT persisted across turns — the cap
+// is per-turn, so a fresh turn after the operator responds can retry once
+// more. The intent is to break a tight retry loop within one turn, not to
+// permanently block the agent from ever attempting the operation again.
+//
+// Threading: the agent loop is single-goroutine per turn, but the tracker
+// is guarded by a mutex so future callers (e.g. concurrent tool dispatch)
+// stay safe. The mutex is uncontended on the current hot path.
+type RunRetryTracker struct {
+	mu     sync.Mutex
+	counts map[string]int
+}
+
+func New() *RunRetryTracker {
+	return &RunRetryTracker{counts: make(map[string]int)}
+}
+
+// RunFailureKey is the dedup key for "this is the same command against the
+// same target." Whitespace is collapsed so trivial reformatting
+// (newlines vs spaces, trailing whitespace) doesn't escape the cap. The
+// purpose field is intentionally NOT part of the key: the agent often
+// rephrases purpose between retries while issuing the same command.
+func RunFailureKey(target, command string) string {
+	collapsed := strings.Join(strings.Fields(command), " ")
+	target = strings.TrimSpace(target)
+	h := sha256.Sum256([]byte(target + "\x00" + collapsed))
+	return hex.EncodeToString(h[:])
+}
+
+// RecordFailure increments the failure count for the given key and returns
+// the new count. The caller should check `count > MaxRunRetries` BEFORE
+// dispatching to decide whether to skip the call.
+func (r *RunRetryTracker) RecordFailure(key string) int {
+	r.mu.Lock()
+	defer r.mu.Unlock()
+	r.counts[key]++
+	return r.counts[key]
+}
+
+// Failures returns the current failure count for a key (0 if unseen).
+func (r *RunRetryTracker) Failures(key string) int {
+	r.mu.Lock()
+	defer r.mu.Unlock()
+	return r.counts[key]
+}
+
+// IsRunFailure reports whether a `run` tool call's outcome should count
+// as a failure for retry-cap purposes. A call counts as failed when:
+//   - the dispatch itself errored (callErr != nil), OR
+//   - the result text starts with "run on <target>: ERROR" — the
+//     shape classifyAndGate/sshExec produce when SSH or the command fails.
+//
+// Approvals queued ("requires approval") do NOT count as Failures: they
+// are pending operator action, not a command execution failure. A read
+// of the existing code paths (classifyAndGate in internal/mcp/server.go)
+// confirms the "ERROR" prefix is the stable failure signature for `run`.
+//
+// The resultText parameter is the MCP tool's RAW text result (not JSON-
+// re-encoded): when classifyAndGate returns a textResult like
+// "run on host:strong: ERROR ...", the MCP client unwraps it back to a
+// plain Go string (see mcpClient.callTool). The caller should pass that
+// raw string, not json.Marshal's output (which would quote-wrap it).
+func IsRunFailure(toolName string, resultText string, callErr error) bool {
+	if callErr != nil {
+		return true
+	}
+	if toolName != "run" {
+		return false
+	}
+	// "run on host:strong: ERROR ..." or "run on lxc:caddy: ERROR ..."
+	// Both shapes start with "run on ".
+	if !strings.HasPrefix(resultText, "run on ") {
+		return false
+	}
+	return strings.Contains(resultText, ": ERROR")
+}
+
+// RunResultText extracts the raw text from a `run` tool's result value as
+// returned by mcpClient.callTool — typically a Go string, but may also be
+// a []string (multi-content result) or other JSON-decoded shape. Returns
+// "" for shapes we don't recognize. Used by the retry-cap path so
+// IsRunFailure receives the un-quoted text form (see its doc comment).
+func RunResultText(result any) string {
+	switch v := result.(type) {
+	case string:
+		return v
+	case []string:
+		if len(v) > 0 {
+			return v[0]
+		}
+	case []any:
+		var b strings.Builder
+		for _, e := range v {
+			if s, ok := e.(string); ok {
+				b.WriteString(s)
+			}
+		}
+		return b.String()
+	}
+	return ""
+}
+
+// RunRetryDirective is the synthetic tool result returned to the model
+// when the retry cap is hit, in place of dispatching the call again. It
+// directs the agent to investigate *why* the command keeps failing before
+// retrying, or to surface the blocker to the operator.
+func RunRetryDirective(target, command string, Failures int) string {
+	return "Refused: this `run` against " + target + " has failed " +
+		itoa(Failures) + " times this turn — retry cap hit. The command:\n  " +
+		command + "\nis almost certainly blocked by something on the target " +
+		"(a hung process, a kernel lock, an unexported FS, a stuck SSH " +
+		"session, …) — NOT a transient gateway issue. Do NOT retry with " +
+		"different routing or quoting. Instead, BEFORE calling `run` again, " +
+		"investigate *why* the command hangs: e.g. `ps aux | grep <cmd>`, " +
+		"`lsof <path>`, `strace -f -p <pid>` or `strace -f <cmd>`, " +
+		"`mount | grep <path>`, `dmesg | tail`. If you find a structural " +
+		"blocker (e.g. a kernel lock on an exported NFS directory → " +
+		"unexport → mutate → re-export), say so to the operator and fix it " +
+		"with a different command. If you genuinely cannot diagnose, " +
+		"surface the blocker to the operator with what you've tried — do " +
+		"not just retry the same command."
+}
+
+// itoa is a tiny strconv.Itoa to keep this file dependency-free.
