{"id":"1f2898da-3fde-419b-8898-4ae9976e1c1b","shortId":"Yb9HHW","kind":"skill","title":"workbench-goal-mode-v2","tagline":"Two-layer autonomous conductor — design loop produces decision packets, dispatch loop routes to bounded issues with dedupe/cooldown/archive controls. Replaces v1's single-agent persistence with a durable control loop that stops only at real blockers.","description":"# Workbench Goal Mode v2\n\nUse this skill when an issue contains `/goal`, `GOAL_MODE: yes`, `GOAL_MODE_V2`,\nor asks the workbench to complete an objective autonomously across multiple\nagents, turns, and evidence gates until a real blocker appears.\n\nGoal Mode v2 is a persistent conductor, not a single-agent persistence wrapper.\nIt splits the workflow into two cooperating layers so the human does not need to\nbabysit windows, monitor every agent, or manually switch contexts.\n\n## Session Communication Bootstrap\n\nAt conductor startup, load `docs/agent-communication-profile.md` and apply:\n\n```text\nApply communication profile docs/agent-communication-profile.md.\nTone: human, direct, bilingual, pushback-ok.\n```\n\nThis controls communication style only. Runtime identity, model, tools, and MCP\navailability still require live proof.\n\n## Architecture\n\n```mermaid\nflowchart TB\n  H[\"Operator shorthand / issue\"] --> GC[\"GOAL_CAPTURED\"]\n  GC --> DSN[\"DESIGNING\"]\n  DSN --> DP[\"DECISION_PACKET\"]\n  DP --> DSP[\"DISPATCHING\"]\n  DSP --> OBS[\"OBSERVING\"]\n  OBS -->|agent done| REV[\"REVIEWING\"]\n  OBS -->|new evidence| DP\n  REV -->|PASS| LRN[\"LEARNING / ARCHIVING\"]\n  REV -->|FLAG| DSP\n  REV -->|BLOCK| BC[\"BLOCKER_CLASSIFIED\"]\n  BC -->|operator| OP[\"OPERATOR_NEEDED\"]\n  BC -->|external| CD[\"COOLDOWN → retry\"]\n  BC -->|permanent| ARC[\"ARCHIVED\"]\n  LRN --> NG[\"NEXT_GOAL or DONE\"]\n  OP -->|human responds| DSN\n  CD -->|cooldown elapsed| DSN\n```\n\n### Layer 1: Design / Decision\n\nOwned by the conductor (Admin or delegated design agent). Produces decision\npackets, not raw execution spam.\n\nResponsibilities:\n- Refine operator intent into a scoped objective.\n- Pull constraints, non-goals, blockers, and evidence expectations from durable\n  memory (SYNTHESIS.md, DECISIONS.md, prior issue evidence).\n- Choose the Friction Tier and applicable gates (SDD, L2 Pressure, Capy check).\n- Maintain taste, scope, blocker semantics, and evidence standards.\n- Ask the operator only when a real design trade-off needs human judgment.\n\n### Layer 2: Dispatch / Operations\n\nOwned by the dispatcher (Admin or Supervisor). Converts decision packets into\nbounded Multica issues, monitors them, and routes follow-ups.\n\nResponsibilities:\n- Convert one decision packet into one or more bounded issues.\n- Assign one owner per issue. Mention specialists only for narrow review.\n- Monitor active / in_review / blocked / done states via `multica issue list`.\n- Re-route only when new evidence changes state — never re-sweep unchanged work.\n- Stop when only operator-call or external-platform blockers remain.