feat: live status-board comment + full-fleet dogfood (#1)

Phase 3: one consolidated, live-updating PR comment aggregating every model's per-lens progress (queued -> running -> finished + verdict), so the swarm's progress is visible at a glance and a watcher can tell when it's done. Opt-in statusWriter in the binary (atomic writes) + a background status-board.sh renderer wired through entrypoint.sh; default on, GADFLY_STATUS_BOARD=0 to disable. Also restores gadfly's dogfood swarm to the full cloud fleet (9 cloud + M5; M1 dropped as too slow) matching mort, and folds in the 3 real bugs the swarm found on its own PR (skip-binary stuck-waiting, panic-stuck lens, busy-loop on bad poll interval). All 36 findings graded via the gadfly MCP (18 real / 18 false-positive). gofmt clean, go vet quiet, go build + go test -race green. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Co-authored-by: Steve Dudenhoeffer <steve@stevedudenhoeffer.com> Co-committed-by: Steve Dudenhoeffer <steve@stevedudenhoeffer.com>
2026-06-27 19:00:12 +00:00
parent 0ad5b66170
commit c3d09d3bd4
11 changed files with 498 additions and 17 deletions
@@ -4,8 +4,8 @@
 # caches :latest, and this build is what carries foreman provider-type support)
 # as a specialist swarm and posts
 # ONE consolidated review comment as gitea-actions. Advisory only — never blocks a
-# merge. Gadfly reviewing its OWN PRs — dogfooding, full fleet (3 cloud + the M1/M5
-# Macs), copied from mort's setup.
+# merge. Gadfly reviewing its OWN PRs — dogfooding, full cloud fleet (9 cloud +
+# the M5 Mac; M1 dropped as too slow), copied from mort's setup.

 name: Adversarial Review (Gadfly)

@@ -41,8 +41,8 @@ jobs:
              || github.actor == 'fizi'
              || github.actor == 'dazed'))
    runs-on: ubuntu-latest
-    # Full fleet (3 cloud + 2 local Macs, all running concurrently) reviewing
-    # every PR with the 3-lens suite — the slow local lanes dominate wall time.
+    # Fleet (9 cloud + 1 local Mac/M5, all running concurrently) reviewing
+    # every PR with the 3-lens suite — the slow local lane dominates wall time.
    timeout-minutes: 90
    steps:
      - uses: docker://gitea.stevedudenhoeffer.com/steve/gadfly:sha-d7f364d
@@ -50,10 +50,12 @@ jobs:
          GITEA_API: ${{ github.server_url }}/api/v1/repos/${{ github.repository }}
          GITEA_TOKEN: ${{ secrets.GITEA_TOKEN }}
          OLLAMA_CLOUD_API_KEY: ${{ secrets.OLLAMA_CLOUD_API_KEY }}
-          # Local Macs, reached through their foreman queues (native Ollama on the
+          # Local Mac (M5), reached through its foreman queue (native Ollama on the
          # wire). Gadfly's GADFLY_ENDPOINT_* form with the "foreman" provider
-          # type: GADFLY_ENDPOINT_M1 registers provider "m1", _M5 registers "m5",
-          # each building a foreman-preset Ollama client at the given URL. Values
+          # type: GADFLY_ENDPOINT_M5 registers provider "m5", building a
+          # foreman-preset Ollama client at the given URL. (M1 is dropped from
+          # gadfly's swarm — too slow/low-signal — so its endpoint isn't mapped.)
+          # Values
          # (host + token) live in gitea secrets, each of the form:
          #   foreman|https://<foreman-host>|<token>
          # (converted from the komodo LLM_* DSNs foreman://<token>@<host>).
@@ -64,12 +66,16 @@ jobs:
          # NOTE: the Mac behind each foreman must still be awake/reachable; if a
          # box is offline, that model's comment shows an error and the others
          # still post. (Gitea secrets aren't auto-exposed — map each explicitly.)
-          GADFLY_ENDPOINT_M1: ${{ secrets.GADFLY_ENDPOINT_M1 }}
          GADFLY_ENDPOINT_M5: ${{ secrets.GADFLY_ENDPOINT_M5 }}
-          # 3 cloud (parallel) + M1 Pro + M5 Max — one consolidated comment each.
-          GADFLY_MODELS: "minimax-m3:cloud,deepseek-v4-flash:cloud,glm-5.2:cloud,m1/qwen3:14b,m5/qwen3.6:35b-mlx"
-          # cloud runs 3 at once; each Mac one at a time; all three lanes parallel.
