elanous 0.2.32 → 0.2.34

package/graphs/implement-loop.yaml

@@ -485,6 +485,12 @@ edges:
     to: decompose
 
   # ⭐ 실행 칸(substrate)이 «Pod 쪽 세 노드»를 켜고 끈다 — local 은 곧장 구현.
+  # ARG L1 r1: 생산 경로가 낸 substrate=pod 관측을 같은 조건 파서로 평가한다.
+  # 아래 기존 map 은 조건 미관측·모호성 때의 안전한 기본 경로다.
+  - from: decompose
+    when: substrate == pod
+    to: pool-admit
+
   - from: decompose
     on: substrate
     map:

package/graphs/journey/SCENARIOS.md

@@ -0,0 +1,33 @@
+# TA 여정 수용 시나리오
+
+## S — 현재
+
+`ta-journey@1`은 `done` 경로를 dry-run으로 통과한다. `entry`·`dispatch-pod`·`author`·`pr` 단위는 기존 하니스 관측을 래핑하며, 실제 실행을 시작하지 않는다. `harvest`는 전역 `verb-harvest@1` 단위를 호출하고, 입력의 `context.observations`에 기존 경로가 남긴 노드별 `outcome`이 있어야 판정한다.
+
+## C — 확인할 경로
+
+그래프를 통과하는 실제 TA 카드가 아직 없어 정상 경로 하나만으로는 각 분기 배선을 확인할 수 없다.
+
+## Q — 무엇을 입증해야 하나
+
+행복 경로와 주요 보류·재작업 출구가 다음 노드 또는 이름 있는 종단으로 닫히며, `--test --dry-run` 실행의 외부 명령 수가 0인가?
+
+## A — 시나리오 표
+
+| 시나리오 | 주입한 단위 출구 | 기대 경로 끝 | 기대 상태·reason |
+|---|---|---|---|
+| 행복 경로 | 모든 단위 `done` | `remember → done` | `done` |
+| 실제 harvest 미리보기 | 전역 `verb-harvest@1`의 `dry_run_command`가 `done` | `harvest → land` | `done` |
+| harvest 완료 출구 계약 | 전역 `verb-harvest@1`에 격리 `printf` fixture가 `done` 반환 | `harvest → done` | `done` |
+| harvest 재작업 출구 계약 | 전역 `verb-harvest@1`에 격리 `printf` fixture가 `rework` 반환 | `harvest → rework` | `failed · rework` |
+| harvest 사람 확인 출구 계약 | 전역 `verb-harvest@1`에 격리 `printf` fixture가 `needs-human` 반환 | `harvest → needs-human` | `failed · needs-human` |
+| 열차 등록 거부 | `land → needs-human` | `land → needs-human` | `failed · needs-human` |
+| 반영 대기·상한 도달 | `reflect → wait` 반복, `prior.attempts` 상한 | `reflect → needs-human` | `failed · needs-human` |
+| 녹색 판정 잔여 | `green → rework` | `green → rework` | `failed · rework` |
+| 직접 입구 | `entry`: `intake=direct`, `launch-gate=pass` | `entry → author` | `done` |
+| 대기열 입구 | `entry`: `intake=queued`, `queue=waiting` | `entry → needs-human` | `failed · needs-human` |
+| 증거 부족 | 단위에 필요한 관측 하나 이상 없음 | `<unit> → needs-human` | `failed · missing-observation` |
+
+reflect의 `wait`는 메인 그래프에서 reflect로 재진입하고 `max_visits: 3`으로 제한한다. 격리 `--test` 시험은 reflect@1 계약과 runner의 호출 경로에서 fixture provider가 `result.attempts`를 반환하고 다음 호출이 `prior.result.attempts`를 읽어 1→2→3으로 진행한 뒤 `needs-human`으로 닫히는 것을 확인한다. harvest 분기는 실제 전역 단위 선언을 복사하고 격리된 `printf` 명령 결과를 `done`·`rework`·`needs-human`으로 주입해 실제 단위 실행 경로와 라우팅을 확인한다. 나머지 분기 fixture는 선언된 출구 순서만 바꾼 `--dry-run`이며 `executed=0`이어야 한다. 모든 fixture는 임시 설정과 복사한 단위 파일을 사용한다. `needs-human`과 `rework`는 실행 상태 `failed`로 매핑하면서 원어 `reason`을 보존한다.
+
+래퍼는 `passJourneyNode`를 통해 기존 노드와 같은 entry/exit 사건을 기록한다. 관측 누락이나 성공·보류를 판정할 수 없는 출구는 `needs-human`이며, 기존 사건이 `failed`·`error`이면 누락된 후속 단계가 있어도 `failed`다. 실제 하니스 실행은 기존 실행 경로의 몫이고, 래퍼는 실행을 대신하지 않는다. `green`·`remember`는 아직 fixture 단위이며 dry-run만 수행한다. 기본 `verb-harvest@1` dry-run은 reviewer를 호출하지 않으며 실제 PR 리뷰·수확을 주장하지 않는다.