\n\n## State Machine\n\n| State | Meaning | Next |\n|-------|---------|------|\n| `GOAL_CAPTURED` | Shorthand or issue received, not yet designed | `DESIGNING` |\n| `DESIGNING` | Conductor is pulling context, refining intent, checking memory | `DECISION_PACKET` or `OPERATOR_NEEDED` |\n| `DECISION_PACKET` | Scoped route produced, ready to dispatch | `DISPATCHING` |\n| `DISPATCHING` | One or more bounded issues created/assigned | `OBSERVING` |\n| `OBSERVING` | Waiting for agent evidence; periodic status scan | `REVIEWING`, `DECISION_PACKET` (re-route), or `BLOCKER_CLASSIFIED` |\n| `REVIEWING` | Evidence evaluated against closeout gates | `LEARNING` (PASS), `DISPATCHING` (FLAG), or `BLOCKER_CLASSIFIED` (BLOCK) |\n| `BLOCKER_CLASSIFIED` | Blocker typed: operator / external / permanent | `OPERATOR_NEEDED`, cooldown → `DESIGNING`, or `ARCHIVED` |\n| `LEARNING / ARCHIVING` | Evidence harvested, noise archived, memory updated | `NEXT_GOAL` or `DONE` |\n| `OPERATOR_NEEDED` | Waiting for human input | `DESIGNING` (on response) |\n| `DONE` | Objective achieved, all gates passed | terminal |\n\n## Decision Packet Format\n\nEvery decision packet must be posted as a structured comment before any issues\nare created:\n\n```text\nDECISION_PACKET\ngoal_id: <goal-shorthand-or-issue-id>\nintent: <one-sentence objective>\nroute: <chosen execution path — which agents, skills, lanes>\nowner: <primary executing agent>\nreviewer: <who verifies evidence>\nconstraints: <non-negotiables, boundaries, blocked lanes>\nevidence_expectations: <what proves success — exact artifacts or checks>\nnon_goals: <explicit exclusions>\nblocker_conditions: <what would block this route>\ntier: fast | standard | heavy\ndedupe_key: <canonical key to prevent duplicate dispatch>\nmax_active: <max concurrent in_progress issues for this goal>\ncooldown_minutes: <minimum minutes before next sweep or re-route>\nverdict: READY_TO_DISPATCH | NEEDS_DESIGN | OPERATOR_NEEDED\n```\n\n## Dedupe, Cooldown, and Noise Controls\n\nThese prevent the controller from accumulating endless sweeper issues (per\nDAS-741 / DAS-743 findings):\n\n### Dedupe\n- Before creating any issue, check if an equivalent issue already exists using\n  the `dedupe_key`. Match on title prefix + active status (not done/cancelled).\n- If an equivalent issue exists and is `in_progress` or `in_review`, do not\n  create a duplicate. Post a status note on the existing issue instead.\n- If an equivalent issue exists and is `blocked`, check if the blocker changed\n  before re-dispatching.\n\n### Cooldown\n- After dispatching a decision packet, set `cooldown_minutes`. Do not scan or\n  re-route for this goal until the cooldown elapses.\n- Default cooldown: 15 minutes for standard work, 30 minutes for heavy work.\n- Cooldown resets when a new decision packet is dispatched.\n\n### Max Active\n- No more than `max_active` (default: 5) issues may be `in_progress` for a\n  single goal at any time.\n- If the limit is reached, new dispatch waits until an active issue moves to\n  `in_review` or `done`.\n\n### Archive and Cancel Semantics\n- Completed (`done`) issues: harvest evidence, then archive — do not re-read or\n  re-route.\n- Stale `in_review` issues (no activity for 48h): post one nudge, then flag for\n  Supervisor.\n- Noise issues (created in error, duplicate, superseded): set to `cancelled`\n  with a one-line reason. Do not leave them in `blocked`.\n- The conductor itself must have a self-cancel condition: if the conductor issue\n  is the only remaining active issue and all dispatched work is `done` or\n  `cancelled`, post the final verdict and stop.\n\n## Autonomy Guards\n\nThe conductor may act autonomously when:\n- The decision packet is `READY_TO_DISPATCH`.\n- No operator-call conditions are triggered.\n- Dedupe confirms no duplicate work exists.\n- The route does not touch blocked lanes (DAS-696, DAS-693, or operator-declared\n  frozen lanes).