-          GADFLY_PROVIDER_CONCURRENCY: "ollama-cloud=3,m1=1,m5=1"
+          # Fleet: 9 cloud (3 at a time) + M5 Max — one consolidated comment each.
+          # Matches mort's cloud set so the model-quality scoreboard is comparable
+          # across both repos. NOTE: M1 Pro is intentionally dropped here (too slow
+          # / low-signal for gadfly's own PRs); mort still runs it.
+          GADFLY_MODELS: "minimax-m3:cloud,glm-5.2:cloud,glm-5.1:cloud,kimi-k2.7-code:cloud,deepseek-v4-pro:cloud,nemotron-3-super:cloud,gpt-oss:120b-cloud,qwen3-coder:480b-cloud,gemma4:cloud,m5/qwen3.6:35b-mlx"
+          # cloud runs 3 at once; the Mac one at a time; both lanes parallel.
+          GADFLY_PROVIDER_CONCURRENCY: "ollama-cloud=3,m5=1"
+          # 3 cloud models x 3 lenses = 9 concurrent ollama-cloud queries (under the 10 budget).
+          GADFLY_PROVIDER_LENS_CONCURRENCY: "ollama-cloud=3"
          # Default => the 3-lens suite (security, correctness, error-handling).
          # Set the repo var GADFLY_SPECIALISTS to override (csv / "all" / "auto").
          GADFLY_SPECIALISTS: ${{ vars.GADFLY_SPECIALISTS || 'security,correctness,error-handling' }}
@@ -39,6 +39,7 @@ cmd/gadfly/            the reviewer binary — pure producer of review markdown
  recheck.go           second-pass verification prompt + verdict recompute
  *_test.go            sandbox, recheck, wrap-up, spec/endpoint-parse, specialist-resolution tests
 scripts/run.sh         fetch PR diff+meta, run the binary, upsert ONE labeled PR comment
+scripts/status-board.sh    render+upsert ONE live status-board comment (per-model/per-lens progress)
 scripts/system-prompt.txt   the reviewer persona + verification discipline (generic, not repo-specific)
 entrypoint.sh          container brains: trigger gating, PR clone, model loop (the logic that
                       used to live in workflow YAML)
@@ -28,5 +28,5 @@ RUN apk add --no-cache bash git curl jq ca-certificates
 COPY --from=build /out/gadfly /usr/local/bin/gadfly
 COPY scripts /app/scripts
 COPY entrypoint.sh /entrypoint.sh
-RUN chmod +x /entrypoint.sh /app/scripts/run.sh /usr/local/bin/gadfly
+RUN chmod +x /entrypoint.sh /app/scripts/run.sh /app/scripts/status-board.sh /usr/local/bin/gadfly
 ENTRYPOINT ["/entrypoint.sh"]
@@ -192,6 +192,30 @@ GADFLY_PROVIDER_LENS_CONCURRENCY: "ollama-cloud=3,m1=1"
 GADFLY_SPECIALISTS: "security,correctness,error-handling"
 ```

+### Live status board
+
+When several models (each with several lenses) review a PR, the individual findings land in
+**one comment per model** — but while that's in flight all you'd see is a row of
+`⏳ Reviewing…` placeholders. So Gadfly also upserts **one consolidated status-board comment**
+that aggregates every model's per-lens progress as it happens:
+
+```
+## 🪰 Gadfly — live review status
+1/3 reviewers finished · updated 2026-06-27 18:14:56Z
+
+#### `glm-5.2:cloud` · ollama-cloud — ⏳ 2/4 lenses
+- ✅ security — No material issues found
+- 🔄 correctness — running
+- ⏸️ performance — queued
+…
+```
+
+Each model process publishes its lenses (queued → running → finished + verdict) to a small
+JSON file, and a background renderer in `entrypoint.sh` re-renders + upserts the single comment
+every `GADFLY_STATUS_POLL_SECS` (default 12s) until the swarm finishes. It's advisory and
+best-effort — the per-model findings comments are unaffected — and entirely separate from those.
+Turn it off with `GADFLY_STATUS_BOARD=0`.
+
 ### Triggers

 1. A **new/reopened/ready** non-draft PR — automatic.
@@ -217,6 +241,7 @@ fixes. This keeps usage down.)
