@gobing-ai/spur 0.3.73 → 0.3.75

package/config/workflows/wrapup-pipeline.yaml

@@ -17,7 +17,10 @@
 #          -> feature-transition (conditional: if vars.feature set)
 #          -> branch-cleanup (conditional: if vars.merge=true)
 #          -> done
-#        (task-resolve with empty list short-circuits to `skipped`).
+#        task-resolve with a VALIDATED empty list short-circuits to `skipped`;
+#        invalid wrap input, an unresolved/non-completed task, a failed metrics
+#        capture, a failed required sync, or a failed feature gate routes to
+#        `failed` (0770) instead of being absorbed by an always-defensive skip.
 #
 # Vars (passed as a JSON object via `--vars`):
 # tasks    — JSON array of WBS strings, passed as a JSON-encoded STRING value
@@ -39,21 +42,46 @@
 # a
 # single model query that both repairs doc drift (sp:doc-evolve) and captures
 #     working learnings (task 0607 R2 — one query answers both; count 2 -> 1).
-# - feature-transition is a soft shell (always exit 0) so a blocked feature sync
-#     does not abort wrap-up after learnings/metrics already landed.
+# - feature-transition is a soft shell (always exit 0); a failed required sync
+#     or feature gate records FAIL and routes to `failed` instead of claiming
+#     success (0770).
+# - 0770 truthful outcomes: vars.tasks is parsed and validated ONCE at
+#     task-resolve (JSON array of non-empty WBS strings, every member resolved
+#     via `task show` to a completed status); siblings consume the normalized
+#     run-scoped artifact and never re-parse raw input (malformed input can no
+#     longer be silently re-interpreted as an empty list). All captures are
+#     run-scoped under .spur/run/<runId>-wrapup-*; a lookup failure is
+#     recorded, never silently omitted as success.
+#   - 0783 validated consumers (audit 0781 F-04): validation accepts only
+#     canonical four-digit WBS strings (whitespace is rejected, not trimmed);
+#     the route writer, every guard, model prompt and operator note after
+#     resolution read the run-scoped capture, never raw vars.tasks, and a
+#     missing/corrupted capture or status refuses progression. Metrics
+#     revalidate the capture, require well-shaped task lookups, serialize rows
+#     with jq (never interpolated printf), and write PASS only after every
+#     required append succeeds. Required feature sync succeeds only for a
+#     valid matching proposal with no gateBlocked/requiresConfirm condition
+#     whose target status is freshly observed; applied:false is a successful
+#     no-op only when from == to and that status is observed — nonzero,
+#     malformed, partial, blocked or unreadable outcomes fail explicitly, and
+#     an affected-feature check can never convert a failed sync into success.
 # - branch-cleanup HITL is exhaustive (yes/no/cancel → done; missing answer
-# defense).
+# defense) and consent-only: it records the decision, performs no git op.
 #   - Wrap-up never mutates task status (consumes completed work only).
 
 "$schema": "@gobing-ai/spur/schemas/state-machine-workflow.schema.json"
 kind: state-machine
 name: wrapup-pipeline
+version: "3"
 description: "Post-execution wrap-up: doc-sync (doc drift + learning capture), metrics, feature-transition, branch-cleanup"
 iterationBound: 10
 initialState: start
 terminalStates:
   - done
   - skipped
+  - failed
+failureStates:
+  - failed
 vars:
   tasks: "[]"
   feature: ""
@@ -64,22 +92,11 @@ vars:
   __runId: ""
   mode: ""
   stepTimeoutMs: "1800000"
-  # Corpus-aware quality gate for the feature-transition hop (R1, task 0625).
-  # The per-task gate (`spur-check`) deliberately excludes corpus-check — the
-  # only
-  # sweep that observes feature-level findings — so a transition that arms one
-  # would go green. This gate is therefore the CORPUS SWEEP ALONE (~29 s), not
-  # `spur-check-new`: every task in the feature already paid a full `spur-check`
-  # in its own pipeline, and the states that run before this hop (`doc-sync`,
-  # `metrics-record`) write only markdown and task sections, which Biome skips
-  # (`ignoreUnknown: true`). Re-running the ~105 s per-task gate here
-  # re-verifies
-  # unchanged code and measures nothing new. Soft by design: the gate reports
-  # PASS/FAIL and lets the operator decide; it never hard-fails the wrap-up
-  # shell.
+  # Recheck the affected feature after sync (D61 / ADR-108), not the corpus
+  # or full code suite. Explicit caller overrides remain supported.
   # TRUSTED CONFIG ONLY — this string is executed via `sh -c` (same surface as
   # task-pipeline's qualityGateCmd). Never interpolate untrusted input into it.
-  featureGateCmd: "bun run corpus-check"
+  featureGateCmd: '$spurBin feature check "$feature"'
   __hitlAnswer: ""
 
 states:
@@ -94,27 +111,95 @@ states:
 