\n- Workbench Max is not in the route.\n\nThe conductor MUST call the operator when:\n- A design trade-off needs human taste judgment (not just more context).\n- A permission, secret, payment, or runtime mutation is required.\n- A blocked lane is the only viable route.\n- Two prior attempts at the same sub-task failed with different approaches.\n- The dedupe key matches an existing active issue with a different decision\n  packet (route conflict).\n\n## Composition With Existing Layers\n\nGoal Mode v2 composes with, but does not replace:\n\n| Layer | How Goal Mode v2 uses it |\n|-------|--------------------------|\n| Friction Tier Router | Selected during DESIGNING; encoded in decision packet tier field |\n| SDD | Decision packet may route heavy work through SDD stages; design layer handles Raw Requirement + Product Design |\n| L2 Pressure Gate | Applied during DESIGNING when objective touches remote/HarnessMax/leaderboard work |\n| Goal Mode v1 | v2 replaces the persistence wrapper; v1 skill remains for simple single-agent /goal tasks |\n| Auto Review Sweeper | Dispatch layer uses sweeper results during OBSERVING → REVIEWING transition |\n| Self-Awareness | Run once at GOAL_CAPTURED for heavy-path goals before DESIGNING begins |\n| Completion Cooling | Applied to dispatched child issues at 75/85/90/100 thresholds |\n\n## Activation\n\nWhen activated by `/goal` or `GOAL_MODE: yes`, the conductor must:\n\n1. Apply `docs/agent-communication-profile.md` before the first user-facing\n   conductor note.\n2. Post `GOAL_LOCK` with objective, owner, non_goals, closeout gates, and\n   operator-call conditions.\n3. If Heavy Path: post `SELF_AWARENESS_BOOTSTRAP` before designing.\n4. If L2_PRESSURE: post `RV_PRESSURE_CHECK` before routing.\n5. Enter `DESIGNING` → produce `DECISION_PACKET` → enter `DISPATCHING`.\n6. Continue the observe → review → archive → next-goal loop until DONE or\n   OPERATOR_NEEDED.\n\n## Closeout Contract\n\n```text\nGOAL_MODE_V2_CLOSEOUT\ngoal_id:\nobjective:\nstate_machine_path: <trace of states visited, e.g. GOAL_CAPTURED→DESIGNING→...→DONE>\ndecision_packets_produced: <count>\nissues_dispatched: <count and IDs>\nevidence_harvested: <summary of completed evidence>\nnoise_cancelled: <count and IDs of cancelled duplicates>\noperator_calls: <count and reasons, if any>\nresidual_risk:\narchive_actions_taken:\nverdict: PASS | FLAG | BLOCK\n```\n\n## Validation\n\nBefore claiming DONE, verify:\n- Every dispatched issue has a verdict (PASS/FLAG/BLOCK) or is properly cancelled.\n- No duplicate active issues share the same dedupe_key.\n- The conductor's self-cancel condition is satisfied.\n- Evidence is posted on the goal issue, not scattered across child comments.\n- `OPERATOR_NEEDED` was raised for every case where the conductor lacked authority.","tags":["workbench","goal","mode","multica","ultimate","fearvox","agent-skills","ai-agents","claude-code","codex","hermes-agent","sdd"],"capabilities":["skill","source-fearvox","skill-workbench-goal-mode-v2","topic-agent-skills","topic-ai-agents","topic-claude-code","topic-codex","topic-hermes-agent","topic-multica","topic-sdd","topic-skills-sh"],"categories":["multica-ultimate-workbench"],"synonyms":[],"warnings":[],"endpointUrl":"https://skills.sh/Fearvox/multica-ultimate-workbench/workbench-goal-mode-v2","protocol":"skill","transport":"skills-sh","auth":{"type":"none","details":{"cli":"npx skills add