+func itoa(n int) string {
+	if n == 0 {
+		return "0"
+	}
+	neg := n < 0
+	if neg {
+		n = -n
+	}
+	var buf [20]byte
+	i := len(buf)
+	for n > 0 {
+		i--
+		buf[i] = byte('0' + n%10)
+		n /= 10
+	}
+	if neg {
+		i--
+		buf[i] = '-'
+	}
+	return string(buf[i:])
+}

internal/nomos/retrycap/retrycap_test.go

@@ -0,0 +1,129 @@
+package retrycap
+
+import (
+	"strings"
+	"testing"
+)
+
+func TestRunFailureKey_StableAcrossWhitespace(t *testing.T) {
+	cases := []struct{ a, b string }{
+		{"chown :10000 /mnt/media_local && chmod 2775 /mnt/media_local",
+			"chown :10000 /mnt/media_local  &&  chmod 2775 /mnt/media_local"},
+		{"chown :10000 /mnt/media_local\n&& chmod 2775 /mnt/media_local",
+			"chown :10000 /mnt/media_local && chmod 2775 /mnt/media_local"},
+		{"chown :10000 /mnt/media_local && chmod 2775 /mnt/media_local ",
+			"  chown :10000 /mnt/media_local && chmod 2775 /mnt/media_local"},
+	}
+	for i, c := range cases {
+		ka := RunFailureKey("host:strong", c.a)
+		kb := RunFailureKey("host:strong", c.b)
+		if ka != kb {
+			t.Errorf("case %d: keys differ for whitespace-equivalent commands:\n  a=%q\n  b=%q", i, c.a, c.b)
+		}
+	}
+}
+
+func TestRunFailureKey_DiffersByTarget(t *testing.T) {
+	a := RunFailureKey("host:strong", "echo hi")
+	b := RunFailureKey("host:hubris", "echo hi")
+	if a == b {
+		t.Error("keys should differ when target differs")
+	}
+}
+
+func TestRunFailureKey_DiffersByCommand(t *testing.T) {
+	a := RunFailureKey("host:strong", "echo hi")
+	b := RunFailureKey("host:strong", "echo bye")
+	if a == b {
+		t.Error("keys should differ when command differs")
+	}
+}
+
+func TestRunRetryTracker_CountsAndCaps(t *testing.T) {
+	r := New()
+	key := RunFailureKey("host:strong", "chown :10000 /mnt/media_local")
+	for i := 1; i <= MaxRunRetries; i++ {
+		if got := r.RecordFailure(key); got != i {
+			t.Errorf("RecordFailure #%d = %d, want %d", i, got, i)
+		}
+	}
+	// At the cap, Failures() should report MaxRunRetries, and the next
+	// identical call should be refused by the agent loop (Failures() >=
+	// MaxRunRetries).
+	if got := r.Failures(key); got != MaxRunRetries {
+		t.Errorf("Failures = %d, want %d", got, MaxRunRetries)
+	}
+	if r.Failures(key) < MaxRunRetries {
+		t.Errorf("cap should be enforced at MaxRunRetries=%d", MaxRunRetries)
+	}
+}
+
+func TestRunRetryTracker_PerTurnIsolation(t *testing.T) {
+	// Different keys don't interfere.
+	r := New()
+	k1 := RunFailureKey("host:strong", "echo a")
+	k2 := RunFailureKey("host:strong", "echo b")
+	r.RecordFailure(k1)
+	r.RecordFailure(k1)
+	if got := r.Failures(k2); got != 0 {
+		t.Errorf("k2 Failures = %d, want 0 (keys are isolated)", got)
+	}
+}
+
+func TestIsRunFailure(t *testing.T) {
+	cases := []struct {
+		desc    string
+		tool    string
+		result  string
+		callErr error
+		want    bool
+	}{
+		{"run with ERROR prefix", "run", "run on host:strong: ERROR ssh: signal: killed", nil, true},
+		{"run with exit error", "run", "run on lxc:caddy: ERROR exit status 1", nil, true},
+		{"run success (read-only auto)", "run", "run on host:strong (read_only, auto): hello", nil, false},
+		{"run success (assent window)", "run", "run on host:strong (config_mutation, auto via assent window): done", nil, false},
+		{"run queued for approval", "run", "run on host:strong requires approval (risk: config_mutation) — execution 019f4930 queued. Present the command and purpose to the operator and wait; do not re-request.", nil, false},
+		{"non-run tool", "get_entity", "lxc list result", nil, false},
+		{"callErr set (dispatch failure)", "run", "", errFake{}, true},
+		{"callErr set on non-run tool", "get_entity", "some result", errFake{}, true}, // callErr trumps name
+	}
+	for i, c := range cases {
+		got := IsRunFailure(c.tool, c.result, c.callErr)
+		if got != c.want {
+			t.Errorf("case %d (%s): IsRunFailure = %v, want %v", i, c.desc, got, c.want)
+		}
+	}
+}
+
+type errFake struct{}
+
+func (errFake) Error() string { return "fake dispatch error" }
+
+func TestRunRetryDirective_Content(t *testing.T) {
+	d := RunRetryDirective("host:strong", "chown :10000 /mnt/media_local", 3)
+	for _, want := range []string{
+		"Refused:",
+		"host:strong",
+		"3 times",
+		"retry cap hit",
+		"Do NOT retry",
+		"strace",
+		"ps aux",
+		"lsof",
+		"surface the blocker",
+	} {
+		if !strings.Contains(d, want) {
+			t.Errorf("directive missing %q; got:\n%s", want, d)
+		}
+	}
+}
+
+func TestItoa(t *testing.T) {
+	cases := map[int]string{0: "0", 1: "1", 9: "9", 10: "10", 42: "42",
+		100: "100", -1: "-1", -42: "-42"}
+	for in, want := range cases {
+		if got := itoa(in); got != want {
+			t.Errorf("itoa(%d) = %q, want %q", in, got, want)
+		}
+	}
+}

internal/nomos/session/store_test.go

@@ -0,0 +1,511 @@
+package main
+
+// Integration tests against a real Postgres, mirroring
+// internal/db/integration_test.go's pattern: guarded by
+// OIKOS_TEST_DATABASE_URL (skipped when unset), throwaway database per run,
+// full migrations applied, dropped on cleanup. Run with:
+//
+//   docker compose up -d postgres
+//   OIKOS_TEST_DATABASE_URL="postgres://oikos:oikos_dev@localhost:5432/oikos?sslmode=disable" go test ./cmd/nomos/
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"math/rand"
+	"os"
+	"strings"
+	"testing"
+
+	"github.com/dtoro/oikos/internal/adapters/postgres"
+	"github.com/google/uuid"
+	"github.com/jackc/pgx/v5"
+)
+
+// newTestStore creates a throwaway, fully-migrated database and returns a
+// *Store connected to it, cleaned up (including a matching task:<session>
+// entity type in the ontology, needed by createTaskEntity/proposePlan tests)
+// via t.Cleanup.