   - id: task-resolve
     description: >
-      Validate the task list and evaluate closed proportional routing (0758 R1-R4).
-      Empty list -> skipped; tasks > 0 && mode == fast -> fast-path (metrics-record,
+      Validate the task list ONCE and evaluate closed proportional routing (0758
+      R1-R4). Validation (0770 + 0783): vars.tasks must be a JSON array of
+      canonical four-digit WBS strings — non-arrays, non-strings, whitespace or
+      other shapes are rejected, not trimmed — deduplicated in first-seen
+      order, with __runId present, and every member must resolve via `spur task
+      show` to a completed status (done/cancelled); failed or malformed
+      lookups fail. Invalid input or an unresolved/non-completed task records
+      FAIL (reason into the run-scoped reason file) and routes to `failed`;
+      the route writer and every guard, prompt and note after resolution read
+      the normalized run-scoped artifact
+      .spur/run/<runId>-wrapup-tasks.json and never re-parse raw input — a
+      missing or corrupted capture refuses progression. A validated empty list
+      -> skipped; length > 0 && mode == fast -> fast-path (metrics-record,
       bypassing doc-sync); missing/unknown/conflict -> doc-sync (safety-path).
       Writes bounded machine-readable reason to reasonFile.
     onEnter:
       - kind: note
         options:
           message: "Resolving task list: ${vars.tasks} (proportional mode: ${vars.mode})"
-      # Run-id confinement (0758 R3). The run-scoped reason file is the only route artifact: a
-      # second copy at the fixed path `.spur/run/wrapup-route-reason.txt` had no reader and was
-      # overwritten by whichever run finished last, so a route claim read from it belonged to no
-      # particular run. The log append carries the run id for the same reason — an unattributed
-      # line is log scraping, which R5 rejects as evidence.
+      # 0770/0783 truthful input: parse and validate vars.tasks exactly once,
+      # here and only here. Malformed JSON, a non-array, a non-string entry,
+      # whitespace or a non-canonical WBS (anything but four digits), a missing
+      # __runId, an unresolved task, a failed or malformed lookup, or a task
+      # not at a completed status records FAIL (routed to `failed` by the first
+      # declared edge) and is never re-interpreted as an empty list by a later
+      # sibling guard. The normalized, deduplicated list is persisted
+      # run-scoped; the route writer, metrics, prompts, notes and guards only
+      # ever read that capture.
+      - kind: shell
+        options:
+          command: >-
+            mkdir -p .spur/run &&
+            RUN_ID="$__runId" &&
+            if [ -z "$RUN_ID" ]; then
+              echo "task-resolve: __runId is empty — refusing the legacy fixed-path fallback" >&2;
+              exit 1;
+            fi &&
+            REASON_FILE=".spur/run/$RUN_ID-route-reason.txt" &&
+            STATUS_FILE=".spur/run/$RUN_ID-wrapup-resolve.status" &&
+            TASKS_FILE=".spur/run/$RUN_ID-wrapup-tasks.json" &&
+            if ! printf '%s' "$tasks" | jq -e 'type == "array" and all(.[]; type == "string" and test("^[0-9]{4}$"))' > /dev/null 2>&1; then
+              echo "task-resolve: tasks must be a JSON array of canonical four-digit WBS strings (whitespace is rejected, not trimmed)" >&2;
+              echo "failed:tasks is not a JSON array of canonical four-digit WBS strings" > "$REASON_FILE";
+              printf 'FAIL\n' > "$STATUS_FILE";
+            else
+              printf '%s' "$tasks" | jq -c 'reduce .[] as $w ([]; if any(.[]; . == $w) then . else . + [$w] end)' > "$TASKS_FILE";
+              RESOLVE_RC=0;
+              for wbs in $(jq -r '.[]' "$TASKS_FILE"); do
+                TASK_JSON=$($spurBin task show "$wbs" --json 2>/dev/null || true);
+                TASK_STATUS=$(printf '%s' "$TASK_JSON" | jq -r '.frontmatter.status // .status // empty' 2>/dev/null || true);
+                if [ "$TASK_STATUS" != "done" ] && [ "$TASK_STATUS" != "cancelled" ]; then
+                  STATUS_DISPLAY="$TASK_STATUS";
+                  if [ -z "$STATUS_DISPLAY" ]; then STATUS_DISPLAY=unresolved; fi;
+                  echo "task-resolve: task $wbs did not resolve to a completed status (status=$STATUS_DISPLAY)" >&2;
+                  RESOLVE_RC=1;
+                fi;