package/graphs/journey/recipes.yaml

@@ -0,0 +1 @@
+{}

package/graphs/journey/ta-journey.yaml

@@ -0,0 +1,84 @@
+graph_id: ta-journey
+version: 1
+launcher: harness-say
+entry_node: entry
+terminal_nodes: [done, failed, needs-human, rework]
+nodes:
+  - node_id: entry
+    kind: subgraph
+    recipe: none
+    call: entry@1
+    max_visits: 1
+    budget: { inner_steps: 10 }
+    contract: { inputs: [ask, context, prior], tools: harness-say, outputs: [result, evidence] }
+  - node_id: author
+    kind: subgraph
+    recipe: none
+    call: author@1
+    max_visits: 1
+    budget: { inner_steps: 10 }
+    contract: { inputs: [ask, context, prior], tools: harness-say, outputs: [result, evidence] }
+  - node_id: dispatch-pod
+    kind: subgraph
+    recipe: none
+    call: dispatch-pod@1
+    max_visits: 1
+    budget: { inner_steps: 10 }
+    contract: { inputs: [ask, context, prior], tools: harness-say, outputs: [result, evidence] }
+  - node_id: pr
+    kind: subgraph
+    recipe: none
+    call: pr@1
+    max_visits: 1
+    budget: { inner_steps: 10 }
+    contract: { inputs: [ask, context, prior], tools: harness-say, outputs: [result, evidence] }
+  - node_id: harvest
+    kind: subgraph
+    recipe: none
+    call: verb-harvest@1
+    max_visits: 1
+    budget: { inner_steps: 10 }
+    contract: { inputs: [ask, context, prior], tools: harness-say, outputs: [result, evidence] }
+  - node_id: land
+    kind: subgraph
+    recipe: none
+    call: land@1
+    max_visits: 1
+    budget: { inner_steps: 10 }
+    contract: { inputs: [ask, context, prior], tools: harness-say, outputs: [result, evidence] }
+  - node_id: reflect
+    kind: subgraph
+    recipe: none
+    call: reflect@1
+    max_visits: 3
+    budget: { inner_steps: 10 }
+    contract: { inputs: [ask, context, prior], tools: harness-say, outputs: [result, evidence] }
+  - node_id: green
+    kind: subgraph
+    recipe: none
+    call: green@1
+    max_visits: 1
+    budget: { inner_steps: 10 }
+    contract: { inputs: [ask, context, prior], tools: harness-say, outputs: [result, evidence] }
+  - node_id: remember
+    kind: subgraph
+    recipe: none
+    call: remember@1
+    max_visits: 1
+    budget: { inner_steps: 10 }
+    contract: { inputs: [ask, context, prior], tools: harness-say, outputs: [result, evidence] }
+  - { node_id: done, kind: gate, recipe: none, max_visits: 1 }
+  - { node_id: failed, kind: gate, recipe: none, max_visits: 1 }
+  - { node_id: needs-human, kind: gate, recipe: none, max_visits: 1 }
+  - { node_id: rework, kind: gate, recipe: none, max_visits: 1 }
+edges:
+  - { from: entry, on: outcome, map: { done: author, rework: rework, needs-human: needs-human, failed: failed } }
+  - { from: author, on: outcome, map: { done: dispatch-pod, rework: rework, needs-human: needs-human, failed: failed } }
+  - { from: dispatch-pod, on: outcome, map: { done: pr, rework: rework, needs-human: needs-human, failed: failed } }
+  - { from: pr, on: outcome, map: { done: harvest, rework: rework, needs-human: needs-human, failed: failed } }
+  - { from: harvest, on: outcome, map: { done: land, rework: rework, needs-human: needs-human, failed: failed } }
+  - { from: land, on: outcome, map: { done: reflect, rework: rework, needs-human: needs-human, failed: failed } }
+  - { from: reflect, on: outcome, map: { done: green, wait: reflect, needs-human: needs-human } }
+  - { from: green, on: outcome, map: { done: remember, rework: rework, needs-human: needs-human, failed: failed } }
+  - { from: remember, on: outcome, map: { done: done, rework: rework, needs-human: needs-human, failed: failed } }
+