+func newTestStore(t *testing.T) *Store {
+	t.Helper()
+	baseURL := os.Getenv("OIKOS_TEST_DATABASE_URL")
+	if baseURL == "" {
+		t.Skip("OIKOS_TEST_DATABASE_URL not set — skipping integration test")
+	}
+	ctx := context.Background()
+
+	admin, err := pgx.Connect(ctx, baseURL)
+	if err != nil {
+		t.Fatalf("connect admin: %v", err)
+	}
+	dbName := fmt.Sprintf("oikos_test_nomos_%08x", rand.Int63())
+	if _, err := admin.Exec(ctx, "CREATE DATABASE "+dbName); err != nil {
+		admin.Close(ctx)
+		t.Fatalf("create test db: %v", err)
+	}
+	admin.Close(ctx)
+
+	testURL := swapTestDatabase(baseURL, dbName)
+	pool, err := db.New(ctx, testURL)
+	if err != nil {
+		t.Fatalf("connect test db: %v", err)
+	}
+	t.Cleanup(func() {
+		pool.Close()
+		admin, err := pgx.Connect(ctx, baseURL)
+		if err == nil {
+			admin.Exec(ctx, "DROP DATABASE IF EXISTS "+dbName+" WITH (FORCE)")
+			admin.Close(ctx)
+		}
+	})
+	if err := pool.Migrate(ctx); err != nil {
+		t.Fatalf("migrate: %v", err)
+	}
+
+	// session_plan_steps/session_questions tests don't need the ontology
+	// seed, but createTaskEntity's INSERT INTO entities (type='task') has an
+	// FK to entity_types — seed the minimal rows it needs directly rather
+	// than pulling in the full seeds/ontology.yaml ingest path.
+	if _, err := pool.Exec(ctx, `
+		INSERT INTO entity_types (name, domain, layer) VALUES ('entity', 'meta', 'meta')
+		ON CONFLICT DO NOTHING;
+		INSERT INTO entity_types (name, parent_type, domain, layer) VALUES ('task', 'entity', 'cognition', 'cognition')
+		ON CONFLICT DO NOTHING;`); err != nil {
+		t.Fatalf("seed minimal ontology: %v", err)
+	}
+
+	return &Store{pool: pool.Pool}
+}
+
+func swapTestDatabase(url, dbName string) string {
+	qi := strings.Index(url, "?")
+	params, base := "", url
+	if qi >= 0 {
+		params = url[qi:]
+		base = url[:qi]
+	}
+	si := strings.LastIndex(base, "/")
+	return base[:si+1] + dbName + params
+}
+
+// TestGetRecentMessages_Truncation is the concrete proof for fix A2 of
+// plans/2026-07-11-nomos-agent-code-review.md: chatWith used to replay a
+// session's ENTIRE history on every turn with no bound. getRecentMessages
+// caps that; this test checks both sides — under the limit, nothing is
+// dropped and truncated=false; over it, only the most recent `limit` come
+// back, in chronological order, with truncated=true.
+func TestGetRecentMessages_Truncation(t *testing.T) {
+	s := newTestStore(t)
+	ctx := context.Background()
+
+	sess, err := s.createSession(ctx, "history window test")
+	if err != nil {
+		t.Fatalf("createSession: %v", err)
+	}
+
+	const total = 35
+	const limit = 30
+	for i := 0; i < total; i++ {
+		role := "user"
+		if i%2 == 1 {
+			role = "assistant"
+		}
+		body := fmt.Appendf(nil, `{"role":%q,"text":"msg-%d"}`, role, i)
+		if err := s.saveMessage(ctx, sess.ID, role, body); err != nil {
+			t.Fatalf("saveMessage %d: %v", i, err)
+		}
+	}
+
+	msgs, truncated, err := s.getRecentMessages(ctx, sess.ID, limit)
+	if err != nil {
+		t.Fatalf("getRecentMessages: %v", err)
+	}
+	if !truncated {
+		t.Errorf("truncated = false, want true (%d messages > limit %d)", total, limit)
+	}
+	if len(msgs) != limit {
+		t.Fatalf("got %d messages, want %d", len(msgs), limit)
+	}
+	// Chronological order: the oldest of the RETAINED messages should be the
+	// (total-limit)-th one saved (msg-5, since msg-0..4 were dropped), and
+	// the last should be the most recently saved (msg-34).