 ```
 cmd/gadfly/            the agentic reviewer binary (majordomo + Ollama Cloud); zero deps beyond stdlib + majordomo
 scripts/run.sh         fetches the PR diff, runs the reviewer, upserts one labeled comment
+scripts/status-board.sh    renders + upserts the single live status-board comment (per-lens progress)
 scripts/system-prompt.txt  the reviewer persona + verification discipline
 entrypoint.sh          the container brains: trigger gating, clone, model loop (logic lives here, not in YAML)
 Dockerfile             multi-stage; build-time module creds (BuildKit secrets) never reach the final image
@@ -252,6 +277,8 @@ The reviewer binary reads these (the stub/entrypoint set sane defaults):
 | `GADFLY_RECHECK` | on | set `0`/`false` to skip the recheck pass |
 | `GADFLY_RECHECK_MAX_STEPS` | 16 | recheck-pass step cap |
 | `GADFLY_MAX_DIFF_CHARS` | 60000 | diff chars embedded in the prompt (full diff via `get_diff`) |
+| `GADFLY_STATUS_BOARD` | on | set `0` to disable the live status-board comment |
+| `GADFLY_STATUS_POLL_SECS` | 12 | how often the status board re-renders/upserts |
 | `GADFLY_TRIGGER_PHRASE` | `@gadfly review` | comment phrase that re-triggers |
 | `GADFLY_ALLOWED_USERS` | *(collaborators)* | comma-separated allow-list for comment triggers |
 | `GADFLY_FINDINGS_URL` | — | gadfly-reports store base URL; set to enable findings telemetry (off when empty) |
@@ -218,6 +218,11 @@ func run() error {
 func runSpecialists(mdl llm.Model, fsTools *repoFS, base string, specialists []Specialist, task, diff string) []specialistResult {
 	results := make([]specialistResult, len(specialists))

+	// Optional live status board: publishes this model's per-lens progress to a
+	// file the entrypoint board renders. Inert (no-op) unless GADFLY_STATUS_FILE
+	// is set, so plain runs are unaffected.
+	sw := newStatusWriter(os.Getenv("GADFLY_MODEL"), modelProvider(), specialists)
+
 	conc := min(lensConcurrency(), len(specialists))

 	sem := make(chan struct{}, conc)
@@ -228,8 +233,21 @@ func runSpecialists(mdl llm.Model, fsTools *repoFS, base string, specialists []S
 		go func(i int, sp Specialist) {
 			defer wg.Done()
 			defer func() { <-sem }()
+			// A panic in one lens must not crash the whole binary (which would
+			// kill every other lens's output) or leave this lens stuck at
+			// "running" on the status board. Recover, record it as an errored
+			// result, and mark the lens finished so the board can complete.
+			defer func() {
+				if r := recover(); r != nil {
+					results[i] = specialistResult{spec: sp, out: fmt.Sprintf("⚠️ This reviewer panicked: %v", r), verdict: verdictUnknown, errored: true}
+					sw.set(sp.Name, lensFinished, "", true)
+				}
+			}()
+			sw.set(sp.Name, lensRunning, "", false)
 			out, errored := reviewWithSpecialist(mdl, fsTools, base, sp, task, diff)
-			results[i] = specialistResult{spec: sp, out: out, verdict: parseVerdict(out), errored: errored}
+			v := parseVerdict(out)
+			results[i] = specialistResult{spec: sp, out: out, verdict: v, errored: errored}
+			sw.set(sp.Name, lensFinished, v.label(), errored)
 		}(i, sp)
 	}
 	wg.Wait()
@@ -0,0 +1,131 @@
+package main
+
+import (
+	"encoding/json"
+	"os"
+	"path/filepath"
+	"strings"
+	"sync"
+	"time"
+)
+
+// Lens states for the live status board. A lens starts queued, becomes running
+// when its pass begins, and ends finished (with a verdict, or errored).
+const (
+	lensQueued   = "queued"
+	lensRunning  = "running"
+	lensFinished = "finished"
+)
+
+// lensStatus is one specialist lens's progress, as rendered by the entrypoint
+// status board (scripts/status-board.sh).
+type lensStatus struct {
+	Name    string `json:"name"`
+	State   string `json:"state"`             // queued | running | finished
+	Verdict string `json:"verdict,omitempty"` // set when finished (the lens's label)
+	Errored bool   `json:"errored,omitempty"` // the lens failed to complete
+}
+
+// modelStatus is the on-disk shape one model process publishes for the live
+// status board: a snapshot of this model's lenses as they progress. The board
+// reads every model's file and renders a single consolidated PR comment.