package/graphs/journey/units/author/author.yaml

@@ -0,0 +1,20 @@
+graph_id: author
+version: 1
+unit:
+  inputs: [ask, context, prior]
+  outputs: [result, evidence]
+  exits: [done, rework, needs-human, failed]
+entry_node: observe
+terminal_nodes: [done, rework, needs-human, failed]
+nodes:
+  - node_id: observe
+    kind: agent
+    recipe: 'cmd:observe'
+    max_visits: 1
+    contract: { inputs: [ask, context, prior], tools: read-only, outputs: [result, evidence] }
+  - { node_id: done, kind: gate, recipe: none, max_visits: 1 }
+  - { node_id: rework, kind: gate, recipe: none, max_visits: 1 }
+  - { node_id: needs-human, kind: gate, recipe: none, max_visits: 1 }
+  - { node_id: failed, kind: gate, recipe: none, max_visits: 1 }
+edges:
+  - { from: observe, on: outcome, map: { done: done, rework: rework, needs-human: needs-human, failed: failed } }

package/graphs/journey/units/author/recipes.yaml

@@ -0,0 +1,2 @@
+observe:
+  command: 'bun "$ELANOUS_GRAPH_DIR/../../../../scripts/graph-units/ta-journey-observe.ts" author'

package/graphs/journey/units/green/green.yaml

@@ -0,0 +1,20 @@
+graph_id: green
+version: 1
+unit:
+  inputs: [ask, context, prior]
+  outputs: [result, evidence]
+  exits: [done, rework, needs-human, failed]
+entry_node: stage
+terminal_nodes: [done, rework, needs-human, failed]
+nodes:
+  - node_id: stage
+    kind: agent
+    recipe: 'cmd:stage'
+    max_visits: 1
+    contract: { inputs: [ask, context, prior], tools: read-only, outputs: [result, evidence] }
+  - { node_id: done, kind: gate, recipe: none, max_visits: 1 }
+  - { node_id: rework, kind: gate, recipe: none, max_visits: 1 }
+  - { node_id: needs-human, kind: gate, recipe: none, max_visits: 1 }
+  - { node_id: failed, kind: gate, recipe: none, max_visits: 1 }
+edges:
+  - { from: stage, on: outcome, map: { done: done, rework: rework, needs-human: needs-human, failed: failed, fail: failed } }

package/graphs/journey/units/green/recipes.yaml

@@ -0,0 +1,2 @@
+stage:
+  command: 'bun "$ELANOUS_GRAPH_DIR/../../../../scripts/graph-units/journey-stub.ts" green'

package/graphs/journey/units/pr/pr.yaml

@@ -0,0 +1,20 @@
+graph_id: pr
+version: 1
+unit:
+  inputs: [ask, context, prior]
+  outputs: [result, evidence]
+  exits: [done, rework, needs-human, failed]
+entry_node: observe
+terminal_nodes: [done, rework, needs-human, failed]
+nodes:
+  - node_id: observe
+    kind: agent
+    recipe: 'cmd:observe'
+    max_visits: 1
+    contract: { inputs: [ask, context, prior], tools: read-only, outputs: [result, evidence] }
+  - { node_id: done, kind: gate, recipe: none, max_visits: 1 }
+  - { node_id: rework, kind: gate, recipe: none, max_visits: 1 }
+  - { node_id: needs-human, kind: gate, recipe: none, max_visits: 1 }
+  - { node_id: failed, kind: gate, recipe: none, max_visits: 1 }
+edges:
+  - { from: observe, on: outcome, map: { done: done, rework: rework, needs-human: needs-human, failed: failed } }

package/graphs/journey/units/pr/recipes.yaml

@@ -0,0 +1,2 @@
+observe:
+  command: 'bun "$ELANOUS_GRAPH_DIR/../../../../scripts/graph-units/ta-journey-observe.ts" pr'

package/graphs/journey/units/remember/recipes.yaml

@@ -0,0 +1,2 @@
+stage:
+  command: 'bun "$ELANOUS_GRAPH_DIR/../../../../scripts/graph-units/journey-stub.ts" remember'

package/graphs/journey/units/remember/remember.yaml

@@ -0,0 +1,20 @@
+graph_id: remember
+version: 1
+unit:
+  inputs: [ask, context, prior]
+  outputs: [result, evidence]
+  exits: [done, rework, needs-human, failed]
+entry_node: stage
+terminal_nodes: [done, rework, needs-human, failed]
+nodes:
+  - node_id: stage
+    kind: agent
+    recipe: 'cmd:stage'
+    max_visits: 1
+    contract: { inputs: [ask, context, prior], tools: read-only, outputs: [result, evidence] }
+  - { node_id: done, kind: gate, recipe: none, max_visits: 1 }
+  - { node_id: rework, kind: gate, recipe: none, max_visits: 1 }
+  - { node_id: needs-human, kind: gate, recipe: none, max_visits: 1 }
+  - { node_id: failed, kind: gate, recipe: none, max_visits: 1 }
+edges:
+  - { from: stage, on: outcome, map: { done: done, rework: rework, needs-human: needs-human, failed: failed, fail: failed } }