+	wantFirst := fmt.Sprintf("msg-%d", total-limit)
+	wantLast := fmt.Sprintf("msg-%d", total-1)
+	if got := extractText(msgs[0].Content); got != wantFirst {
+		t.Errorf("first retained message = %q, want %q", got, wantFirst)
+	}
+	if got := extractText(msgs[len(msgs)-1].Content); got != wantLast {
+		t.Errorf("last retained message = %q, want %q", got, wantLast)
+	}
+
+	// Under the limit: nothing dropped.
+	sess2, err := s.createSession(ctx, "small session")
+	if err != nil {
+		t.Fatalf("createSession: %v", err)
+	}
+	for i := 0; i < 5; i++ {
+		body := fmt.Appendf(nil, `{"role":"user","text":"msg-%d"}`, i)
+		if err := s.saveMessage(ctx, sess2.ID, "user", body); err != nil {
+			t.Fatalf("saveMessage: %v", err)
+		}
+	}
+	msgs2, truncated2, err := s.getRecentMessages(ctx, sess2.ID, limit)
+	if err != nil {
+		t.Fatalf("getRecentMessages (small): %v", err)
+	}
+	if truncated2 {
+		t.Errorf("truncated = true for a 5-message session under a %d limit, want false", limit)
+	}
+	if len(msgs2) != 5 {
+		t.Errorf("got %d messages, want 5", len(msgs2))
+	}
+}
+
+// TestProposePlan_RefuseInFlight is the concrete proof for the plan-drift
+// fix (2026-07-14, "plan added twice in the sidebar"): proposePlan must
+// REPLACE the step list only while every existing step is still 'pending'
+// (a genuine pre-execution revision), and REFUSE the call once any step has
+// started. The prior append-mode safety net (commit 5384499) preserved
+// history but duplicated the plan in the sidebar when the agent re-proposed
+// on "proceed". Refusing is the correct default — the agent must advance
+// with update_plan_step + run.
+func TestProposePlan_RefuseInFlight(t *testing.T) {
+	s := newTestStore(t)
+	ctx := context.Background()
+
+	sess, err := s.createSession(ctx, "plan refuse test")
+	if err != nil {
+		t.Fatalf("createSession: %v", err)
+	}
+
+	// First call: no steps exist yet — must persist as-is (replace mode,
+	// trivially: nothing to replace).
+	out1, err := s.proposePlan(ctx, sess.ID, []PlanStepInput{{Title: "Step A"}})
+	if err != nil {
+		t.Fatalf("proposePlan #1: %v", err)
+	}
+	if len(out1) != 1 || out1[0]["seq"] != 1 {
+		t.Fatalf("proposePlan #1 = %+v, want one step at seq 1", out1)
+	}
+	if out1[0]["generation"] != 1 {
+		t.Fatalf("proposePlan #1 generation = %v, want 1", out1[0]["generation"])
+	}
+
+	// Mark step 1 as started.
+	if err := s.updatePlanStep(ctx, sess.ID, 1, "running", "", ""); err != nil {
+		t.Fatalf("updatePlanStep: %v", err)
+	}
+
+	// Second call, simulating a model that re-proposes mid-flight (the
+	// operator-reported "proceed" bug): since step 1 has left 'pending',
+	// this MUST refuse with errPlanInFlight, not append or replace.
+	_, err = s.proposePlan(ctx, sess.ID, []PlanStepInput{{Title: "Step B"}})
+	if !errors.Is(err, errPlanInFlight) {
+		t.Fatalf("proposePlan #2: err = %v, want errPlanInFlight (refuse mid-flight re-proposal)", err)
+	}
+
+	// The original step 1 must be untouched — not erased, not appended to.
+	steps, err := s.getPlanSteps(ctx, sess.ID, false)
+	if err != nil {
+		t.Fatalf("getPlanSteps: %v", err)
+	}
+	if len(steps) != 1 {
+		t.Fatalf("got %d persisted steps, want 1 (refused call must not mutate the plan)", len(steps))
+	}
+	if steps[0].Title != "Step A" || steps[0].Status != "running" {
+		t.Errorf("step 1 = %+v, want Step A still running (refused call must not touch it)", steps[0])
+	}
+
+	// Third call BEFORE anything runs on a fresh session: every step is
+	// still pending, so this must REPLACE (mark the prior plan `replaced`),
+	// not refuse. The new plan becomes generation 2.
+	sess2, err := s.createSession(ctx, "plan replace test")
+	if err != nil {
+		t.Fatalf("createSession: %v", err)
+	}
+	if _, err := s.proposePlan(ctx, sess2.ID, []PlanStepInput{{Title: "Original"}}); err != nil {
+		t.Fatalf("proposePlan (initial): %v", err)
+	}
+	if _, err := s.proposePlan(ctx, sess2.ID, []PlanStepInput{{Title: "Revised"}}); err != nil {
+		t.Fatalf("proposePlan (revise before execution): %v", err)
+	}
+	// Default (current generation) view: only the revised step.