+type modelStatus struct {
+	Model    string       `json:"model"`
+	Provider string       `json:"provider"`
+	Started  int64        `json:"started"` // unix seconds
+	Updated  int64        `json:"updated"` // unix seconds, bumped on every change
+	Done     bool         `json:"done"`    // all lenses finished
+	Lenses   []lensStatus `json:"lenses"`
+}
+
+// statusWriter maintains a model's status file as its lenses progress. It is
+// purely opt-in: when GADFLY_STATUS_FILE is unset the writer's path is empty and
+// every method is a no-op, so a plain run (and the unit tests) never touch the
+// filesystem and behave exactly as before. Writes are atomic (temp file +
+// rename within the same dir) so the board never reads a half-written file even
+// though lenses can finish concurrently.
+type statusWriter struct {
+	path string
+	mu   sync.Mutex
+	st   modelStatus
+}
+
+// newStatusWriter seeds a writer with every lens queued and flushes the initial
+// snapshot. model/provider are echoed into the file so the board can render
+// them without re-deriving from the filename (which is sanitized). The status
+// file path comes from GADFLY_STATUS_FILE (set by run.sh per model); when empty
+// the writer is inert.
+func newStatusWriter(model, provider string, specialists []Specialist) *statusWriter {
+	w := &statusWriter{path: strings.TrimSpace(os.Getenv("GADFLY_STATUS_FILE"))}
+	w.st = modelStatus{
+		Model:    model,
+		Provider: provider,
+		Started:  time.Now().Unix(),
+	}
+	for _, sp := range specialists {
+		w.st.Lenses = append(w.st.Lenses, lensStatus{Name: sp.Name, State: lensQueued})
+	}
+	w.flush()
+	return w
+}
+
+// set transitions a lens to a new state (and verdict/errored when finished),
+// recomputes the overall done flag, and atomically rewrites the file. Unknown
+// lens names are ignored. Safe for concurrent callers (one goroutine per lens).
+func (w *statusWriter) set(name, state, verdict string, errored bool) {
+	if w == nil || w.path == "" {
+		return
+	}
+	w.mu.Lock()
+	defer w.mu.Unlock()
+	for i := range w.st.Lenses {
+		if w.st.Lenses[i].Name == name {
+			w.st.Lenses[i].State = state
+			w.st.Lenses[i].Verdict = verdict
+			w.st.Lenses[i].Errored = errored
+			break
+		}
+	}
+	done := true
+	for _, l := range w.st.Lenses {
+		if l.State != lensFinished {
+			done = false
+			break
+		}
+	}
+	w.st.Done = done
+	w.flush()
+}
+
+// flush writes the current snapshot atomically. Best-effort: any error is
+// swallowed (the status board is advisory and must never affect the review).
+func (w *statusWriter) flush() {
+	if w.path == "" {
+		return
+	}
+	w.st.Updated = time.Now().Unix()
+	data, err := json.MarshalIndent(&w.st, "", "  ")
+	if err != nil {
+		return
+	}
+	dir := filepath.Dir(w.path)
+	tmp, err := os.CreateTemp(dir, ".status-*.tmp")
+	if err != nil {
+		return
+	}
+	tmpName := tmp.Name()
+	if _, err := tmp.Write(data); err != nil {
+		tmp.Close()
+		os.Remove(tmpName)
+		return
+	}
+	if err := tmp.Close(); err != nil {
+		os.Remove(tmpName)
+		return
+	}
+	// Rename is atomic within the same filesystem, so the board reader sees
+	// either the old complete file or the new complete file — never a partial.
+	if err := os.Rename(tmpName, w.path); err != nil {
+		os.Remove(tmpName)
+	}
+}
@@ -0,0 +1,103 @@
+package main
+
+import (
+	"encoding/json"
+	"os"
+	"path/filepath"
+	"testing"
+)
+
+// readStatus loads a modelStatus written by the statusWriter.
+func readStatus(t *testing.T, path string) modelStatus {
+	t.Helper()
+	data, err := os.ReadFile(path)
+	if err != nil {
+		t.Fatalf("read status file: %v", err)
+	}
+	var st modelStatus
+	if err := json.Unmarshal(data, &st); err != nil {
+		t.Fatalf("unmarshal status: %v", err)
+	}
+	return st
+}
+
+func TestStatusWriterLifecycle(t *testing.T) {
+	path := filepath.Join(t.TempDir(), "glm.json")
+	t.Setenv("GADFLY_STATUS_FILE", path)
+
+	specs := []Specialist{
+		{Name: "security", Title: "Security"},
+		{Name: "correctness", Title: "Correctness"},
+	}
+	w := newStatusWriter("glm-5.2:cloud", "ollama-cloud", specs)
+
+	// Initial snapshot: both lenses queued, model not done, metadata echoed.