package/graphs/operate-loop.yaml

@@ -13,8 +13,8 @@
 graph_id: operate-loop
 version: 1
 entry_node: operate-plan
-terminal_nodes: [merge, operate-blocked]
-runner_inputs: [goal, worktree, branch, review-intent, head-sha, base-branch]
+terminal_nodes: [landed, operate-blocked, needs-human]
+runner_inputs: [goal, worktree, branch, review-intent, head-sha, base-branch, ask, context, prior]
 docs_only_gate_skip: false
 state: [spec, changed_files, review_summary]
 
@@ -129,6 +129,17 @@ nodes:
       tools: gh-pr-merge
       outputs: [merged]
 
+  - node_id: recover
+    kind: subgraph
+    recipe: none
+    max_visits: 1
+    call: recover@1
+    budget: { inner_steps: 4 }
+    contract: { inputs: [ask, context, prior], tools: read-only, outputs: [result, evidence] }
+
+  - { node_id: landed, kind: gate, recipe: none, max_visits: 1 }
+  - { node_id: needs-human, kind: gate, recipe: none, max_visits: 1 }
+
 edges:
   - from: operate-plan
     to: operate-verify
@@ -171,3 +182,11 @@ edges:
 
   - from: open-pr
     to: merge
+
+  - from: merge
+    on: outcome
+    map: { merged: landed, land-refused: recover, ok: needs-human, fail: needs-human }
+
+  - from: recover
+    on: outcome
+    map: { done: needs-human, rework: needs-human, needs-human: needs-human, failed: needs-human }

package/graphs/overlays/stop-heal-ladder.yaml

@@ -5,7 +5,7 @@
 #   stop_class == main-sync          → rebase-regate
 #   stop_class == pod-died           → resume-from-salvage
 #   stop_class == evidence-uncovered → derive-evidence-rejudge
-# ⛔ 이 넷 밖의 갈래(review-repeat · review-out-of-scope · fabric · launch-failed · unclassified)는 선언이 «없다» —
+# ⛔ 이 넷 밖의 갈래(review-repeat · review-out-of-scope · scope-escape · fabric · launch-failed · unclassified)는 선언이 «없다» —
 #   처방은 `needs-owner` 이고 그것은 코드 기본(`AUTOHEAL_ACTION_BY_CLASS`)이 말한다.
 #
 # 🔑 노드 id 가 곧 처방 이름이다. `src/self-implement/stop-autoheal.ts` 는 그 갈래에 얹힌 오버레이가 «더한 노드 id» 로

package/graphs/release-rehearsal/recipes.yaml

@@ -9,4 +9,4 @@ tick:
     const result=JSON.parse(stdout);
     if (result.ok !== true) throw new Error(result.error ?? "rehearsal tick failed");
     console.log(JSON.stringify({outcome:"ok",tick:result}));'
-  timeout_ms: 5400000
+  timeout_ms: 18000000

package/graphs/release-rehearsal/release-rehearsal.yaml

@@ -1,7 +1,7 @@
 graph_id: release-rehearsal
 loop:
-  title: 판 컷 전 가벼운 RC 리허설
-  description: 판 일정의 컷 −7h · −150m 단계를 15분마다 살펴서 단계당 한 번만 실행
+  title: 판 컷 전 RC·하드 게이트 리허설
+  description: 판 일정의 컷 −12h 하드(기본 간격) · −7h RC · −5h 하드 · −150m RC를 15분마다 살펴서 단계당 한 번만 실행
   owner: OP
   trigger:
     cron: '*/15 * * * *'

package/graphs/research-loop.yaml

@@ -10,8 +10,8 @@
 graph_id: research-loop
 version: 5
 entry_node: investigate
-terminal_nodes: [merge]
-runner_inputs: [goal, worktree, branch, review-intent, head-sha, base-branch]
+terminal_nodes: [landed, needs-human]
+runner_inputs: [goal, worktree, branch, review-intent, head-sha, base-branch, ask, context, prior]
 # ⛔ 키 이름은 파서(`src/self-implement/graph-yaml.ts:201`)가 아는 것뿐이다 — 다른 이름은 «조용히» 무시된다.
 docs_only_gate_skip: true
 
@@ -109,6 +109,18 @@ nodes:
       tools: gh-pr-merge
       outputs: [merged]
 