+	revisedSteps, err := s.getPlanSteps(ctx, sess2.ID, false)
+	if err != nil {
+		t.Fatalf("getPlanSteps: %v", err)
+	}
+	if len(revisedSteps) != 1 || revisedSteps[0].Title != "Revised" {
+		t.Fatalf("got %+v, want a single 'Revised' step (current-generation view)", revisedSteps)
+	}
+	if revisedSteps[0].Seq != 1 {
+		t.Fatalf("revised step seq = %d, want 1 (seq is generation-relative, resets to 1..N)", revisedSteps[0].Seq)
+	}
+	if revisedSteps[0].Generation != 2 {
+		t.Fatalf("revised step generation = %d, want 2 (prior pending plan is replaced, not deleted, so the counter increments)", revisedSteps[0].Generation)
+	}
+	// all=true audit view: both generations, the original marked `replaced`.
+	allSteps, err := s.getPlanSteps(ctx, sess2.ID, true)
+	if err != nil {
+		t.Fatalf("getPlanSteps(all): %v", err)
+	}
+	if len(allSteps) != 2 {
+		t.Fatalf("all=true got %d steps, want 2 (Original replaced gen1 + Revised gen2)", len(allSteps))
+	}
+	if allSteps[0].Title != "Original" || allSteps[0].Status != "replaced" || allSteps[0].Generation != 1 {
+		t.Errorf("gen1 step = %+v, want Original/replaced/gen1", allSteps[0])
+	}
+	if allSteps[1].Title != "Revised" || allSteps[1].Generation != 2 || allSteps[1].Seq != 1 {
+		t.Errorf("gen2 step = %+v, want Revised/gen2/seq1", allSteps[1])
+	}
+}
+
+// TestUpdatePlanStep_GenerationRelative is the P0.1 regression proof: after a
+// re-plan, update_plan_step(seq=N) — using the 1-based number the model
+// naturally carries — must address the CURRENT generation and never resurrect
+// a superseded generation's `replaced` row. Before the fix, seq was globally
+// increasing across generations, so seq=1 after a re-plan flipped the gen-1
+// `replaced` step back to `running`/`done` while the real gen-2 work went
+// unrecorded.
+func TestUpdatePlanStep_GenerationRelative(t *testing.T) {
+	s := newTestStore(t)
+	ctx := context.Background()
+
+	sess, err := s.createSession(ctx, "gen-relative seq test")
+	if err != nil {
+		t.Fatalf("createSession: %v", err)
+	}
+	// Generation 1: two steps.
+	if _, err := s.proposePlan(ctx, sess.ID, []PlanStepInput{{Title: "A"}, {Title: "B"}}); err != nil {
+		t.Fatalf("proposePlan #1: %v", err)
+	}
+	// Re-plan: setGoal marks the gen-1 plan `replaced`, proposePlan starts gen 2.
+	if err := s.setGoal(ctx, sess.ID, "follow-up sub-task"); err != nil {
+		t.Fatalf("setGoal: %v", err)
+	}
+	if _, err := s.proposePlan(ctx, sess.ID, []PlanStepInput{{Title: "C"}, {Title: "D"}}); err != nil {
+		t.Fatalf("proposePlan #2: %v", err)
+	}
+
+	// The model addresses the new plan with 1-based seq. seq=1 must hit
+	// gen-2 "C", leaving gen-1 "A" (replaced) untouched.
+	if err := s.updatePlanStep(ctx, sess.ID, 1, "running", "", ""); err != nil {
+		t.Fatalf("updatePlanStep(seq=1, running): %v", err)
+	}
+	if err := s.updatePlanStep(ctx, sess.ID, 1, "done", "", ""); err != nil {
+		t.Fatalf("updatePlanStep(seq=1, done): %v", err)
+	}
+
+	all, err := s.getPlanSteps(ctx, sess.ID, true)
+	if err != nil {
+		t.Fatalf("getPlanSteps(all): %v", err)
+	}
+	byTitle := map[string]planStep{}
+	for _, st := range all {
+		byTitle[st.Title] = st
+	}
+	// gen-1 steps stay `replaced` — NOT resurrected to running/done.
+	if byTitle["A"].Status != "replaced" || byTitle["A"].Generation != 1 {
+		t.Errorf("A = %+v, want replaced/gen1 (a superseded row must never be touched)", byTitle["A"])
+	}
+	if byTitle["B"].Status != "replaced" || byTitle["B"].Generation != 1 {
+		t.Errorf("B = %+v, want replaced/gen1", byTitle["B"])
+	}
+	// gen-2 seq=1 advanced; seq=2 untouched.
+	if byTitle["C"].Status != "done" || byTitle["C"].Generation != 2 || byTitle["C"].Seq != 1 {
+		t.Errorf("C = %+v, want done/gen2/seq1 (the 1-based update must address the current generation)", byTitle["C"])
+	}
+	if byTitle["D"].Status != "pending" || byTitle["D"].Seq != 2 {
+		t.Errorf("D = %+v, want pending/seq2", byTitle["D"])
+	}
+
+	// Out-of-range seq must be refused (no current-gen step there).
+	if err := s.updatePlanStep(ctx, sess.ID, 99, "running", "", ""); !errors.Is(err, errPlanStepNotFound) {
+		t.Fatalf("updatePlanStep(seq=99) err = %v, want errPlanStepNotFound", err)
+	}
+}
+
+// TestCompleteTask_AutoCloseEmitsEvents is the P1.1 regression proof:
+// completeTask's bulk auto-close of in-flight steps must emit one
+// plan.step.finished event per closed step (so the live panel converges
+// instead of freezing on "running" after the task completes) and must stamp
+// started_at so no closed step is left un-timestamped (P0.1 fix 5).