+	st := readStatus(t, path)
+	if st.Model != "glm-5.2:cloud" || st.Provider != "ollama-cloud" {
+		t.Fatalf("model/provider not echoed: %+v", st)
+	}
+	if len(st.Lenses) != 2 {
+		t.Fatalf("want 2 lenses, got %d", len(st.Lenses))
+	}
+	for _, l := range st.Lenses {
+		if l.State != lensQueued {
+			t.Fatalf("lens %q want queued, got %q", l.Name, l.State)
+		}
+	}
+	if st.Done {
+		t.Fatal("model marked done while lenses still queued")
+	}
+	if st.Started == 0 {
+		t.Fatal("started timestamp not set")
+	}
+
+	// Transition one lens through running -> finished; model not yet done.
+	w.set("security", lensRunning, "", false)
+	if got := readStatus(t, path); got.Lenses[0].State != lensRunning {
+		t.Fatalf("security want running, got %q", got.Lenses[0].State)
+	}
+	w.set("security", lensFinished, "No material issues found", false)
+	st = readStatus(t, path)
+	if st.Lenses[0].State != lensFinished || st.Lenses[0].Verdict != "No material issues found" {
+		t.Fatalf("security finish not recorded: %+v", st.Lenses[0])
+	}
+	if st.Done {
+		t.Fatal("model marked done with one lens still queued")
+	}
+
+	// Finish the second lens (errored) -> model done.
+	w.set("correctness", lensFinished, "Reviewed", true)
+	st = readStatus(t, path)
+	if !st.Done {
+		t.Fatal("model should be done after all lenses finished")
+	}
+	if !st.Lenses[1].Errored {
+		t.Fatal("errored flag not recorded for correctness")
+	}
+	if st.Updated < st.Started {
+		t.Fatalf("updated (%d) should be >= started (%d)", st.Updated, st.Started)
+	}
+}
+
+// With GADFLY_STATUS_FILE unset the writer is inert: no file, no panic.
+func TestStatusWriterDisabled(t *testing.T) {
+	t.Setenv("GADFLY_STATUS_FILE", "")
+	w := newStatusWriter("m", "p", []Specialist{{Name: "security"}})
+	w.set("security", lensFinished, "Minor issues", false)
+	// Nothing to assert beyond "did not panic / did not write" — a nil-safe set
+	// on the disabled writer is the contract.
+	if w.path != "" {
+		t.Fatalf("expected empty path when disabled, got %q", w.path)
+	}
+}
+
+// set must ignore unknown lens names rather than panic or append.
+func TestStatusWriterUnknownLens(t *testing.T) {
+	path := filepath.Join(t.TempDir(), "s.json")
+	t.Setenv("GADFLY_STATUS_FILE", path)
+	w := newStatusWriter("m", "p", []Specialist{{Name: "security"}})
+	w.set("does-not-exist", lensRunning, "", false)
+	if st := readStatus(t, path); len(st.Lenses) != 1 || st.Lenses[0].State != lensQueued {
+		t.Fatalf("unknown lens mutated state: %+v", st.Lenses)
+	}
+}
@@ -155,6 +155,12 @@ DEFAULT_CONC="${GADFLY_CONCURRENCY:-1}"

 provider_of() { case "$1" in */*) echo "${1%%/*}";; *) echo "${GADFLY_PROVIDER:-ollama-cloud}";; esac; }

+# Per-model status file path for the live board. The model id can contain '/'
+# and ':' (e.g. m1/qwen3:14b), so sanitize to a flat filename; the JSON inside
+# carries the real model/provider, so this just needs to be unique per model.