+  # Only a PR landing refusal supplies the pinned context required by recover@1.
+  - node_id: recover
+    kind: subgraph
+    recipe: none
+    max_visits: 1
+    call: recover@1
+    budget: { inner_steps: 4 }
+    contract: { inputs: [ask, context, prior], tools: read-only, outputs: [result, evidence] }
+
+  - { node_id: landed, kind: gate, recipe: none, max_visits: 1 }
+  - { node_id: needs-human, kind: gate, recipe: none, max_visits: 1 }
+
 edges:
   # ⛔⭐ 유일하게 implement-loop 과 «다른» 자리 — gate 를 «건너뛸 수 있다».
   #   조건은 코드가 낸다: 변경 파일에 «코드»가 있나(routeGate). 못 세면 «돈다»(fail-safe).
@@ -150,3 +162,12 @@ edges:
 
   - from: open-pr
     to: merge
+
+  - from: merge
+    on: outcome
+    map: { merged: landed, land-refused: recover, ok: needs-human, fail: needs-human }
+
+  # The train/TA owns retry effects; the unit only classifies the refusal.
+  - from: recover
+    on: outcome
+    map: { done: needs-human, rework: needs-human, needs-human: needs-human, failed: needs-human }

package/graphs/units/dispatch-pod/dispatch-pod.yaml

@@ -0,0 +1,50 @@
+graph_id: dispatch-pod
+version: 1
+unit:
+  inputs: [ask, context, prior]
+  outputs: [result, evidence]
+  exits: [done, rework, needs-human, failed]
+entry_node: pool-admit
+terminal_nodes: [done, rework, needs-human, failed]
+nodes:
+  - node_id: pool-admit
+    group: dispatch-pod
+    journey_stage: dispatch
+    kind: gate
+    recipe: 'cmd:pod-pool-admission'
+    max_visits: 4
+    terminal_stages: []
+    contract:
+      inputs: [substrate, memory_tier]
+      tools: read-only
+      outputs: [pool_member, result, evidence]
+  - node_id: image-sync
+    group: dispatch-pod
+    journey_stage: dispatch
+    kind: gate
+    recipe: 'cmd:pod-pool-image-sync'
+    max_visits: 4
+    terminal_stages: []
+    contract:
+      inputs: [pool_member]
+      tools: read-only
+      outputs: [image_current, result, evidence]
+  - node_id: credential-refresh
+    group: dispatch-pod
+    journey_stage: dispatch
+    kind: observe
+    recipe: 'cmd:pod-credential-prerefresh'
+    max_visits: 4
+    terminal_stages: []
+    contract:
+      inputs: [pool_member]
+      tools: read-only
+      outputs: [usable_accounts, result, evidence]
+  - { node_id: done, kind: gate, recipe: none, max_visits: 1 }
+  - { node_id: rework, kind: gate, recipe: none, max_visits: 1 }
+  - { node_id: needs-human, kind: hitl, recipe: none, max_visits: 1 }
+  - { node_id: failed, kind: observe, recipe: none, max_visits: 1 }
+edges:
+  - { from: pool-admit, on: outcome, map: { done: image-sync, rework: rework, needs-human: needs-human, failed: failed } }
+  - { from: image-sync, on: outcome, map: { done: credential-refresh, rework: rework, needs-human: needs-human, failed: failed } }
+  - { from: credential-refresh, on: outcome, map: { done: done, rework: rework, needs-human: needs-human, failed: failed } }

