LINBEE-30174 | fix: backfill a missed grade at SessionEnd (v1.1.1)

.claude-plugin/marketplace.json

@@ -2,7 +2,7 @@
   "name": "linearb-ai",
   "metadata": {
     "description": "LinearB plugins for AI coding agents: engineering context from LinearB, right inside your agent.",
-    "version": "1.1.0"
+    "version": "1.1.1"
   },
   "owner": {
     "name": "LinearB",
@@ -12,7 +12,7 @@
     {
       "name": "agentic-advisor",
       "description": "Before writing code, grades how fragile the target is from LinearB health signals + local git history and holds the agent to a matching effort level (LOW/MEDIUM/HIGH) — lean on calm repos, defensive on fragile ones.",
-      "version": "1.1.0",
+      "version": "1.1.1",
       "author": {
         "name": "LinearB",
         "email": "support@linearb.io"

plugins/agentic-advisor/.claude-plugin/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "agentic-advisor",
-  "version": "1.1.0",
+  "version": "1.1.1",
   "description": "Before writing code, grades how fragile the target is from LinearB health signals + local git history (rework, incidents, fix/revert density, code ownership) and holds the agent to a matching effort level (LOW/MEDIUM/HIGH) — lean on calm repos, defensive on fragile ones. Auto-triggers via a bundled hook and can report each effort decision to your LinearB org as usage telemetry.",
   "author": {
     "name": "LinearB",

plugins/agentic-advisor/README.md

@@ -40,7 +40,7 @@ Bundled hooks can report each effort decision to LinearB's reported-metrics API
 
 | event | when | `value` | token tag |
 | --- | --- | --- | --- |
-| **decision** (`phase=decision`) | on `Stop`, right after the verdict — once per repo + effort level per session | `1` / `2` / `3` = LOW / MEDIUM / HIGH | `grading_tokens` — output tokens spent producing the verdict |
+| **decision** (`phase=decision`) | on `Stop`, right after the verdict (or at `SessionEnd`, tagged `backfill: session_end`, if that turn was interrupted) — once per repo + effort level per session | `1` / `2` / `3` = LOW / MEDIUM / HIGH | `grading_tokens` — output tokens spent producing the verdict |
 | **tokens** (`phase=tokens`) | on `SessionEnd`, best-effort | `0` (not a grade — exclude from grade aggregates) | `coding_tokens` — output tokens spent on the work after the verdict |
 
 Count adoption and grades from **`phase=decision`** rows; use **`phase=tokens`** rows only for the coding-token measurement. (If you opt into the `LINEARB_BASELINE_HOLDOUT_PCT` experiment, held-out sessions also send a `value=0`, `label=baseline` event with `coding_tokens`; it's off by default.)

plugins/agentic-advisor/hooks/agentic-advisor-report.sh

@@ -8,8 +8,9 @@
 #     session that never ends cleanly (closed terminal, Ctrl-C, crash). coding_tokens
 #     is intentionally omitted here: it isn't complete until the session is over.
 #
-#   SessionEnd (once, best-effort) -> coding_tokens.
-#     graded   -> a TOKENS event (value 0, effort_level kept, phase=tokens) carrying
+#   SessionEnd (once, best-effort) -> backfill + coding_tokens.
+#     graded   -> first, any DECISION Stop never sent (an interrupted turn never fires
+#                 Stop), tagged backfill=session_end; then a TOKENS event (value 0, effort_level kept, phase=tokens) carrying
 #                 the whole-session coding_tokens for the efficiency experiment.
 #     baseline -> a beacon (value 0, label=baseline) carrying coding_tokens; the skill
 #                 was withheld so there is no verdict to report earlier.
@@ -110,7 +111,6 @@ if [ "$label" = "graded" ]; then
     vcount="$(grep -cE 'LinearB: .+ (LOW|MEDIUM|HIGH) effort' "$transcript" 2>/dev/null)"; vcount="${vcount:-0}"
     vseen="$marker_dir/${session_id:-default}.vseen"
     [ -f "$vseen" ] && [ "$(cat "$vseen" 2>/dev/null)" = "$vcount" ] && exit 0
-    printf '%s' "$vcount" > "$vseen" 2>/dev/null || true
   fi
 fi
 