Fearvox/multica-ultimate-workbench","source_repo":"https://github.com/Fearvox/multica-ultimate-workbench","install_from":"skills.sh"}},"qualityScore":"0.456","qualityRationale":"deterministic score 0.46 from registry signals: · indexed on github topic:agent-skills · 12 github stars · SKILL.md body (9,155 chars)","verified":false,"liveness":"unknown","lastLivenessCheck":null,"agentReviews":{"count":0,"score_avg":null,"cost_usd_avg":null,"success_rate":null,"latency_p50_ms":null,"narrative_summary":null,"summary_updated_at":null},"enrichmentModel":"deterministic:skill-github:v1","enrichmentVersion":1,"enrichedAt":"2026-05-18T19:07:13.504Z","embedding":null,"createdAt":"2026-05-13T01:05:02.725Z","updatedAt":"2026-05-18T19:07:13.504Z","lastSeenAt":"2026-05-18T19:07:13.504Z","tsv":"'-693':885 '-696':883 '-741':602 '-743':604 '/goal':54,1040,1084 '1':233,1092 '15':698 '2':312,1103 '3':1119 '30':703 '4':1129 '48h':783 '5':725,1139 '6':1147 '75/85/90/100':1078 'accumul':596 'achiev':509 'across':70,1252 'act':852 'action':1203 'activ':359,575,626,718,723,748,781,831,955,1080,1082,1227 'admin':240,319 'agent':30,72,93,115,183,244,445,543,1039 'alreadi':616 'appear':81 'appli':129,131,1016,1072,1093 'applic':282 'approach':948 'arc':216 'architectur':158 'archiv':195,217,485,487,491,756,766,1152,1202 'artifact':561 'ask':62,297 'assign':347 'attempt':938 'author':1266 'auto':1042 'autonom':9,69,853 'autonomi':847 'avail':153 'awar':1056,1125 'babysit':111 'bc':201,204,209,214 'begin':1069 'bilingu':138 'block':200,362,472,553,663,812,880,929,1208 'blocker':42,80,202,265,292,394,457,470,473,475,566,667 'bootstrap':122,1126 'bound':20,326,345,438 'boundari':552 'call':389,865,902,1117,1194 'cancel':758,800,821,840,1192,1224,1239 'capi':287 'captur':168,402,1061,1181 'case':1261 'cd':211,228 'chang':376,668 'check':288,418,563,611,664,1136 'child':1075,1253 'choos':277 'chosen':539 'claim':1211 'classifi':203,458,471,474 'closeout':463,1112,1162,1168 'comment':526,1254 'communic':121,132,144 'complet':66,760,1070 'compos':971 'composit':964 'condit':567,822,866,1118,1240 'conductor':10,88,124,239,412,814,825,850,900,1090,1101,1235,1264 'confirm':870 'conflict':963 'constraint':261,548 'contain':53 'context':119,415,918 'continu':1148 'contract':1163 'control':24,35,143,590,594 'convert':322,337 'cool':1071 'cooldown':212,229,482,576,587,673,680,694,697,708 'cooper':102 'count':1195 'creat':531,608,644,793 'created/assigned':440 'das':601,603,882,884 'decis':14,174,235,246,323,339,420,425,451,514,518,533,677,713,856,960,992,997,1143,1184 'decisions.md':273 'declar':889 'dedup':572,586,606,620,869,950,1232 'dedupe/cooldown/archive':23 'default':696,724 'deleg':242 'design':11,171,234,243,304,409,410,411,483,504,583,907,989,1006,1012,1018,1068,1128,1141,1182 'differ':947,959 'direct':137 'dispatch':16,178,313,318,432,433,434,467,581,672,675,716,744,835,861,1045,1074,1146,1188,1215 'docs/agent-communication-profile.md':127,134,1094 'done':184,223,363,497,507,755,761,838,1158,1183,1212 'done/cancelled':629 'dp':173,176,190 'dsn':170,172,227,231 'dsp':177,179,198 'duplic':646,796,872,1226 'durabl':34,270 'e.g':1179 'elaps':230,695 'encod':990 'endless':597 'enter':1140,1145 'equival':614,632,658 'error':795 'evalu':461 'everi':114,517,1214,1260 'evid':75,189,267,276,295,375,446,460,488,555,764,1189,1243 'exact':560 'execut':250,540 'exist':617,634,653,660,874,954,966 'expect':268,556 'extern':210,392,478 'external-platform':391 'face':1100 'fail':945 