package/graphs/units/dispatch-pod/recipes.yaml

@@ -0,0 +1,14 @@
+# These are observations only. No pod admission, image update, or credential
+# refresh is performed by the unit.
+pod-pool-admission:
+  command: >-
+    bun -e 'const args=["harness","queue","backpressure","--json"],p=Bun.spawnSync(["eln",...args]),raw=p.stdout.toString();let parsed;try{parsed=JSON.parse(raw)}catch{};const allowed=p.exitCode===0&&Number.isFinite(parsed?.launchBudget)&&parsed.launchBudget>0;const observed=parsed&&typeof parsed==="object"?{launchBudget:parsed.launchBudget??null,reasons:parsed.reasons??[],signals:Object.fromEntries(Object.entries(parsed.signals??{}).filter(([key])=>["queueDepth","admitWaiting","admitStarvedOver30m","podRecommended","staleParents"].includes(key)))}:{lineCount:raw.split("\n").filter(Boolean).length};console.log(JSON.stringify({outcome:p.exitCode!==0?"failed":allowed?"done":"needs-human",pool_member:{admission:observed},result:{observed},evidence:{command:["eln",...args].join(" "),exitCode:p.exitCode,launchBudget:parsed?.launchBudget??null}}))'
+  timeout_ms: 30000
+pod-pool-image-sync:
+  command: >-
+    bun -e 'const path=require("node:path"),exe=Bun.which("eln"),link=exe?Bun.spawnSync(["readlink",path.resolve(path.dirname(exe),"..","current")]):null,args=["config","get","harness.imageCommit"],p=Bun.spawnSync(["eln",...args]),pin=p.stdout.toString().trim(),installed=link?.stdout.toString().trim()??"",installedPin=path.basename(installed).split("-").at(-1)??"";const ok=p.exitCode===0&&link?.exitCode===0&&!!pin&&!!installed&&installedPin===pin;console.log(JSON.stringify({outcome:p.exitCode!==0||link?.exitCode!==0?"failed":ok?"done":"needs-human",image_current:{configPin:pin||null,installedPath:installed||null,installedPin:installedPin||null},result:{configPin:pin||null,installedPath:installed||null,installedPin:installedPin||null},evidence:{command:["eln",...args].join(" "),readlinkExit:link?.exitCode??null,configExit:p.exitCode,reason:ok?undefined:"installed current link and configured image pin do not exactly match or could not be observed"}}))'
+  timeout_ms: 30000
+pod-credential-prerefresh:
+  command: >-
+    bun -e 'const args=["provider","codex","status","--json"],p=Bun.spawnSync(["eln",...args]),raw=p.stdout.toString();let parsed;try{parsed=JSON.parse(raw)}catch{};const current=parsed&&typeof parsed==="object"&&parsed.current&&typeof parsed.current==="object"?{signalFresh:parsed.current.signalFresh??null,signalAgeMinutes:parsed.current.signalAgeMinutes??null,hasCredits:parsed.current.hasCredits??null}:null;console.log(JSON.stringify({outcome:"needs-human",usable_accounts:{status:current,expiry:"not exposed by a read-only command"},result:{status:"needs-human",reason:"credential expiry cannot be verified without refresh"},evidence:{command:["eln",...args].join(" "),exitCode:p.exitCode,observedKeys:parsed&&typeof parsed==="object"?Object.keys(parsed):[]}}))'
+  timeout_ms: 30000

package/graphs/units/entry/entry.yaml

@@ -0,0 +1,50 @@
+graph_id: entry
+version: 1
+unit:
+  inputs: [ask, context, prior]
+  outputs: [result, evidence]
+  exits: [done, rework, needs-human, failed]
+entry_node: intake
+terminal_nodes: [done, rework, needs-human, failed]
+nodes:
+  - node_id: intake
+    group: entry
+    journey_stage: entry
+    kind: observe
+    recipe: 'cmd:harness-entrance'
+    max_visits: 1
+    terminal_stages: []
+    contract:
+      inputs: [ask, context, prior]
+      tools: read-only
+      outputs: [entrance, journey_key, result, evidence]
+  - node_id: queue
+    group: entry
+    journey_stage: entry
+    kind: gate
+    recipe: 'cmd:harness-queue'
+    max_visits: 1
+    terminal_stages: []
+    contract:
+      inputs: [entrance, seat]
+      tools: read-only
+      outputs: [queue_item, queue_outcome, result, evidence]
+  - node_id: launch-gate
+    group: entry
+    journey_stage: entry
+    kind: gate
+    recipe: 'cmd:pre-launch-gate'
+    max_visits: 1
+    terminal_stages: []
+    contract:
+      inputs: [goal_id, budget]
+      tools: read-only
+      outputs: [launch_decision, result, evidence]
+  - { node_id: done, kind: gate, recipe: none, max_visits: 1 }
+  - { node_id: rework, kind: gate, recipe: none, max_visits: 1 }
+  - { node_id: needs-human, kind: hitl, recipe: none, max_visits: 1 }
+  - { node_id: failed, kind: observe, recipe: none, max_visits: 1 }
+edges:
+  - { from: intake, on: outcome, map: { done: queue, rework: rework, needs-human: needs-human, failed: failed } }
+  - { from: queue, on: outcome, map: { done: launch-gate, rework: rework, needs-human: needs-human, failed: failed } }
+  - { from: launch-gate, on: outcome, map: { done: done, rework: rework, needs-human: needs-human, failed: failed } }