+              done;
+              if [ "$RESOLVE_RC" -eq 0 ]; then
+                printf 'PASS\n' > "$STATUS_FILE";
+              else
+                echo "failed:unresolved or non-completed task (see $TASKS_FILE)" > "$REASON_FILE";
+                printf 'FAIL\n' > "$STATUS_FILE";
+              fi;
+            fi;
+            exit 0
+      # Route reason writer (0758 R3/R4 run-attribution pins intact). Runs AFTER
+      # validation and reads ONLY the validated run-scoped capture — never raw
+      # vars.tasks (0783 R2/R5); the fixed-path fallback is gone because
+      # validation refuses an empty __runId above. A FAIL resolve exits without
+      # writing: the failed reason written by validation stands, and the skip
+      # reason can never mask it. A missing/corrupted capture yields N=-1 and
+      # no skipped claim; the failed defense edge owns the run.
       - kind: shell
         options:
           command: >-
             mkdir -p .spur/run .spur/memory &&
             RUN_ID="$__runId" &&
-            if [ -z "$RUN_ID" ]; then RUN_ID="wrapup"; fi &&
+            if [ -z "$RUN_ID" ]; then
+              echo "task-resolve: __runId is empty — refusing to write a route reason" >&2;
+              exit 1;
+            fi &&
             REASON_FILE=".spur/run/$RUN_ID-route-reason.txt" &&
-            if [ "$(echo "$tasks" | jq length 2>/dev/null || echo 0)" -eq 0 ]; then
+            STATUS_FILE=".spur/run/$RUN_ID-wrapup-resolve.status" &&
+            TASKS_FILE=".spur/run/$RUN_ID-wrapup-tasks.json" &&
+            if [ "$(cat "$STATUS_FILE" 2>/dev/null)" = FAIL ]; then
+              exit 0;
+            fi &&
+            N=$(jq length "$TASKS_FILE" 2>/dev/null || echo -1) &&
+            if [ "$N" -eq 0 ]; then
               echo "skipped:empty task list" > "$REASON_FILE";
             elif [ "$mode" = "fast" ]; then
               echo "fast:evidence complete+consistent" > "$REASON_FILE";
@@ -146,88 +231,115 @@ states:
         options:
           agent: ${vars.agent}
           input: >-
-            First run Skill(skill="sp:doc-evolve", args="wrapup tasks=${vars.tasks}") to repair drift in
+            First run Skill(skill="sp:doc-evolve", args="wrapup") to repair drift in
             docs/00_ADR.md, docs/03_ARCHITECTURE.md, docs/04_DESIGN.md, docs/design/* per constitution edit
-            rules; do not write task/feature corpus. THEN extract working learnings from tasks ${vars.tasks} —
+            rules; do not write task/feature corpus. The batch task list is the normalized, validated
+            capture at .spur/run/${vars.__runId}-wrapup-tasks.json — read it for the WBS ids; never
+            re-parse raw input. THEN extract working learnings from those tasks —
             conventions, errors fixed, patterns, gotchas, grouped by date and task WBS — as raw markdown
-            (no fences) and END your final message with that markdown (it is captured to .spur/run/wrapup-learnings.md).
+            (no fences) and END your final message with that markdown (it is captured to the
+            run-scoped .spur/run/${vars.__runId}-wrapup-learnings.md).
           # Declared Layer-1 role (0538 R2): routing reason beside the agent:
           # pin.
           role: coder
-          answerFile: .spur/run/wrapup-learnings.md
-          expectFile: .spur/run/wrapup-learnings.md
+          answerFile: .spur/run/${vars.__runId}-wrapup-learnings.md
+          expectFile: .spur/run/${vars.__runId}-wrapup-learnings.md
           timeoutMs: ${vars.stepTimeoutMs}
       - kind: shell
         options:
           command: >-
-            mkdir -p .spur/memory &&
-            if test -s .spur/run/wrapup-learnings.md; then
-              cat .spur/run/wrapup-learnings.md >> .spur/memory/learnings.md &&
+            mkdir -p .spur/run .spur/memory &&
+            LEARNINGS_FILE=".spur/run/$__runId-wrapup-learnings.md" &&
+            if test -s "$LEARNINGS_FILE"; then
+              cat "$LEARNINGS_FILE" >> .spur/memory/learnings.md &&
               printf '\n' >> .spur/memory/learnings.md;
             else
-              echo "wrapup: learning capture missing or empty at $PWD/.spur/run/wrapup-learnings.md - append skipped" >&2;
+              echo "wrapup: learning capture missing or empty at $PWD/$LEARNINGS_FILE - append skipped" >&2;
             fi &&
             exit 0
 