'fast':569 'field':995 'final':843 'find':605 'first':1097 'flag':197,468,788,1207 'flowchart':160 'follow':334 'follow-up':333 'format':516 'friction':279,984 'frozen':890 'gate':76,283,464,511,1015,1113 'gc':166,169 'goal':3,44,55,58,82,167,221,264,401,495,535,565,691,734,968,979,1024,1060,1066,1086,1105,1111,1155,1165,1169,1180,1248 'guard':848 'h':162 'handl':1008 'harvest':489,763,1190 'heavi':571,706,1001,1064,1121 'heavy-path':1063 'human':106,136,225,309,502,912 'id':536,1170 'ident':148 'input':503 'instead':655 'intent':255,417,537 'issu':21,52,165,275,328,346,351,367,405,439,529,599,610,615,633,654,659,726,749,762,779,792,826,832,956,1076,1187,1216,1228,1249 'judgment':310,914 'key':573,621,951,1233 'l2':285,1013,1131 'lack':1265 'lane':545,554,881,891,930 'layer':8,103,232,311,967,977,1007,1046 'learn':194,465,486 'leav':809 'limit':740 'line':805 'list':368 'live':156 'load':126 'lock':1106 'loop':12,17,36,1156 'lrn':193,218 'machin':397,1173 'maintain':289 'manual':117 'match':622,952 'max':574,717,722,893 'may':727,851,999 'mcp':152 'mean':399 'memori':271,419,492 'mention':352 'mermaid':159 'minut':577,681,699,704 'mode':4,45,56,59,83,969,980,1025,1087,1166 'model':149 'monitor':113,329,358 'move':750 'multica':327,366 'multipl':71 'must':520,816,901,1091 'mutat':925 'narrow':356 'need':109,208,308,424,481,499,582,585,911,1161,1256 'negoti':551 'never':378 'new':188,374,712,743 'next':220,400,494,1154 'next-goal':1153 'ng':219 'nois':490,589,791,1191 'non':263,550,564,1110 'non-goal':262 'non-negoti':549 'note':650,1102 'nudg':786 'ob':180,182,187 'object':68,259,508,1020,1108,1171 'observ':181,441,442,1051,1150 'ok':141 'one':338,342,348,435,785,804 'one-lin':803 'op':206,224 'oper':163,205,207,254,299,314,388,423,477,480,498,584,864,888,904,1116,1160,1193,1255 'operator-cal':387,863,1115 'operator-declar':887 'own':236,315 'owner':349,546,1109 'packet':15,175,247,324,340,421,426,452,515,519,534,678,714,857,961,993,998,1144,1185 'pass':192,466,512,1206 'pass/flag/block':1220 'path':541,1065,1122,1174 'payment':922 'per':350,600 'period':447 'perman':215,479 'permiss':920 'persist':31,87,94,1030 'platform':393 'post':522,647,784,841,1104,1123,1133,1245 'prefix':625 'pressur':286,1014,1132,1135 'prevent':592 'prior':274,937 'produc':13,245,429,1142,1186 'product':1011 'profil':133 'progress':638,730 'proof':157 'proper':1223 'prove':558 'pull':260,414 'pushback':140 'pushback-ok':139 'rais':1258 'raw':249,1009 're':370,380,454,671,687,770,774 're-dispatch':670 're-read':769 're-rout':369,453,686,773 're-sweep':379 'reach':742 'read':771 'readi':430,579,859 'real':41,79,303 'reason':806,1197 'receiv':406 'refin':253,416 'remain':395,830,1034 'remote/harnessmax/leaderboard':1022 'replac':25,976,1028 'requir':155,927,1010 'reset':709 'residu':1200 'respond':226 'respons':252,336,506 'result':1049 'retri':213 'rev':185,191,196,199 'review':186,357,361,450,459,547,641,753,778,1043,1052,1151 'risk':1201 'rout':18,332,371,428,455,538,688,775,876,898,935,962,1000,1138 'router':986 'run':1057 'runtim':147,924 'rv':1134 'satisfi':1242 'scan':449,684 'scatter':1251 'scope':258,291,427 'sdd':284,996,1004 'secret':921 'select':987 'self':820,1055,1124,1238 'self-awar':1054 'self-cancel':819,1237 'semant':293,759 'session':120 'set':679,798 'share':1229 'shorthand':164,403 'simpl':1036 'singl':29,92,733,1038 'single-ag':28,91,1037 'skill':49,544,1033 'skill-workbench-goal-mode-v2' 'source-fearvox' 'spam':251 