feat: implement issue #1008 — Onboard idea→initiative pipeline reusables into canary-rollout (planner/triage ready; enhancer/feature-ideation blocked)

scripts/canary-rollout.sh

@@ -187,16 +187,33 @@ _reusable_differs() {
 # unless state is BLOCKED (then REGRESSION | PRE_EXISTING).
 _frontier_state() {
   local agent="$1"
-  local cand chans frontier=""
+  local cand chans frontier="" prev_on=()
+  local soak_start use_soak_as_candidate=false next_health_signal
+  soak_start="$(_agent_field "$agent" soak_start_ring)"
+  [ "$soak_start" = "null" ] && soak_start=""
+  next_health_signal="$(_agent_field "$agent" next_tier_health_signal)"
+  [ "$next_health_signal" = "null" ] && next_health_signal=""
+
   cand="$(channel_commit "$agent" next)"
+  # soak_start_ring agents have no @next channel caller — the innermost soaked ring is
+  # the candidate source.  Fall back only when next genuinely has no tag.
+  if [ -z "$cand" ] && [ -n "$soak_start" ]; then
+    cand="$(channel_commit "$agent" "$soak_start")"
+    use_soak_as_candidate=true
+  fi
   chans="$(ordered_channels "$agent")"
 
+  # Collect the rings already on the candidate (starting at next / soak_start) and find the frontier.
   local chan_array=()
   IFS=, read -r -a chan_array <<< "$chans"
   local ch
   for ch in "${chan_array[@]}"; do
     local c; c="$(channel_commit "$agent" "$ch")"
-    if [ "$ch" = "next" ] || [ "$c" = "$cand" ]; then :; else frontier="$ch"; break; fi
+    if [ "$ch" = "next" ] || { [ "$use_soak_as_candidate" = "true" ] && [ "$ch" = "$soak_start" ]; } || [ "$c" = "$cand" ]; then
+      prev_on+=("$ch")
+    else
+      frontier="$ch"; break
+    fi
   done
   if [ -z "$frontier" ]; then
     echo "$cand - - COMPLETE 0 0 0 0 0 0 -"; return 0
@@ -207,18 +224,36 @@ _frontier_state() {
   source="${transition%%->*}"
   cut_z="$(candidate_cut_date "$agent" "$cand")"
   if [ -z "$cut_z" ]; then
-    # Cannot determine the per-candidate window start — fail closed to prevent unbounded history queries.
-    echo "$cand $frontier $transition BLOCKED 0 0 0 0 0 0 -"; return 0
+    if [ -z "$soak_start" ]; then
+      # Cannot determine the per-candidate window start — fail closed to prevent unbounded history queries.
+      echo "$cand $frontier $transition BLOCKED 0 0 0 0 0 0 -"; return 0
+    fi
+    # soak_start_ring agents may lack a vX.Y.Z release tag; proceed with an unbounded window.
   fi
   now_epoch="$(date -u +%s)"
 
   # Source-tier repos (the tier currently running the candidate).
-  local src_repos=() r
-  while IFS= read -r r; do [ -n "$r" ] && src_repos+=("$r"); done < <(resolve_members "$agent" "$source")
+  # When the frontier IS the soak_start_ring (transition is empty — no source ring below it),
+  # use the host repo with next_tier_health_signal as the pre-soak health source.
+  local src_repos=() r use_health_signal=false
+  if [ -z "$transition" ] && [ -n "$next_health_signal" ]; then
+    use_health_signal=true
+    src_repos=("$(_agent_field "$agent" host)")
+  else
+    while IFS= read -r r; do [ -n "$r" ] && src_repos+=("$r"); done < <(resolve_members "$agent" "$source")
+  fi
 
   # Sample on the source tier over the per-candidate window.
   local sample earliest
-  read -r sample earliest < <(_tier_sample "$agent" "$cut_z" "${src_repos[@]}")
+  if [ "$use_health_signal" = "true" ]; then
+    local hs_json
+    hs_json="$(_run_json "${src_repos[0]}" "$next_health_signal" "$cut_z")"
+    sample=$(jq '[.[]?|select(.conclusion=="success" or .conclusion=="failure")]|length' 2>/dev/null <<< "$hs_json" || echo 0)
+    earliest="$(jq -r '[.[]?|select(.conclusion=="success" or .conclusion=="failure")|.createdAt?]|min // empty' 2>/dev/null <<< "$hs_json")"
+    [ -z "$earliest" ] && earliest="-"
+  else
+    read -r sample earliest < <(_tier_sample "$agent" "$cut_z" "${src_repos[@]}")
+  fi
 