   - id: metrics-record
     description: >
       Append one JSONL row per task to .spur/memory/wrapup-metrics.jsonl deterministically.
+      Revalidates the normalized run-scoped capture from task-resolve (0770 + 0783):
+      a missing, corrupted or non-canonical capture records FAIL; every member must
+      produce a successful, well-shaped `task show` lookup (a missing metrics row is
+      never silently absorbed as success); rows are serialized with jq — never
+      interpolated printf JSON — so escaped fields stay parseable; and PASS is
+      written only after every required append succeeds. A missing per-task verdict
+      is UNKNOWN telemetry, never proof of completion; previously written valid rows
+      survive a failure.
     onEnter:
       - kind: shell
         options:
           command: >-
-            mkdir -p .spur/memory &&
-            printf '%s' "$tasks" | jq -r '(if type == "string" then (try fromjson catch .) else . end) | if type == "array" then .[] else empty end' |
-            while IFS= read -r wbs; do
-              if [ -n "$wbs" ]; then
+            mkdir -p .spur/run .spur/memory &&
+            STATUS_FILE=".spur/run/$__runId-wrapup-metrics.status" &&
+            TASKS_FILE=".spur/run/$__runId-wrapup-tasks.json" &&
+            METRICS_RC=0 &&
+            if ! jq -e 'type == "array" and all(.[]; type == "string" and test("^[0-9]{4}$"))' "$TASKS_FILE" > /dev/null 2>&1; then
+              echo "metrics-record: run-scoped task capture missing, corrupted or non-canonical — refusing to record metrics" >&2;
+              printf 'FAIL\n' > "$STATUS_FILE";
+            else
+              for wbs in $(jq -r '.[]' "$TASKS_FILE"); do
                 task_json=$($spurBin task show "$wbs" --json 2>/dev/null || true);
-                if [ -n "$task_json" ]; then
-                  feat=$(printf '%s' "$task_json" | jq -r '.frontmatter.feature_id // .feature_id // empty' 2>/dev/null || true);
-                  stat=$(printf '%s' "$task_json" | jq -r '.frontmatter.status // .status // "unknown"' 2>/dev/null || true);
+                if printf '%s' "$task_json" | jq -e '(.frontmatter.status // .status) != null' > /dev/null 2>&1; then
+                  feat=$(printf '%s' "$task_json" | jq -r '.frontmatter.feature_id // .feature_id // ""');
+                  stat=$(printf '%s' "$task_json" | jq -r '.frontmatter.status // .status // "unknown"');
                   verdict="UNKNOWN";
                   if [ -f ".spur/run/$wbs-verdict.json" ]; then
                     v=$(jq -r '.verdict // "UNKNOWN"' ".spur/run/$wbs-verdict.json" 2>/dev/null || true);
                     if [ -n "$v" ]; then verdict="$v"; fi;
                   fi;
                   ts=$(date -u +"%Y-%m-%dT%H:%M:%SZ");
-                  printf '{"wbs":"%s","feature_id":"%s","status":"%s","verdict":"%s","timestamp":"%s"}\n' "$wbs" "$feat" "$stat" "$verdict" "$ts" >> .spur/memory/wrapup-metrics.jsonl;
+                  if ! jq -cn --arg wbs "$wbs" --arg feature_id "$feat" --arg status "$stat" --arg verdict "$verdict" --arg timestamp "$ts" '{wbs:$wbs, feature_id:$feature_id, status:$status, verdict:$verdict, timestamp:$timestamp}' >> .spur/memory/wrapup-metrics.jsonl; then
+                    echo "metrics-record: metrics append failed for task $wbs — recording FAIL instead of claiming the row landed" >&2;
+                    METRICS_RC=1;
+                  fi;
+                else
+                  echo "metrics-record: task $wbs lookup failed or was malformed — recording FAIL instead of silently omitting its metrics row" >&2;
+                  METRICS_RC=1;
                 fi;
-              fi;
-            done &&
+              done;
+              if [ "$METRICS_RC" -eq 0 ]; then printf 'PASS\n' > "$STATUS_FILE"; else printf 'FAIL\n' > "$STATUS_FILE"; fi;
+            fi;
             exit 0
 