+STATUS_DIR="${WORKDIR}/status"
+status_file_for() { echo "${STATUS_DIR}/$(echo "$1" | tr -c '[:alnum:]._-' '_').json"; }
+
 provider_cap() { # provider -> concurrency (override map "p=N,...", else default)
  local p="$1" item k v
  IFS=',' read -ra _caps <<< "${GADFLY_PROVIDER_CONCURRENCY:-}"
@@ -167,8 +173,19 @@ provider_cap() { # provider -> concurrency (override map "p=N,...", else default
 }

 review_one() {
+  local sf=""
+  [ "${GADFLY_STATUS_BOARD:-1}" != "0" ] && sf="$(status_file_for "$1")"
  PROVIDER=ollama MODEL="$1" GADFLY_BIN="/usr/local/bin/gadfly" GADFLY_REPO_DIR="$REPO_DIR" \
+    GADFLY_STATUS_FILE="$sf" \
    bash "${SCRIPTS_DIR}/run.sh" || log "model $1 failed (continuing)"
+  # If the binary never wrote real status (run.sh skipped it: empty diff, no key,
+  # binary missing), the pre-seed stays {started:0, done:false} and the board
+  # would show this model "waiting to start" forever and never reach N/N. Mark
+  # such a never-started file done so the board can complete. The binary stamps a
+  # nonzero `started`, so that reliably distinguishes "ran" from "skipped".
+  if [ -n "$sf" ] && [ -f "$sf" ] && [ "$(jq -r '.started // 0' "$sf" 2>/dev/null)" = "0" ]; then
+    tmp="$(jq '.done = true' "$sf" 2>/dev/null)" && printf '%s' "$tmp" > "$sf"
+  fi
 }

 # Normalize the model list (trim, drop blanks) into MODEL_LIST.
@@ -197,10 +214,44 @@ run_lane() { # $1=provider: run its models, at most `cap` at a time
  wait
 }

+# --- live status board (optional, default on) ------------------------------
+# Each model process publishes per-lens progress to STATUS_DIR/<model>.json; a
+# background renderer (status-board.sh) upserts ONE consolidated PR comment so
+# progress across all models/lenses is visible at a glance — and a watcher can
+# tell when the whole swarm is finished. Advisory/best-effort; the per-model
+# findings still land in each model's own comment. Disable with
+# GADFLY_STATUS_BOARD=0.
+BOARD_PID=""
+if [ "${GADFLY_STATUS_BOARD:-1}" != "0" ]; then
+  rm -rf "$STATUS_DIR"; mkdir -p "$STATUS_DIR"
+  # Pre-seed every model as queued so the board shows the full swarm from t=0,
+  # even models still waiting on their provider lane's concurrency cap. Each
+  # binary overwrites its own file with real per-lens detail once it starts.
+  for m in "${MODEL_LIST[@]}"; do
+    jq -n --arg model "$m" --arg provider "$(provider_of "$m")" \
+      '{model:$model, provider:$provider, started:0, updated:0, done:false, lenses:[]}' \
+      > "$(status_file_for "$m")" 2>/dev/null || true
+  done
+  GITEA_API="$GITEA_API" GITEA_TOKEN="$GITEA_TOKEN" PR="$PR" GADFLY_STATUS_DIR="$STATUS_DIR" \
+    bash "${SCRIPTS_DIR}/status-board.sh" &
+  BOARD_PID=$!
+  log "status board started (pid ${BOARD_PID})"
+fi
+
 log "providers: ${PROVIDERS:-none}"
-# Each provider lane runs in parallel; cap is enforced within each lane.
+# Each provider lane runs in parallel; cap is enforced within each lane. Track
+# the lane PIDs so we wait ONLY for the review work — not the status board,
+# which intentionally runs until we signal it below.
+LANE_PIDS=()
 for p in $PROVIDERS; do
  run_lane "$p" &
+  LANE_PIDS+=("$!")
 done
-wait
+[ "${#LANE_PIDS[@]}" -gt 0 ] && wait "${LANE_PIDS[@]}"
+
+# Reviews are done: signal the board to render the final state once and exit.
+if [ -n "$BOARD_PID" ]; then
+  touch "${STATUS_DIR}/.done" 2>/dev/null || true
+  wait "$BOARD_PID" 2>/dev/null || true
+fi
 log "done"
@@ -65,6 +65,10 @@ jobs:
          #   GADFLY_PROVIDER_LENS_CONCURRENCY: "ollama-cloud=3,m1=1"
          # GADFLY_LENS_CONCURRENCY: ${{ vars.GADFLY_LENS_CONCURRENCY }}
          # GADFLY_PROVIDER_LENS_CONCURRENCY: ${{ vars.GADFLY_PROVIDER_LENS_CONCURRENCY }}
+          # Live status board (optional; ON by default): one consolidated comment
+          # showing every model's per-lens progress as it runs. Disable with
+          # GADFLY_STATUS_BOARD=0; tune the refresh with GADFLY_STATUS_POLL_SECS.