package/graphs/units/entry/recipes.yaml

@@ -0,0 +1,14 @@
+# Read-only observations of the existing entry group. The command result is
+# wrapped in the unit contract so callers retain both result and evidence.
+harness-entrance:
+  command: >-
+    bun -e 'const args=["harness","doors","--json"],p=Bun.spawnSync(["eln",...args]),raw=p.stdout.toString();let parsed;try{parsed=JSON.parse(raw)}catch{};const observed=parsed&&typeof parsed==="object"?{keys:Object.keys(parsed),rows:Array.isArray(parsed.rows)?parsed.rows.length:undefined}:{lineCount:raw.split("\n").filter(Boolean).length};const result={command:["eln",...args].join(" "),observed};console.log(JSON.stringify({outcome:p.exitCode===0?"done":"failed",entrance:observed,journey_key:process.env.ELANOUS_RUN_ID??null,result,evidence:{exitCode:p.exitCode}}))'
+  timeout_ms: 30000
+harness-queue:
+  command: >-
+    bun -e 'const args=["harness","queue","list"],p=Bun.spawnSync(["eln",...args]),raw=p.stdout.toString(),observed={lineCount:raw.split("\n").filter(Boolean).length};console.log(JSON.stringify({outcome:p.exitCode===0?"done":"failed",queue_item:observed,queue_outcome:p.exitCode===0?"read":"unavailable",result:{command:["eln",...args].join(" "),observed},evidence:{exitCode:p.exitCode}}))'
+  timeout_ms: 30000
+pre-launch-gate:
+  command: >-
+    bun -e 'const args=["harness","budget","--json"],p=Bun.spawnSync(["eln",...args]),raw=p.stdout.toString();let parsed;try{parsed=JSON.parse(raw)}catch{};const allowed=p.exitCode===0&&parsed?.outcome==="proceed";const observed=parsed&&typeof parsed==="object"?{outcome:parsed.outcome??null,keys:Object.keys(parsed)}:{lineCount:raw.split("\n").filter(Boolean).length};console.log(JSON.stringify({outcome:p.exitCode!==0?"failed":allowed?"done":"needs-human",launch_decision:observed,result:{command:["eln",...args].join(" "),observed},evidence:{exitCode:p.exitCode,decision:parsed?.outcome??null}}))'
+  timeout_ms: 30000

package/graphs/units/harvest/harvest.yaml

@@ -0,0 +1,25 @@
+graph_id: verb-harvest
+version: 1
+entry_node: review
+terminal_nodes: [done, rework, needs-human, failed]
+unit:
+  inputs: [ask, context, prior]
+  outputs: [result, evidence]
+  exits: [done, rework, needs-human, failed]
+nodes:
+  - node_id: review
+    kind: judge
+    recipe: 'cmd:harvest-review'
+    max_visits: 1
+    contract:
+      inputs: [ask, context, prior]
+      tools: read-only
+      outputs: [outcome, result, evidence]
+  - { node_id: done, kind: gate, recipe: none, max_visits: 1 }
+  - { node_id: rework, kind: gate, recipe: none, max_visits: 1 }
+  - { node_id: needs-human, kind: hitl, recipe: none, max_visits: 1 }
+  - { node_id: failed, kind: observe, recipe: none, max_visits: 1 }
+edges:
+  - from: review
+    on: outcome
+    map: { done: done, rework: rework, needs-human: needs-human, failed: failed }

package/graphs/units/harvest/recipes.yaml

@@ -0,0 +1,4 @@
+harvest-review:
+  command: 'bun scripts/graph-harvest-unit.ts'
+  dry_run_command: 'bun scripts/graph-harvest-unit.ts --dry-run'
+  timeout_ms: 330000

package/graphs/units/land/land.yaml

@@ -0,0 +1,45 @@
+graph_id: land
+version: 1
+unit:
+  inputs: [ask, context, prior]
+  outputs: [result, evidence]
+  exits: [done, rework, needs-human, failed]
+policy:
+  express:
+    by: [queueFlag, predecessorOf, multiplierTier, p0InCutWindow2h]
+    waitsForBatch: false
+    directive: "2026-10-19 · prioritize queued P0s, predecessors, multiplier tier, and near-cut work"
+  batch:
+    min: 2
+    max: 4
+    groupBy: [area, cellPrefix]
+    maxWaitMin: 30
+    order: [express, predecessor, enqueued]
+    directive: "2026-10-14 · group by area and cell prefix, with batch size 2–4 and a 30-minute wait"
+  scenarioPrebuild:
+    source: [src/ux-sim, scenarios/ux-sim, fake-llm, fake-harness]
+    mock: required
+    directive: "2026-10-19 · prebuild scenarios with UX simulation, fake LLM, and fake harness inputs"
+  gate:
+    oncePerBatch: true
+    directive: "2026-10-19 · run one gate per batch"
+  land:
+    perPr: squash
+    pinnedHead: true
+    ejectOnFail: true
+    integrationPr: false
+    directive: "2026-10-19 · squash each PR at its pinned head and eject failures"
+entry_node: enqueue
+terminal_nodes: [done, rework, needs-human, failed]
+nodes:
+  - node_id: enqueue
+    kind: agent
+    recipe: 'cmd:land-enqueue'
+    max_visits: 1
+    contract: { inputs: [ask, context, prior], tools: external-write, outputs: [result, evidence] }
+  - { node_id: done, kind: gate, recipe: none, max_visits: 1 }
+  - { node_id: rework, kind: gate, recipe: none, max_visits: 1 }
+  - { node_id: needs-human, kind: hitl, recipe: none, max_visits: 1 }
+  - { node_id: failed, kind: observe, recipe: none, max_visits: 1 }
+edges:
+  - { from: enqueue, on: outcome, map: { done: done, rework: rework, needs-human: needs-human, failed: failed } }