   - id: feature-transition
     description: >
       If vars.feature is set, sync feature status via bounded feature-sync-bounded
-      (0411). Soft shell — blocked/failed sync must not abort wrap-up after metrics.
-      Empty vars.feature is NOT a blocked sync but a mis-invocation (the engine only
-      enters this state when vars.feature is set) and fails loud (dogfood 2026-08-15,
-      feature I3: a silent exit 0 made the wrap look complete with no transition).
+      (0411). The shell terminates `exit 0` so artifacts already landed are never
+      discarded, but the outcome is truthful (0770 + 0783 R4): sync succeeds only
+      with a valid proposal matching vars.feature, no gateBlocked/requiresConfirm
+      condition, and a freshly observed feature status equal to proposal.to;
+      applied:false is a successful explicit no-change only when from == to and
+      that status is observed. A nonzero, malformed, partial, blocked or
+      unreadable outcome records FAIL in the run-scoped sync status and routes
+      to `failed` — an affected-feature check cannot convert a failed sync into
+      success. Empty vars.feature is NOT a blocked sync but a mis-invocation
+      (the engine only enters this state when vars.feature is set) and fails
+      loud (dogfood 2026-08-15, feature I3).
     onEnter:
-      # Genuinely soft: the terminating `exit 0` must be reached on BOTH
-      # branches, so
-      # the sync is chained with `;` — with `&&` a non-zero sync skipped `exit
-      # 0` and
-      # aborted the run at this state, discarding the learnings/metrics that
-      # already
-      # landed and never reaching branch-cleanup/done (the exact failure the
-      # state
-      # description promises against). `feature-sync-bounded.ts` is a
-      # Spur-monorepo
-      # path (`spur init` does not scaffold `plugins/sp/`), so seeded projects
-      # fall
-      # back to the plain `spur feature sync` verb.
-      # R1 (0625): a feature transition that changed state may have armed a
-      # feature-level finding the per-task fast gate (spur-check) never sees.
-      # Run the corpus-aware gate after an applied sync OR a non-zero exit that
-      # may follow a partial multi-hop transition, then report PASS/FAIL. Soft
-      # by design: the operator decides, and wrap-up never hard-fails here.
+      # The terminating `exit 0` must be reached on every branch — the sync is
+      # chained with `;`, because `&&` would abort the state, discarding the
+      # learnings/metrics that already landed. Soft TERMINATION is not soft
+      # TRUTH: the result is classified per 0783 R4 and only a verified sync
+      # writes PASS. `feature-sync-bounded.ts` is a Spur-monorepo path (`spur
+      # init` does not scaffold `plugins/sp/`), so seeded projects fall back to
+      # the plain `spur feature sync` verb.
+      # R1 (0625): a feature transition that changed state may have left
+      # cross-task fallout the per-task fast gates never see. The
+      # affected-feature gate (featureGateCmd) runs after an applied sync OR a
+      # non-zero exit that may follow a partial multi-hop transition. It is a
+      # diagnostic only: a gate PASS can never convert a failed sync into
+      # success, and a gate FAIL fails the step (0770 failure routing).
       # NOTE: do not add
       # `#` comments INSIDE the folded shell string — YAML `>-` folds them onto
       # the following statement line and comments it out.
       - kind: shell
         options:
           command: >-
+            mkdir -p .spur/run &&
             if [ -z "$feature" ]; then
               echo "feature-transition: vars.feature is empty — refusing no-op feature sync (mis-invocation, not a blocked sync)" >&2;
               exit 1;
@@ -241,47 +353,96 @@ states:
             fi);
             SYNC_RC=$?;
             printf '%s\n' "$SYNC_OUTPUT";
-            APPLIED=$(printf '%s' "$SYNC_OUTPUT" | jq -r '.applied // false' 2>/dev/null || echo false);
+            OBSERVED=$($spurBin feature show "$feature" --json 2>/dev/null | jq -r '.status // .frontmatter.status // empty' 2>/dev/null || true);
+            if [ -z "$OBSERVED" ]; then OBSERVED=unreadable; fi;
+            APPLIED=unreadable;
+            SYNC_OK=0;
+            REASON="";
+            if [ "$SYNC_RC" -ne 0 ]; then
+              REASON="sync exited nonzero (rc=$SYNC_RC)";
+            elif ! printf '%s' "$SYNC_OUTPUT" | jq -e 'type == "object" and (.proposal | type == "object") and (.proposal.featureId | type == "string") and (.proposal.from | type == "string") and (.proposal.to | type == "string") and (.applied | type == "boolean")' > /dev/null 2>&1; then
+              REASON="malformed or unreadable sync result";
+            else
+              P_FROM=$(printf '%s' "$SYNC_OUTPUT" | jq -r '.proposal.from // ""');
+              P_TO=$(printf '%s' "$SYNC_OUTPUT" | jq -r '.proposal.to // ""');
+              APPLIED=$(printf '%s' "$SYNC_OUTPUT" | jq -r '.applied');
+              if [ "$(printf '%s' "$SYNC_OUTPUT" | jq -r '.proposal.featureId // ""')" != "$feature" ]; then
+                REASON="sync proposal does not match feature $feature";
+              elif [ "$(printf '%s' "$SYNC_OUTPUT" | jq -r '.proposal.gateBlocked // false')" = "true" ]; then
+                REASON="sync proposal is gate-blocked — a blocked sync is not a no-change success";
+              elif [ "$(printf '%s' "$SYNC_OUTPUT" | jq -r '.proposal.requiresConfirm // false')" = "true" ]; then
+                REASON="sync proposal requires operator confirmation";
+              elif [ "$APPLIED" = "true" ] && [ "$OBSERVED" != "$P_TO" ]; then
+                REASON="applied sync did not land on the proposal target (observed=$OBSERVED, to=$P_TO)";
+              elif [ "$APPLIED" = "false" ] && { [ "$P_FROM" != "$P_TO" ] || [ "$OBSERVED" != "$P_TO" ]; }; then
+                REASON="sync applied nothing without a from==to observed no-op (from=$P_FROM, to=$P_TO, observed=$OBSERVED)";
+              else
+                SYNC_OK=1;
+              fi;
+            fi;
+            GATE="skipped";
             if [ "$APPLIED" = "true" ] || [ "$SYNC_RC" -ne 0 ]; then
-              echo "feature-transition: sync applied or failed after a possible partial transition for $feature — running corpus-aware gate: $featureGateCmd";
+              echo "feature-transition: sync applied or failed after a possible partial transition for $feature — running feature gate: $featureGateCmd";
               if sh -c "$featureGateCmd"; then
-                echo "feature-transition: corpus-aware gate PASS for feature $feature";
+                GATE="PASS";
+                echo "feature-transition: feature gate PASS for feature $feature";
               else
-                echo "feature-transition: corpus-aware gate FAIL for feature $feature — inspect findings before reporting the transition complete" >&2;
+                GATE="FAIL";
+                echo "feature-transition: feature gate FAIL for feature $feature — inspect findings before reporting the transition complete" >&2;
               fi;
             else
-              echo "feature-transition: sync did not apply a transition (rc=$SYNC_RC, applied=$APPLIED) — corpus-aware gate skipped";
+              echo "feature-transition: sync did not apply a transition (rc=$SYNC_RC, applied=$APPLIED) — feature gate skipped";
+            fi;
+            SYNC_STATUS="PASS";
+            if [ "$SYNC_OK" -ne 1 ] || [ "$GATE" = "FAIL" ]; then
+              SYNC_STATUS="FAIL";
+              echo "feature-transition: required synchronization failed for $feature — $REASON; gate=$GATE" >&2;
+            elif [ "$APPLIED" = "false" ]; then
+              echo "feature-transition: feature sync verified for $feature (from==to observed at $OBSERVED, gate=$GATE) — explicit no-change";
+            else
+              echo "feature-transition: feature sync verified for $feature (applied, observed=$OBSERVED, gate=$GATE)";
+            fi;
+            if [ "$SYNC_STATUS" = "PASS" ]; then
+              printf 'PASS\n' > ".spur/run/$__runId-wrapup-sync.status";
+            else
+              printf 'FAIL\n' > ".spur/run/$__runId-wrapup-sync.status";
             fi;
             exit 0
 