+          # GADFLY_STATUS_BOARD: ${{ vars.GADFLY_STATUS_BOARD }}
          # --- Models & providers (optional; default = Ollama Cloud) ----------
          # Gadfly is majordomo-powered, so it can target other backends. Set a
          # provider for bare model ids; point at a different endpoint with a
@@ -25,6 +25,8 @@
 #
 # Optional:
 #   MAX_DIFF_CHARS     diff truncation cap for the prompt (default 60000)
+#   GADFLY_STATUS_FILE per-model JSON path for the live status board (set by
+#                      entrypoint.sh; empty/unset disables status publishing)
 #
 # This script is advisory: it never fails the job for review content. It exits
 # non-zero only on a usage/configuration error.
@@ -161,6 +163,7 @@ case "$PROVIDER" in
          GADFLY_TITLE="$TITLE" \
          GADFLY_BODY="$BODY" \
          GADFLY_MAX_DIFF_CHARS="$MAX_DIFF_CHARS" \
+          GADFLY_STATUS_FILE="${GADFLY_STATUS_FILE:-}" \
          "$BIN" 2>"$ERR_FILE"
        )"
        rc=$?
@@ -0,0 +1,137 @@
+#!/usr/bin/env bash
+# Live status board for a Gadfly review.
+#
+# Each model process (the cmd/gadfly binary) publishes its per-lens progress to
+# $GADFLY_STATUS_DIR/<model>.json as lenses go queued -> running -> finished.
+# This script polls that directory and upserts ONE consolidated PR comment that
+# aggregates every model's per-lens status, so a human (or an agent watching the
+# PR) can see the whole swarm's progress at a glance and know when it's done —
+# instead of staring at N separate "⏳ Reviewing…" placeholders.
+#
+# It is advisory and best-effort: a failed render/post is logged and retried on
+# the next tick; nothing here can fail the review or block a merge. It runs in
+# the background from entrypoint.sh and exits once the $GADFLY_STATUS_DIR/.done
+# sentinel appears (the entrypoint touches it after all model lanes finish),
+# after one final render.
+#
+# Required env:
+#   GITEA_API         https://HOST/api/v1/repos/OWNER/REPO
+#   GITEA_TOKEN       token with repo write access (posts the comment)
+#   PR                pull request number
+#   GADFLY_STATUS_DIR directory holding the per-model <model>.json files
+# Optional:
+#   GADFLY_STATUS_POLL_SECS  render/upsert interval (default 12)
+set -uo pipefail
+
+: "${GITEA_API:?GITEA_API required}"
+: "${GITEA_TOKEN:?GITEA_TOKEN required}"
+: "${PR:?PR required}"
+: "${GADFLY_STATUS_DIR:?GADFLY_STATUS_DIR required}"
+
+POLL="${GADFLY_STATUS_POLL_SECS:-12}"
+# Guard against a non-numeric poll interval: with `set -uo pipefail` (no set -e)
+# a bad `sleep "$POLL"` would fail silently and the `while :` loop would spin,
+# hammering the Gitea API. Coerce anything non-integer (or <1) back to 12.
+case "$POLL" in ''|*[!0-9]*) POLL=12 ;; esac
+[ "$POLL" -ge 1 ] 2>/dev/null || POLL=12
+DONE_FILE="${GADFLY_STATUS_DIR}/.done"
+MARKER="<!-- gadfly-status-board -->"
+API_TIMEOUT="--connect-timeout 20 --max-time 30"
+BOARD_ID="" # cached comment id, so we PATCH in place instead of re-searching
+
+say() { echo "[gadfly-status-board] $*" >&2; }
+
+command -v jq >/dev/null 2>&1 || { say "jq not found; status board disabled"; exit 0; }
+
+# render_section FILE -> markdown for one model (its header + per-lens bullets).
+# Reads the JSON the binary writes; tolerates a half-written/missing file by
+# emitting nothing (jq exits non-zero -> caller skips it this tick).