'specialist':353 'split':97 'stage':1005 'stale':776 'standard':296,570,701 'startup':125 'state':364,377,396,398,1172,1177 'status':448,627,649 'still':154 'stop':38,384,846 'structur':525 'style':145 'sub':943 'sub-task':942 'success':559 'supersed':797 'supervisor':321,790 'sweep':381 'sweeper':598,1044,1048 'switch':118 'synthesis.md':272 'taken':1204 'task':944,1041 'tast':290,913 'tb':161 'termin':513 'text':130,532,1164 'threshold':1079 'tier':280,568,985,994 'time':737 'titl':624 'tone':135 'tool':150 'topic-agent-skills' 'topic-ai-agents' 'topic-claude-code' 'topic-codex' 'topic-hermes-agent' 'topic-multica' 'topic-sdd' 'topic-skills-sh' 'touch':879,1021 'trace':1175 'trade':306,909 'trade-off':305,908 'transit':1053 'trigger':868 'turn':73 'two':7,101,936 'two-lay':6 'type':476 'unchang':382 'up':335 'updat':493 'use':47,618,982,1047 'user':1099 'user-fac':1098 'v1':26,1026,1032 'v2':5,46,60,84,970,981,1027,1167 'valid':1209 'verdict':578,844,1205,1219 'verifi':1213 'via':365 'viabl':934 'visit':1178 'wait':443,500,745 'window':112 'work':383,702,707,836,873,1002,1023 'workbench':2,43,64,892 'workbench-goal-mode-v2':1 'workflow':99 'wrapper':95,1031 'yes':57,1088 'yet':408","prices":[{"id":"35415226-fb1a-497a-822e-6bc97c8bfb71","listingId":"1f2898da-3fde-419b-8898-4ae9976e1c1b","amountUsd":"0","unit":"free","nativeCurrency":null,"nativeAmount":null,"chain":null,"payTo":null,"paymentMethod":"skill-free","isPrimary":true,"details":{"org":"Fearvox","category":"multica-ultimate-workbench","install_from":"skills.sh"},"createdAt":"2026-05-13T01:05:02.725Z"}],"sources":[{"listingId":"1f2898da-3fde-419b-8898-4ae9976e1c1b","source":"github","sourceId":"Fearvox/multica-ultimate-workbench/workbench-goal-mode-v2","sourceUrl":"https://github.com/Fearvox/multica-ultimate-workbench/tree/main/skills/workbench-goal-mode-v2","isPrimary":false,"firstSeenAt":"2026-05-13T01:05:02.725Z","lastSeenAt":"2026-05-18T19:07:13.504Z"}],"details":{"listingId":"1f2898da-3fde-419b-8898-4ae9976e1c1b","quickStartSnippet":null,"exampleRequest":null,"exampleResponse":null,"schema":null,"openapiUrl":null,"agentsTxtUrl":null,"citations":[],"useCases":[],"bestFor":[],"notFor":[],"kindDetails":{"org":"Fearvox","slug":"workbench-goal-mode-v2","github":{"repo":"Fearvox/multica-ultimate-workbench","stars":12,"topics":["agent-skills","ai-agents","claude-code","codex","hermes-agent","multica","sdd","skills-sh"],"license":null,"html_url":"https://github.com/Fearvox/multica-ultimate-workbench","pushed_at":"2026-05-14T23:12:50Z","description":"Public multi-agent workbench skill pack for SDD, review gates, context discipline, VM lanes, Sanity context, and autonomous agent coordination.","skill_md_sha":"56eff3875f05279114ea122a16d89052acca5354","skill_md_path":"skills/workbench-goal-mode-v2/SKILL.md","default_branch":"main","skill_tree_url":"https://github.com/Fearvox/multica-ultimate-workbench/tree/main/skills/workbench-goal-mode-v2"},"layout":"multi","source":"github","category":"multica-ultimate-workbench","frontmatter":{"name":"workbench-goal-mode-v2","description":"Two-layer autonomous conductor — design loop produces decision packets, dispatch loop routes to bounded issues with dedupe/cooldown/archive controls. Replaces v1's single-agent persistence with a durable control loop that stops only at real blockers."},"skills_sh_url":"https://skills.sh/Fearvox/multica-ultimate-workbench/workbench-goal-mode-v2"},"updatedAt":"2026-05-18T19:07:13.504Z"}}