   # Dwell is always measured from the candidate's own cut (tagger date), per #548 spec.
   local dwell_h=0

standards/canary-rings.json

@@ -1,12 +1,18 @@
 {
   "version": 1,
-  "description": "Ring membership for agent canary rollout (initiative #495, issues #500/#501). The promotion automation (scripts/canary-rollout.sh) reads this as the single source of truth. `next` is host-relative: it resolves to the repo that HOSTS the reusable; `ring0` covers the other org-infra repo (the host already sits in `next`).",
+  "description": "Ring membership for agent canary rollout (initiative #495, issues #500/#501; pipeline onboarding #1008). The promotion automation (scripts/canary-rollout.sh) reads this as the single source of truth. `next` is host-relative: it resolves to the repo that HOSTS the reusable; `ring0` covers the other org-infra repo (the host already sits in `next`). An agent may set `soak_start_ring` to name the innermost ring that actually has a channel caller — used when the host dogfoods the reusable via a local `./` ref (so there is no `@next`/`ring0` channel caller) and no org-infra repo consumes it as a channel member; soak then begins at that ring. `unmanaged` records reusable-less / non-channel-promotable workflows (e.g. direct `gh workflow run` dispatch) so the automation does not try to gate them.",
   "org_infra_repos": ["petry-projects/.github", "petry-projects/.github-private"],
   "member_tokens": {
     "$host": "the agent's host repo (the repo that owns the reusable)",
     "$org_infra": "org_infra_repos minus the host (the host is already in `next`)",
     "*": "every other consumer not named in an earlier ring"
   },
+  "unmanaged": {
+    "initiative-driver": {
+      "reason": "Direct `gh workflow run` dispatch to petry-projects/.github-private; no reusable workflow, so it is not channel-promotable. Recorded here (out of model, #1008) so the canary automation does not attempt to gate it.",
+      "health_source": ".github/workflows/initiative-driver-canary.yml"
+    }
+  },
   "agents": {
     "dev-lead": {
       "host": "petry-projects/.github-private",
@@ -40,6 +46,28 @@
           }
         }
       }
+    },
+    "initiative-planner": {
+      "host": "petry-projects/.github",
+      "reusable": ".github/workflows/initiative-planner-reusable.yml",
+      "run_workflow": "Initiative Planner — Approval Trigger",
+      "soak_start_ring": "ring1",
+      "next_tier_health_signal": ".github/workflows/initiative-planner-canary.yml",
+      "rings": [
+        { "channel": "ring1",  "order": 2, "members": ["petry-projects/TalkTerm", "petry-projects/bmad-bgreat-suite"] },
+        { "channel": "stable", "order": 3, "members": ["*"] }
+      ]
+    },
+    "idea-triage": {
+      "host": "petry-projects/.github",
+      "reusable": ".github/workflows/idea-triage-reusable.yml",
+      "run_workflow": "Idea Triage — Weekly Shortlist",
+      "soak_start_ring": "ring1",
+      "next_tier_health_signal": ".github/workflows/idea-triage-canary.yml",
+      "rings": [
+        { "channel": "ring1",  "order": 2, "members": ["petry-projects/TalkTerm", "petry-projects/bmad-bgreat-suite"] },
+        { "channel": "stable", "order": 3, "members": ["*"] }
+      ]
     }
   }
 }