+render_section() {
+  jq -r '
+    def icon(state; errored):
+      if state == "finished" then (if errored then "⚠️" else "✅" end)
+      elif state == "running" then "🔄"
+      else "⏸️" end;
+    def lensline:
+      "- " + icon(.state; (.errored // false)) + " **" + .name + "** — " +
+      ( if .state == "finished" then (if (.errored // false) then "could not complete" else (.verdict // "done") end)
+        elif .state == "running" then "running"
+        else "queued" end );
+    ( [.lenses[] | select(.state == "finished")] | length ) as $fin
+    | ( .lenses | length ) as $tot
+    | ( if .done then "✅ done"
+        elif $tot == 0 then "⏳ waiting to start"
+        else "⏳ " + ($fin|tostring) + "/" + ($tot|tostring) + " lenses" end ) as $sum
+    | "#### `" + .model + "` · " + .provider + " — " + $sum + "\n"
+    + ( if $tot == 0 then "- ⏸️ _no lenses reported yet_"
+        else ([.lenses[] | lensline] | join("\n")) end )
+  ' "$1" 2>/dev/null
+}
+
+# render_body -> the full consolidated comment body (marker + header + sections).
+render_body() {
+  local f sections="" total=0 done=0 ts
+  shopt -s nullglob
+  local files=("${GADFLY_STATUS_DIR}"/*.json)
+  shopt -u nullglob
+  for f in "${files[@]}"; do
+    local sec
+    sec="$(render_section "$f")" || continue
+    [ -z "$sec" ] && continue
+    total=$((total + 1))
+    if [ "$(jq -r 'if .done then 1 else 0 end' "$f" 2>/dev/null)" = "1" ]; then
+      done=$((done + 1))
+    fi
+    sections="${sections}${sec}"$'\n\n'
+  done
+  ts="$(date -u '+%Y-%m-%d %H:%M:%SZ')"
+  if [ "$total" -eq 0 ]; then
+    sections="_Waiting for reviewers to start…_"$'\n'
+  fi
+  printf '%s\n## 🪰 Gadfly — live review status\n\n%d/%d reviewers finished · updated %s\n\n%s\n<sub>Live status board. Findings are posted in each model'\''s own comment. Advisory only — does not block merge.</sub>' \
+    "$MARKER" "$done" "$total" "$ts" "$sections"
+}
+
+# find_existing -> id of the board comment if it already exists (paginate by
+# marker). Used once, to recover the comment across container restarts.
+find_existing() {
+  local page=1 cmts id
+  while [ "$page" -le 10 ]; do
+    cmts="$(curl $API_TIMEOUT -fsS -H "Authorization: token ${GITEA_TOKEN}" \
+      "${GITEA_API}/issues/${PR}/comments?limit=50&page=${page}" 2>/dev/null || echo '[]')"
+    [ "$(echo "$cmts" | jq 'length' 2>/dev/null || echo 0)" = "0" ] && break
+    id="$(echo "$cmts" | jq -r --arg m "$MARKER" \
+      '.[] | select(.body != null and (.body | startswith($m))) | .id' 2>/dev/null | head -n1)"
+    [ -n "$id" ] && { echo "$id"; return; }
+    page=$((page + 1))
+  done
+  echo ""
+}
+
+# upsert BODY — PATCH the cached/known board comment, else POST a new one and
+# cache its id. A failed PATCH (e.g. comment deleted) clears the cache so the
+# next tick re-discovers or re-creates it.
+upsert() {
+  local body="$1" post_body resp
+  post_body="$(jq -n --arg b "$body" '{body:$b}')"
+  [ -z "$BOARD_ID" ] && BOARD_ID="$(find_existing)"
+  if [ -n "$BOARD_ID" ]; then
+    if ! curl $API_TIMEOUT -fsS -X PATCH -H "Authorization: token ${GITEA_TOKEN}" -H "Content-Type: application/json" \
+      "${GITEA_API}/issues/comments/${BOARD_ID}" -d "$post_body" >/dev/null 2>&1; then
+      say "patch of comment ${BOARD_ID} failed; will re-discover"
+      BOARD_ID=""
+    fi
+  else
+    resp="$(curl $API_TIMEOUT -fsS -X POST -H "Authorization: token ${GITEA_TOKEN}" -H "Content-Type: application/json" \
+      "${GITEA_API}/issues/${PR}/comments" -d "$post_body" 2>/dev/null || echo '{}')"
+    BOARD_ID="$(echo "$resp" | jq -r '.id // ""' 2>/dev/null)"
+  fi
+}
+
+say "starting (poll ${POLL}s, dir ${GADFLY_STATUS_DIR})"
+while :; do
+  upsert "$(render_body)"
+  [ -f "$DONE_FILE" ] && break
+  sleep "$POLL"
+done
+say "done"