+func TestCompleteTask_AutoCloseEmitsEvents(t *testing.T) {
+	s := newTestStore(t)
+	ctx := context.Background()
+
+	sess, err := s.createSession(ctx, "auto-close events test")
+	if err != nil {
+		t.Fatalf("createSession: %v", err)
+	}
+	if _, err := s.proposePlan(ctx, sess.ID, []PlanStepInput{{Title: "A"}, {Title: "B"}}); err != nil {
+		t.Fatalf("proposePlan: %v", err)
+	}
+	// A is running, B still pending at completion time.
+	if err := s.updatePlanStep(ctx, sess.ID, 1, "running", "", ""); err != nil {
+		t.Fatalf("updatePlanStep(1, running): %v", err)
+	}
+	if err := s.completeTask(ctx, sess.ID, "success", "done"); err != nil {
+		t.Fatalf("completeTask: %v", err)
+	}
+
+	// Every auto-closed step should now carry both a started_at and a
+	// finished_at (no NULL-started `done` step).
+	steps, err := s.getPlanSteps(ctx, sess.ID, true)
+	if err != nil {
+		t.Fatalf("getPlanSteps: %v", err)
+	}
+	for _, st := range steps {
+		if st.Status == "done" && st.StartedAt == nil {
+			t.Errorf("step %q done but started_at is NULL (P0.1 fix 5: stamp it)", st.Title)
+		}
+	}
+
+	// Exactly two plan.step.finished events — one per closed step (A and B).
+	var finished int
+	if err := s.pool.QueryRow(ctx,
+		`SELECT COUNT(*) FROM events WHERE type = 'plan.step.finished' AND correlation_id = $1`,
+		sess.ID).Scan(&finished); err != nil {
+		t.Fatalf("count events: %v", err)
+	}
+	if finished != 2 {
+		t.Fatalf("plan.step.finished events = %d, want 2 (one per auto-closed step)", finished)
+	}
+}
+
+// TestHadDiscoveryAndWriteback is the Store-level proof for D.1 (refuse
+// complete_task when discovery ran without writeback). hadDiscovery must
+// report true only after a successful `run` call; hadEntityWriteback must
+// report true only after a successful update_entity_attributes or
+// create_relationship call. The D.1 gate in tasks.go combines these: refuse
+// success when hadDiscovery && !hadEntityWriteback.
+func TestHadDiscoveryAndWriteback(t *testing.T) {
+	s := newTestStore(t)
+	ctx := context.Background()
+
+	sess, err := s.createSession(ctx, "discovery test")
+	if err != nil {
+		t.Fatalf("createSession: %v", err)
+	}
+
+	// Before any tool calls: no discovery, no writeback.
+	if s.hadDiscovery(ctx, sess.ID) {
+		t.Fatal("hadDiscovery = true before any tool calls, want false")
+	}
+	if s.hadEntityWriteback(ctx, sess.ID) {
+		t.Fatal("hadEntityWriteback = true before any tool calls, want false")
+	}
+
+	// A `run` call (discovery) — should set hadDiscovery, not hadEntityWriteback.
+	agentID := uuid.New()
+	s.logActivity(ctx, agentID, sess.ID, "run", nil, "", "uptime output", 100, true, "corr-1", 0)
+	if !s.hadDiscovery(ctx, sess.ID) {
+		t.Fatal("hadDiscovery = false after a successful run call, want true")
+	}
+	if s.hadEntityWriteback(ctx, sess.ID) {
+		t.Fatal("hadEntityWriteback = true after only a run call, want false")
+	}
+
+	// A failed run call should NOT count as discovery (no facts learned).
+	sess2, err := s.createSession(ctx, "failed discovery test")
+	if err != nil {
+		t.Fatalf("createSession: %v", err)
+	}
+	s.logActivity(ctx, agentID, sess2.ID, "run", nil, "", "ssh timeout", 100, false, "corr-2", 0)
+	if s.hadDiscovery(ctx, sess2.ID) {
+		t.Fatal("hadDiscovery = true after a failed run call, want false (no facts learned)")
+	}
+
+	// A get_entity call should NOT count as discovery (DB lookup, not live state).
+	sess3, err := s.createSession(ctx, "lookup test")
+	if err != nil {
+		t.Fatalf("createSession: %v", err)
+	}
+	s.logActivity(ctx, agentID, sess3.ID, "get_entity", nil, "", "entity row", 10, true, "corr-3", 0)
+	if s.hadDiscovery(ctx, sess3.ID) {
+		t.Fatal("hadDiscovery = true after get_entity, want false (DB lookups are not discovery)")
+	}
+
+	// update_entity_attributes sets hadEntityWriteback.
+	sess4, err := s.createSession(ctx, "writeback test")
+	if err != nil {
+		t.Fatalf("createSession: %v", err)
+	}
+	s.logActivity(ctx, agentID, sess4.ID, "update_entity_attributes", nil, "", "ok", 10, true, "corr-4", 0)
+	if !s.hadEntityWriteback(ctx, sess4.ID) {
+		t.Fatal("hadEntityWriteback = false after update_entity_attributes, want true")
+	}
+	// And the discovery+writeback combination (the conv3 scenario).