tests/canary_rollout.bats

@@ -168,6 +168,69 @@ setup() {
   [ "$status" -eq 0 ]
 }
 
+# ── canary-rings.json: idea→initiative pipeline onboarding (#1008) ─────────────
+# initiative-planner + idea-triage are onboarded as managed agents whose reusable
+# lives cross-repo in petry-projects/.github. Their host dogfoods via a local `./`
+# ref (no @next caller) and .github-private runs a full central copy (not a channel
+# consumer), so there is no next/ring0 channel caller — soak begins at ring1.
+_pipeline_agent_shape() {
+  # $1 = agent key. Asserts the shared onboarding shape for planner/triage.
+  local agent="$1"
+  # hosted in the public repo (reusable lives there, not in .github-private)
+  run jq -e --arg a "$agent" '.agents[$a].host == "petry-projects/.github"' "$RINGS"
+  [ "$status" -eq 0 ]
+  # soak starts at ring1 (owner-confirmed: no channel caller below ring1)
+  run jq -e --arg a "$agent" '.agents[$a].soak_start_ring == "ring1"' "$RINGS"
+  [ "$status" -eq 0 ]
+  # only ring1 + stable channels are modelled, ordered ring1 -> stable
+  run jq -r --arg a "$agent" '.agents[$a].rings | sort_by(.order) | map(.channel) | join(",")' "$RINGS"
+  [ "$output" = "ring1,stable" ]
+  # ring1 = the two low-traffic consumers
+  run jq -e --arg a "$agent" '.agents[$a].rings[] | select(.channel=="ring1") | (.members | index("petry-projects/TalkTerm")) and (.members | index("petry-projects/bmad-bgreat-suite"))' "$RINGS"
+  [ "$status" -eq 0 ]
+  # stable = ["*"] (un-enrolled today; populated by the promotion automation)
+  run jq -e --arg a "$agent" '.agents[$a].rings[] | select(.channel=="stable") | .members == ["*"]' "$RINGS"
+  [ "$status" -eq 0 ]
+  # .github-private runs a full central copy — never a channel member of these agents
+  run jq -e --arg a "$agent" '[.agents[$a].rings[].members[]] | index("petry-projects/.github-private") | not' "$RINGS"
+  [ "$status" -eq 0 ]
+}
+
+@test "canary-rings.json: initiative-planner onboarded (soak_start_ring=ring1, cross-repo host)" {
+  _pipeline_agent_shape "initiative-planner"
+  # its pre-ring1 health signal is the in-repo canary workflow
+  run jq -e '.agents["initiative-planner"].next_tier_health_signal == ".github/workflows/initiative-planner-canary.yml"' "$RINGS"
+  [ "$status" -eq 0 ]
+}
+
+@test "canary-rings.json: idea-triage onboarded (soak_start_ring=ring1, cross-repo host)" {
+  _pipeline_agent_shape "idea-triage"
+  # same pattern as initiative-planner: needs next_tier_health_signal for pre-ring1 gate
+  run jq -e '.agents["idea-triage"].next_tier_health_signal == ".github/workflows/idea-triage-canary.yml"' "$RINGS"
+  [ "$status" -eq 0 ]
+}
+
+@test "canary-rings.json: initiative-driver is recorded as unmanaged, not a channel agent" {
+  # no reusable -> not channel-promotable -> excluded from .agents so the
+  # automation never tries to gate it.
+  run jq -e '.agents["initiative-driver"] == null' "$RINGS"
+  [ "$status" -eq 0 ]
+  # but explicitly recorded under .unmanaged with a rationale + health source.
+  run jq -e '.unmanaged["initiative-driver"].reason | type == "string" and (. | length > 0)' "$RINGS"
+  [ "$status" -eq 0 ]
+  run jq -e '.unmanaged["initiative-driver"].health_source == ".github/workflows/initiative-driver-canary.yml"' "$RINGS"
+  [ "$status" -eq 0 ]
+}
+
+@test "canary-rings.json: blocked/hold reusables are NOT onboarded" {
+  # idea-enhancer (zero tags/callers) and feature-ideation (open bug #571 + missing
+  # channel tags) must stay out of the SoT entirely — neither agent nor unmanaged.
+  run jq -e '.agents["idea-enhancer"] == null and .agents["feature-ideation"] == null' "$RINGS"
+  [ "$status" -eq 0 ]
+  run jq -e '.unmanaged["idea-enhancer"] == null and .unmanaged["feature-ideation"] == null' "$RINGS"
+  [ "$status" -eq 0 ]
+}
+
 # ── orchestrator: resolve_members (host-relative tokens) ──────────────────────
 @test "orchestrator: resolve_members expands \$host / \$org_infra / * " {
   run bash -c "source '$ORCH' && CANARY_RINGS='$RINGS' resolve_members dev-lead next"
@@ -220,6 +283,74 @@ GHEOF
   [[ "$output" == *"stable"* ]]
 }
 