package/graphs/units/land/recipes.yaml

@@ -0,0 +1,3 @@
+land-enqueue:
+  command: 'bun scripts/graph-units/land-enqueue.ts'
+  timeout_ms: 60000

package/graphs/units/recipes.yaml

@@ -1,3 +1,10 @@
 plan:
   command: 'bun scripts/release-loop/gate-recheck-plan-node.ts'
   timeout_ms: 60000
+recover-plan:
+  command: 'bun scripts/graph-units/recover-plan.ts'
+  timeout_ms: 30000
+harvest-review:
+  command: 'bun scripts/graph-harvest-unit.ts'
+  dry_run_command: 'bun scripts/graph-harvest-unit.ts --dry-run'
+  timeout_ms: 330000

package/graphs/units/recover.yaml

@@ -0,0 +1,39 @@
+graph_id: recover
+version: 1
+unit:
+  inputs: [ask, context, prior]
+  outputs: [result, evidence]
+  exits: [done, rework, needs-human, failed]
+  entries: [plan, request-review, wait-retry, human-review]
+entry_node: plan
+terminal_nodes: [done, rework, needs-human, failed]
+nodes:
+  - node_id: plan
+    kind: agent
+    recipe: 'cmd:recover-plan'
+    max_visits: 1
+    contract: { inputs: [ask, context, prior], tools: read-only, outputs: [result, evidence] }
+  - node_id: request-review
+    kind: agent
+    recipe: 'cmd:recover-plan'
+    max_visits: 1
+    contract: { inputs: [ask, context, prior], tools: read-only, outputs: [result, evidence] }
+  - node_id: wait-retry
+    kind: agent
+    recipe: 'cmd:recover-plan'
+    max_visits: 1
+    contract: { inputs: [ask, context, prior], tools: read-only, outputs: [result, evidence] }
+  - node_id: human-review
+    kind: agent
+    recipe: 'cmd:recover-plan'
+    max_visits: 1
+    contract: { inputs: [ask, context, prior], tools: read-only, outputs: [result, evidence] }
+  - { node_id: done, kind: gate, recipe: none, max_visits: 1 }
+  - { node_id: rework, kind: gate, recipe: none, max_visits: 1 }
+  - { node_id: needs-human, kind: gate, recipe: none, max_visits: 1 }
+  - { node_id: failed, kind: gate, recipe: none, max_visits: 1 }
+edges:
+  - { from: plan, on: outcome, map: { done: done, rework: rework, needs-human: needs-human, failed: failed, fail: failed } }
+  - { from: request-review, on: outcome, map: { done: done, rework: rework, needs-human: needs-human, failed: failed, fail: failed } }
+  - { from: wait-retry, on: outcome, map: { done: done, rework: rework, needs-human: needs-human, failed: failed, fail: failed } }
+  - { from: human-review, on: outcome, map: { done: done, rework: rework, needs-human: needs-human, failed: failed, fail: failed } }

package/graphs/units/reflect/recipes.yaml

@@ -0,0 +1,3 @@
+reflect:
+  command: 'bun scripts/graph-units/reflect-node.ts'
+  timeout_ms: 30000

package/graphs/units/reflect/reflect.yaml

@@ -0,0 +1,15 @@
+graph_id: reflect
+version: 1
+unit:
+  inputs: [ask, context, prior]
+  outputs: [result, evidence]
+  exits: [done, wait, needs-human]
+entry_node: reflect
+terminal_nodes: [done, wait, needs-human]
+nodes:
+  - { node_id: reflect, kind: agent, recipe: 'cmd:reflect', max_visits: 1 }
+  - { node_id: done, kind: gate, recipe: none, max_visits: 1 }
+  - { node_id: wait, kind: gate, recipe: none, max_visits: 1 }
+  - { node_id: needs-human, kind: gate, recipe: none, max_visits: 1 }
+edges:
+  - { from: reflect, on: outcome, map: { done: done, wait: wait, needs-human: needs-human } }