   - id: branch-cleanup
     description: >
-      Irreversible HITL gate. If vars.merge=true, dispatch branch cleanup (merge or
-      delete). This gate ALWAYS pauses — even under --auto — because branch operations
-      are irreversible (Auto-Decision Principle #6). The operator must explicitly confirm.
+      Irreversible-operation HITL gate, consent-only (0770). If vars.merge=true,
+      branch cleanup (merge or delete) is a candidate — but this gate records
+      the operator decision and performs NO git operation itself; an operator
+      or a separate approved surface executes any actual merge/delete. This
+      gate ALWAYS pauses — even under --auto — because branch operations are
+      irreversible (Auto-Decision Principle #6).
     pause: true
     onEnter:
       - kind: hitl.confirm
         options:
-          prompt: "Branch cleanup for tasks ${vars.tasks}. This is IRREVERSIBLE (merge or delete). Confirm to proceed?"
+          prompt: "Branch cleanup for the validated task list (.spur/run/${vars.__runId}-wrapup-tasks.json). This is IRREVERSIBLE (merge or delete). Confirm to proceed?"
 
   - id: done
     description: >
-      Terminal — wrap-up complete. Output summary: tasks wrapped, learnings captured,
-      metrics recorded, feature advanced (if applicable), branch cleaned (if applicable).
+      Terminal — wrap-up complete. Summary: tasks wrapped, learnings captured,
+      metrics recorded, feature advanced or explicitly no-change (if applicable),
+      branch-cleanup decision recorded (consent-only; no git operation performed).
     onEnter:
       - kind: note
         options:
-          message: "Wrap-up pipeline complete for tasks: ${vars.tasks}. Learnings at .spur/memory/learnings.md, metrics at .spur/memory/wrapup-metrics.jsonl."
-      # Checkpoint write: record session state for resume (Phase 4, task 0171
-      # R3)
-      - kind: shell
-        options:
-          command: 'mkdir -p .spur/memory/sessions && echo "checkpoint: wrapup-pipeline done tasks=$tasks ts=$(date -u +%Y-%m-%dT%H:%M:%SZ)" > .spur/memory/sessions/wrapup-checkpoint.md'
+          message: "Wrap-up pipeline complete for the validated task list at .spur/run/${vars.__runId}-wrapup-tasks.json. Learnings at .spur/memory/learnings.md, metrics at .spur/memory/wrapup-metrics.jsonl. Branch cleanup: consent recorded only (merge=${vars.merge}); no git operation performed by wrap-up."
 