+	s.logActivity(ctx, agentID, sess4.ID, "run", nil, "", "apt-get update output", 100, true, "corr-5", 0)
+	if !s.hadDiscovery(ctx, sess4.ID) {
+		t.Fatal("hadDiscovery = false after run+writeback, want true")
+	}
+	if !s.hadEntityWriteback(ctx, sess4.ID) {
+		t.Fatal("hadEntityWriteback = false after run+writeback, want true")
+	}
+}
+
+// TestSetGoal_SupersessionEvent is the Store-level proof for P1.4 from
+// plans/2026-07-18-session-review-three-sessions.md: when setGoal is called
+// and a non-empty prior goal already exists with a DIFFERENT value, a
+// task.superseded event must be emitted (so the audit trail records the
+// pivot — the row's goal column will be overwritten, losing the prior intent
+// without this event). When the goal is identical OR no prior goal exists,
+// no supersession event is emitted.
+//
+// Background: session 55927f0a had two set_goal calls; the first was
+// implicitly abandoned when the operator said "lets just keep ludo-library
+// then." Without the event, the prior goal silently disappeared.
+func TestSetGoal_SupersededEvent(t *testing.T) {
+	s := newTestStore(t)
+	ctx := context.Background()
+
+	sess, err := s.createSession(ctx, "goal pivot test")
+	if err != nil {
+		t.Fatalf("createSession: %v", err)
+	}
+
+	// First set_goal — no prior, no supersession event expected.
+	if err := s.setGoal(ctx, sess.ID, "Fix sabnzbd download folder to use ludo-lvm"); err != nil {
+		t.Fatalf("setGoal #1: %v", err)
+	}
+	if n := countEvents(ctx, s, sess.ID, "task.superseded"); n != 0 {
+		t.Errorf("after first set_goal: %d task.superseded events, want 0", n)
+	}
+
+	// Second set_goal with a DIFFERENT goal — supersession event expected.
+	if err := s.setGoal(ctx, sess.ID, "Add NFS export of ludo-lvm to ZimaOS"); err != nil {
+		t.Fatalf("setGoal #2: %v", err)
+	}
+	if n := countEvents(ctx, s, sess.ID, "task.superseded"); n != 1 {
+		t.Errorf("after second set_goal with a different goal: %d task.superseded events, want 1", n)
+	}
+
+	// Third set_goal with the SAME goal as the second — no new supersession
+	// event (idempotent: same goal is a no-op, not a pivot).
+	if err := s.setGoal(ctx, sess.ID, "Add NFS export of ludo-lvm to ZimaOS"); err != nil {
+		t.Fatalf("setGoal #3: %v", err)
+	}
+	if n := countEvents(ctx, s, sess.ID, "task.superseded"); n != 1 {
+		t.Errorf("after third set_goal with same goal as second: %d task.superseded events, want 1 (no new pivot)", n)
+	}
+
+	// The session's current goal must be the latest one set.
+	got, err := s.getSession(ctx, sess.ID)
+	if err != nil {
+		t.Fatalf("getSession: %v", err)
+	}
+	if got.Goal != "Add NFS export of ludo-lvm to ZimaOS" {
+		t.Errorf("session goal = %q, want the second (latest) goal", got.Goal)
+	}
+}
+
+// countEvents counts observability events of the given type correlated to
+// the given session. Used by TestSetGoal_SupersededEvent to assert the
+// task.superseded audit-trail signal was emitted.
+func countEvents(ctx context.Context, s *Store, sessionID, eventType string) int {
+	var n int
+	s.pool.QueryRow(ctx,
+		`SELECT COUNT(*) FROM events WHERE correlation_id = $1 AND type = $2`,
+		sessionID, eventType).Scan(&n)
+	return n
+}

internal/nomos/turngate/turngate.go

@@ -0,0 +1,90 @@
+package turngate
+
+import (
+	"sync"
+	"time"
+)
+
+// TurnGate enforces at most one in-flight agent turn per session.
+//
+// Why this exists (plan 2026-08-03, F1): handleChat runs a turn in the HTTP
+// request goroutine, and every "resume" path (the empty-message reconnect,
+// the auto-continuation worker, the idle sweep, answer-question, the /resume
+// endpoint) launches ANOTHER goroutine running a full turn. Nothing prevented
+// two turns for the SAME session at once, so a network blip that triggered a
+// reconnect would spawn a duplicate resumeSession while the original turn was
+// still alive — their tool calls interleaved on the wire and in the persisted
+// transcript, which is the root cause behind the "parallel/nesting/sequence
+// is off" and "task didn't end / flaky" reports.
+//
+// Model: one permit (buffered-1 channel seeded with a single token) per
+// session id. Acquiring consumes the token; releasing puts it back.
+//   - Background/best-effort callers (resumeSession and everything it backs)
+//     use a non-blocking Acquire and SKIP when busy — a duplicate nudge while a
+//     turn is already running adds nothing, and the continuation/idle tickers
+//     will retry on their own.
+//   - The live chat path (an operator message) waits briefly for a finishing
+//     background turn, then bails with an actionable error if still busy — see
+//     handleChat.
+//
+// The permits map grows one entry per session id seen. For this single-agent
+// homelab process that set is small and bounded by real sessions; cleanup is
+// intentionally omitted (a sweep would race with Acquire/Release and the
+// memory is negligible).