@@ -293,8 +293,10 @@ fi
 
 # ---- graded path ----
 # (GATE 1 — skill actually ran — was already checked in the cheap fast-exit block above.)
-# Don't lock in a repo-less event (needs repo_url to join to the PR).
+# Don't lock in a repo-less event (needs repo_url to join to the PR). Mark this verdict
+# count seen only after that, so a Stop before the repo resolves retries on the next Stop.
 [ -n "$repo_url" ] || exit 0
+[ "$event" = "Stop" ] && [ -n "${vseen:-}" ] && { printf '%s' "$vcount" > "$vseen" 2>/dev/null || true; }
 
 # GATE 2 — a real assistant-authored verdict line exists, as text or as the hold's
 # `printf` Bash call (grep, not python, so it still works without python3).
@@ -316,9 +318,11 @@ parse_verdict() { # $1 = line
 }
 effort_val() { case "$1" in LOW) echo 1 ;; MEDIUM) echo 2 ;; HIGH) echo 3 ;; *) echo 0 ;; esac; }
 
-if [ "$event" = "Stop" ]; then
-  # DECISION event(s): report the grade the moment it exists — reliable, per turn,
-  # once per (repo|effort). No coding_tokens (not complete until the session ends).
+# DECISION event(s), once per (repo|effort) per session. Sent on Stop the moment the grade
+# exists; SessionEnd re-runs it to backfill any grade Stop missed (e.g. an interrupted turn
+# never fires Stop). No coding_tokens here (not complete until the session ends).
+report_decisions() {
+  local backfill=""; [ "$event" = "SessionEnd" ] && backfill="session_end"
   while IFS= read -r line; do
     [ -n "$line" ] || continue
     parse_verdict "$line" || continue
@@ -331,7 +335,8 @@ if [ "$event" = "Stop" ]; then
       --arg effort "$veffort" --arg repo "$vrepo" --arg ev "$vev" \
       --arg email "$email" --arg repourl "$repo_url" --arg branch "$branch" \
       --arg actual "$effort_actual" --arg ticket "$ticket" --arg sname "$session_name" \
-      --arg gdur "$grading_duration" --arg model "$model" --arg gtok "$grading_tokens" --arg pver "$plugin_version" --arg idsrc "$identity_source" '
+      --arg gdur "$grading_duration" --arg model "$model" --arg gtok "$grading_tokens" --arg pver "$plugin_version" --arg idsrc "$identity_source" \
+      --arg backfill "$backfill" '
       {
         metric_name: $m, source: $src, timestamp: $ts, value: $val,
         entity: (
@@ -348,15 +353,18 @@ if [ "$event" = "Stop" ]; then
           (if $gdur   != "" then {grading_duration_s: $gdur}    else {} end) +
           (if $model  != "" then {model: $model}                else {} end) +
           (if $gtok   != "" then {grading_tokens: $gtok}        else {} end) +
-          (if $pver   != "" then {plugin_version: $pver}        else {} end)
+          (if $pver   != "" then {plugin_version: $pver}        else {} end) +
+          (if $backfill != "" then {backfill: $backfill}        else {} end)
         )
       }' 2>/dev/null)"
     [ -n "$payload" ] || continue
     post "$payload"
     printf '%s\n' "$key" >> "$state" 2>/dev/null || true
   done <<< "$verdicts"
-  exit 0
-fi
+}
+
+report_decisions
+[ "$event" = "Stop" ] && exit 0
 
 # event == SessionEnd, graded: TOKENS event — the whole-session coding_tokens, attributed
 # to the final (last) verdict. value 0 so it doesn't double-count on grade charts; the