   - id: skipped
     description: Terminal — wrap-up skipped (empty task list or operator abort).
 
+  - id: failed
+    description: >
+      Terminal — wrap-up failed (0770): invalid wrap input, an unresolved or
+      non-completed task, a failed metrics capture, a failed required sync, or
+      a failed feature gate. Already-written artifacts (learnings, metrics
+      rows, route reason) are preserved; nothing is rolled back.
+
 transitions:
   # ── start -> task-resolve ──
   - from: start
@@ -290,31 +451,47 @@ transitions:
     guard:
       kind: always
 
-  # ── task-resolve: proportional route table (task 0758 R1-R4) ──
+  # ── task-resolve: validation fail + proportional route table (0758 R1-R4, 0770) ──
+  - from: task-resolve
+    to: failed
+    description: >
+      Validation FAIL (0770): malformed wrap input, a non-string/empty entry, an
+      unresolved or non-completed task, or an empty __runId. Declared before the
+      route edges so invalid input can never be absorbed by the empty-list skip.
+    guard:
+      kind: shell
+      options:
+        command: 'test "$(cat .spur/run/$__runId-wrapup-resolve.status 2>/dev/null)" = FAIL'
+  # Route edges key on the validated run-scoped capture, never raw vars.tasks
+  # (0783 R2). A missing or corrupted capture yields -1, which satisfies none
+  # of the numeric edges, so the run falls through to the always-defense
+  # `failed` edge instead of claiming a skip or a route.
   - from: task-resolve
     to: skipped
-    description: Task list is empty — skip wrap-up.
+    description: Validated task list is empty — skip wrap-up (only a validated [] may skip).
     guard:
       kind: shell
       options:
-        command: 'test "$(echo "$tasks" | jq length 2>/dev/null || echo 0)" -eq 0'
+        command: 'test "$(jq length .spur/run/$__runId-wrapup-tasks.json 2>/dev/null || echo -1)" -eq 0'
   - from: task-resolve
     to: metrics-record
     description: Proportional fast path — complete and consistent evidence bypasses doc-sync.
     guard:
       kind: shell
       options:
-        command: 'test "$(echo "$tasks" | jq length 2>/dev/null || echo 0)" -gt 0 && test "$mode" = fast'
+        command: 'test "$(jq length .spur/run/$__runId-wrapup-tasks.json 2>/dev/null || echo -1)" -gt 0 && test "$mode" = fast'
   - from: task-resolve
     to: doc-sync
     description: Proportional safety path — missing/unknown/conflicting evidence routes to full doc-sync.
     guard:
       kind: shell
       options:
-        command: 'test "$(echo "$tasks" | jq length 2>/dev/null || echo 0)" -gt 0 && test "$mode" != fast'
+        command: 'test "$(jq length .spur/run/$__runId-wrapup-tasks.json 2>/dev/null || echo -1)" -gt 0 && test "$mode" != fast'
   - from: task-resolve
-    to: skipped
-    description: Defense — task list unparseable; skip rather than fail with no-passing-transition.
+    to: failed
+    description: >
+      Defense (0770) — no PASS/FAIL resolve status (shell crashed or never ran):
+      route to `failed` rather than claiming a skip.
     guard:
       kind: always
 