+@test "orchestrator: evaluate uses next_tier_health_signal when prev_on is empty (soak_start_ring agent)" {
+  _make_stub_bin
+  # initiative-planner/next is the candidate; ring1 is not yet on it → prev_on=[]
+  cat > "$STUB_BIN/git" <<'GITEOF'
+#!/usr/bin/env bash
+case "$*" in
+  *"rev-parse"*"initiative-planner/next"*)   echo "aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa" ;;
+  *"rev-parse"*"initiative-planner/ring1"*)  echo "bbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbb" ;;
+  *"rev-parse"*"initiative-planner/stable"*) echo "cccccccccccccccccccccccccccccccccccccccc" ;;
+  *) : ;;
+esac
+GITEOF
+  chmod +x "$STUB_BIN/git"
+  # Healthy runs from the host canary; ring member queries return empty (low traffic)
+  cat > "$STUB_BIN/gh" <<'GHEOF'
+#!/usr/bin/env bash
+case "$*" in
+  *"run list"*"initiative-planner-canary"*)
+    printf '['; for i in $(seq 1 15); do [ "$i" -gt 1 ] && printf ','; printf '{"conclusion":"success"}'; done; printf ']'
+    ;;
+  *"run list"*) echo "[]" ;;
+  *) echo "{}" ;;
+esac
+GHEOF
+  chmod +x "$STUB_BIN/gh"
+
+  run env CANARY_RINGS="$RINGS" bash "$ORCH" evaluate initiative-planner
+  [ "$status" -eq 0 ]
+  [[ "$output" == *"initiative-planner"* ]]
+  [[ "$output" == *"ring1"* ]]
+  # 15 healthy runs from canary, baseline=0 → ceil(0/7)=0 min needed → PROMOTE
+  [[ "$output" == *"PROMOTE"* ]]
+}
+
+@test "orchestrator: evaluate uses ring1 health when soak_start_ring is the candidate source (no next tag)" {
+  _make_stub_bin
+  # initiative-planner has soak_start_ring=ring1 and no @next channel caller.
+  # ring1 tag exists (aaaa = candidate); stable tag is on a different commit (bbbb).
+  # _frontier_state must treat ring1 as the candidate source and identify stable as
+  # the frontier (not get stuck or falsely report COMPLETE).
+  cat > "$STUB_BIN/git" <<'GITEOF'
+#!/usr/bin/env bash
+case "$*" in
+  *"rev-parse"*"initiative-planner/ring1"*)  echo "aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa" ;;
+  *"rev-parse"*"initiative-planner/stable"*) echo "bbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbb" ;;
+  *) : ;;
+esac
+GITEOF
+  chmod +x "$STUB_BIN/git"
+  # ring1 members return healthy runs; stable frontier members are wildcards (skipped)
+  cat > "$STUB_BIN/gh" <<'GHEOF'
+#!/usr/bin/env bash
+case "$*" in
+  *"run list"*) echo '[{"conclusion":"success"},{"conclusion":"success"},{"conclusion":"success"},{"conclusion":"success"},{"conclusion":"success"},{"conclusion":"success"},{"conclusion":"success"},{"conclusion":"success"},{"conclusion":"success"},{"conclusion":"success"}]' ;;
+  *) echo "{}" ;;
+esac
+GHEOF
+  chmod +x "$STUB_BIN/gh"
+
+  run env CANARY_RINGS="$RINGS" bash "$ORCH" evaluate initiative-planner
+  [ "$status" -eq 0 ]
+  [[ "$output" == *"initiative-planner"* ]]
+  # stable is the frontier (ring1 = soak_start_ring = already on candidate)
+  [[ "$output" == *"stable"* ]]
+  # 10+ healthy ring1 runs, baseline (stable/*) = 0 → min_healthy=0, healthy>0 → PROMOTE
+  [[ "$output" == *"PROMOTE"* ]]
+}
+
 @test "orchestrator: promote --override --dry-run shows the move but never pushes" {
   _make_stub_bin
   cat > "$STUB_BIN/gh" <<'GHEOF'