+type TurnGate struct {
+	mu      sync.Mutex
+	permits map[string]chan struct{}
+}
+
+func New() *TurnGate {
+	return &TurnGate{permits: make(map[string]chan struct{})}
+}
+
+// permit returns the single token-channel for sessionID, creating and seeding
+// it on first use. Creation is guarded so two concurrent first-callers for the
+// same id share one channel.
+func (g *TurnGate) permit(sessionID string) chan struct{} {
+	g.mu.Lock()
+	defer g.mu.Unlock()
+	ch, ok := g.permits[sessionID]
+	if !ok {
+		ch = make(chan struct{}, 1)
+		ch <- struct{}{}
+		g.permits[sessionID] = ch
+	}
+	return ch
+}
+
+// Acquire takes the session's permit. With wait <= 0 it is non-blocking
+// (returns false immediately if a turn is active). With wait > 0 it blocks up
+// to wait for the permit, returning false on timeout. Every true return MUST
+// be paired with exactly one Release.
+func (g *TurnGate) Acquire(sessionID string, wait time.Duration) bool {
+	ch := g.permit(sessionID)
+	if wait <= 0 {
+		select {
+		case <-ch:
+			return true
+		default:
+			return false
+		}
+	}
+	t := time.NewTimer(wait)
+	defer t.Stop()
+	select {
+	case <-ch:
+		return true
+	case <-t.C:
+		return false
+	}
+}
+
+// Release returns the session's permit. Idempotent: a Release with no matching
+// Acquire (or a double Release) is a no-op rather than a blocking send.
+func (g *TurnGate) Release(sessionID string) {
+	ch := g.permit(sessionID)
+	select {
+	case ch <- struct{}{}:
+	default:
+	}
+}

internal/nomos/turngate/turngate_test.go

@@ -0,0 +1,114 @@
+package turngate
+
+import (
+	"sync"
+	"sync/atomic"
+	"testing"
+	"time"
+)
+
+func TestTurnGate_NonBlockingSkipsWhenBusy(t *testing.T) {
+	g := New()
+	if !g.Acquire("s1", 0) {
+		t.Fatal("first non-blocking Acquire should succeed on a free session")
+	}
+	// A second non-blocking Acquire (a background resume) must skip, not queue.
+	if g.Acquire("s1", 0) {
+		t.Fatal("second non-blocking Acquire should fail while a turn is active")
+	}
+	// A different session is independent.
+	if !g.Acquire("s2", 0) {
+		t.Fatal("Acquire on a different session should succeed")
+	}
+	g.Release("s2")
+	g.Release("s1")
+	// After Release, the session is free again.
+	if !g.Acquire("s1", 0) {
+		t.Fatal("Acquire should succeed again after Release")
+	}
+	g.Release("s1")
+}
+
+func TestTurnGate_BlockingAcquireWaitsForRelease(t *testing.T) {
+	g := New()
+	if !g.Acquire("s1", 0) {
+		t.Fatal("first Acquire should succeed")
+	}
+
+	got := make(chan bool, 1)
+	go func() { got <- g.Acquire("s1", 2*time.Second) }()
+
+	select {
+	case <-got:
+		t.Fatal("blocking Acquire should wait, not return before Release")
+	case <-time.After(50 * time.Millisecond):
+		// expected: still waiting
+	}
+
+	g.Release("s1")
+	select {
+	case ok := <-got:
+		if !ok {
+			t.Fatal("blocking Acquire should succeed after Release")
+		}
+	case <-time.After(time.Second):
+		t.Fatal("blocking Acquire did not return after Release")
+	}
+	g.Release("s1")
+}
+
+func TestTurnGate_BlockingAcquireTimesOut(t *testing.T) {
+	g := New()
+	g.Acquire("s1", 0) // hold the permit
+
+	start := time.Now()
+	if g.Acquire("s1", 60*time.Millisecond) {
+		t.Fatal("Acquire should time out while permit is held")
+	}
+	if elapsed := time.Since(start); elapsed < 50*time.Millisecond {
+		t.Fatalf("Acquire returned too fast (%v); expected to wait ~60ms", elapsed)
+	}
+	g.Release("s1")
+}
+
+// TestTurnGate_SingleFlightConcurrent is the core F1 guarantee: many concurrent
+// background acquirers on the SAME session, exactly one runs at a time. This is
+// the property that prevents two turns interleaving tool calls.
+func TestTurnGate_SingleFlightConcurrent(t *testing.T) {
+	g := New()
+	const n = 50
+	var inFlight, maxInFlight int64
+	var runs int64
+	var wg sync.WaitGroup
+	wg.Add(n)
+	start := make(chan struct{})
+	for i := 0; i < n; i++ {
+		go func() {
+			defer wg.Done()
+			<-start
+			if !g.Acquire("shared", 0) { // background-style: skip if busy
+				return
+			}
+			defer g.Release("shared")
+			cur := atomic.AddInt64(&inFlight, 1)
+			for {
+				m := atomic.LoadInt64(&maxInFlight)
+				if cur <= m || atomic.CompareAndSwapInt64(&maxInFlight, m, cur) {
+					break
+				}
+			}
+			atomic.AddInt64(&runs, 1)
+			time.Sleep(2 * time.Millisecond)
+			atomic.AddInt64(&inFlight, -1)
+		}()
+	}
+	close(start)
+	wg.Wait()
+
+	if maxInFlight != 1 {
+		t.Fatalf("max in-flight turns = %d, want 1 (turns must not overlap)", maxInFlight)
+	}
+	if runs == 0 {
+		t.Fatal("expected at least one turn to run")
+	}
+}