@@ -325,47 +502,67 @@ transitions:
     guard:
       kind: always
 
-  # ── metrics-record: conditional routing based on feature and merge ──
-  # Declaration order matters: feature-transition is tried first (if feature is
-  # set),
-  # then branch-cleanup (if merge=true but no feature), then done (neither).
+  # ── metrics-record: fail edge, then conditional routing (feature → merge → done) ──
+  # Declaration order matters: the FAIL edge is tried first (0770), then
+  # feature-transition (if feature is set), then branch-cleanup (if merge=true
+  # but no feature), then done (neither). Routing edges additionally require
+  # the metrics status PASS so a missing status can never claim success.
+  - from: metrics-record
+    to: failed
+    description: >
+      Metrics capture failed for at least one task — a missing metrics row is
+      recorded as failure, never silently omitted as success (0770).
+    guard:
+      kind: shell
+      options:
+        command: 'test "$(cat .spur/run/$__runId-wrapup-metrics.status 2>/dev/null)" = FAIL'
   - from: metrics-record
     to: feature-transition
-    description: Feature id is set — advance feature through legal lifecycle edges.
+    description: Metrics PASS and feature id is set — advance feature through legal lifecycle edges.
     guard:
       kind: shell
       options:
-        command: 'test -n "$feature"'
+        command: 'test "$(cat .spur/run/$__runId-wrapup-metrics.status 2>/dev/null)" = PASS && test -n "$feature"'
   - from: metrics-record
     to: branch-cleanup
-    description: No feature id, but merge is requested — go to branch cleanup.
+    description: Metrics PASS, no feature id, but merge is requested — go to branch cleanup.
     guard:
       kind: shell
       options:
-        command: 'test -z "$feature" && test "$merge" = true'
+        command: 'test "$(cat .spur/run/$__runId-wrapup-metrics.status 2>/dev/null)" = PASS && test -z "$feature" && test "$merge" = true'
   - from: metrics-record
     to: done
-    description: No feature id and no merge — wrap-up is done.
+    description: Metrics PASS, no feature id and no merge — wrap-up is done.
     guard:
       kind: shell
       options:
-        command: 'test -z "$feature" && test "$merge" != true'
+        command: 'test "$(cat .spur/run/$__runId-wrapup-metrics.status 2>/dev/null)" = PASS && test -z "$feature" && test "$merge" != true'
 
-  # ── feature-transition: -> branch-cleanup (if merge) or done ──
+  # ── feature-transition: fail edge, then branch-cleanup (if merge) or done ──
+  - from: feature-transition
+    to: failed
+    description: >
+      Required synchronization failed (0770): non-zero sync exit, an invalid
+      sync result, or a failed feature gate. The failed status carries the
+      actual reason; already-written learnings/metrics stay on disk.
+    guard:
+      kind: shell
+      options:
+        command: 'test "$(cat .spur/run/$__runId-wrapup-sync.status 2>/dev/null)" = FAIL'
   - from: feature-transition
     to: branch-cleanup
-    description: Feature advanced — merge is requested.
+    description: Sync PASS — merge is requested.
     guard:
       kind: shell
       options:
-        command: 'test "$merge" = true'
+        command: 'test "$(cat .spur/run/$__runId-wrapup-sync.status 2>/dev/null)" = PASS && test "$merge" = true'
   - from: feature-transition
     to: done
-    description: Feature advanced — no merge requested.
+    description: Sync PASS — no merge requested.
     guard:
       kind: shell
       options:
-        command: 'test "$merge" != true'
+        command: 'test "$(cat .spur/run/$__runId-wrapup-sync.status 2>/dev/null)" = PASS && test "$merge" != true'
 
   # ── branch-cleanup: HITL exhaustive routing (yes / no / cancel → done) ──
   # No irreversible git op is wired yet; confirmation is recorded, then wrap

package/package.json

@@ -1,6 +1,6 @@
 {
     "name": "@gobing-ai/spur",
-    "version": "0.3.73",
+    "version": "0.3.75",
     "description": "Spur CLI — local-first harness for mainstream coding agents: constraint checking, workflow orchestration, agent health, and history analytics. Bun-native; exposes the `spur` command.",
     "keywords": [
         "spur",