feat(goal-conductor): Diamondnode control-plane bootstrap (evt-diamondnode-qubo-first-prompt)#3
feat(goal-conductor): Diamondnode control-plane bootstrap (evt-diamondnode-qubo-first-prompt)#3Igor Holt (igor-holt) wants to merge 9 commits into
Conversation
…evt-diamondnode-qubo-first-prompt) Scaffolds goal_conductor_live_artifact, budget_policy, openclaw_dispatch_policy, haiku_worker_prompt, hermes_openclaw_bridge (design-only), utilization_ledger, session_registry, decision_log. QUBO portfolio NOT computed: strategist skill absent — VPD fields null rather than fabricated. Ref evt-diamondnode-qubo-first-prompt. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
|
You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard. |
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request scaffolds the Diamondnode Goal Conductor control plane, introducing budget policies, decision logs, live artifacts, worker prompt templates, and dispatch policies. The review feedback identifies several configuration inconsistencies and potential bugs: a mismatch between gate references and definitions in the dispatch policy, inconsistent routing keys, unquoted colons in YAML model tags that could cause parsing errors, an inconsistent ledger file path in the bridge configuration, and a schema discrepancy between the worker's output and the utilization ledger.
Important
The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.
| - { task_class: cheap_classification,route: local, gate: arbiter_vram_thermal } | ||
|
|
||
| gates: | ||
| arbiter: |
There was a problem hiding this comment.
The routing table on lines 18-19 references gate: arbiter_vram_thermal, but the gate configuration block here is named arbiter. Rename this block to arbiter_vram_thermal to match the references and prevent dispatch resolution failures.
arbiter_vram_thermal:| - { task_class: architecture, route: opus, policy_gate: true } | ||
| - { task_class: security, route: opus, policy_gate: true } | ||
| - { task_class: long_horizon_plan, route: opus, policy_gate: true } |
There was a problem hiding this comment.
The routing table uses policy_gate: true for opus routed tasks, but uses gate: arbiter_vram_thermal for local routed tasks. To maintain consistency and allow the dispatcher to dynamically resolve gate configurations, use the gate key pointing to the defined opus_policy_gate.
- { task_class: architecture, route: opus, gate: opus_policy_gate }
- { task_class: security, route: opus, gate: opus_policy_gate }
- { task_class: long_horizon_plan, route: opus, gate: opus_policy_gate }| strategist: { model: claude-opus-4-8, use: architecture/security/long-horizon } | ||
| default_worker: { model: claude-haiku-4-5, caps: { soft_usd: 0.25, hard_usd: 1.00, max_turns: 6, wall_min: 12, in_tok: 60000, out_tok: 8000 } } | ||
| implementation: { model: sonnet } | ||
| local: { models: [llama3.2:3b, qwen2:0.5b, hermes-g5-real-test], gate: arbiter_vram_thermal } |
There was a problem hiding this comment.
Unquoted colons in flow sequences (e.g., llama3.2:3b and qwen2:0.5b) can cause parsing issues with some YAML parsers. It is safer to wrap these model identifiers in double quotes.
local: { models: ["llama3.2:3b", "qwen2:0.5b", "hermes-g5-real-test"], gate: arbiter_vram_thermal }| plane: worker | ||
| template: haiku_worker_prompt.md | ||
| on_complete: | ||
| - append: utilization_ledger.jsonl |
There was a problem hiding this comment.
The path to the utilization ledger is inconsistent with other files (e.g., budget_policy.yaml and haiku_worker_prompt.md), which use goal-conductor/utilization_ledger.jsonl. Update this path to include the goal-conductor/ prefix to ensure consistency and avoid writing to the root directory.
- append: goal-conductor/utilization_ledger.jsonl| 1. Stay strictly inside the task contract. No scope expansion. | ||
| 2. Never print secrets. Reference env vars by name only. | ||
| 3. If confidence < 0.72 OR you fail twice → STOP and emit an escalation handoff (do not retry endlessly). | ||
| 4. On completion, emit a utilization record (append to goal-conductor/utilization_ledger.jsonl) and a one-line artifact patch summary. |
There was a problem hiding this comment.
The instruction tells the worker to append a utilization record directly to goal-conductor/utilization_ledger.jsonl. However, the specified output JSON format on lines 26-37 does not match the schema of the ledger (which contains fields like record_type, agent, surface, goal_vectors, model, etc.). Please clarify whether the worker should format the record to match the ledger schema before appending, or if openclaw handles the ledger formatting using the worker's output.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…rketplace entry Skill moved from ~/.claude/skills (symlinked back for CLI loading). QUBO/knapsack VPD portfolio selector, exact or annealing solver, heuristic value manifest, honest evt- emission. Registered in .claude-plugin/marketplace.json. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…kills, and README/index.ts route drift note Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…ctive port energy/elapsed_ms were fabricated (hardcoded base + Math.random) yet Ed25519-signed into the audit stream; callers could also override decision fields via body spread. Now: real vram-efficiency sigmoid + constraint checks ported from unified_inference/optimizer.py, measured elapsed_ms, caller telemetry quarantined under inputs/telemetry_used, mode=deterministic-objective-v1 declared in the signed payload. 10 new vitest cases. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…uth gate; honest README POST /notion/* now 503-until-provisioned / 401-on-bad-bearer (no silent open endpoints). Env extended with optional notion proxy fields (fixes pre-existing notion.ts typecheck errors). README endpoint table reflects reality: BotID disclosed as disabled no-op, placeholder handlers labeled. 7 new route tests. deploy --dry-run verified; live deploy left to operator. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
… report Measures real signals (vitest counts, typecheck, deploy:dry, gpu benchmark suite), weights them on the first-prompt value axis, costs from ledger or operator input. First run: value 10.0/10 measured, $4.50 session estimate -> VPD 2.22 (acceptable). value_basis=measured-internal-v0.1; external baselines roadmap included. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…assessor + first run records) assess_activity.sh gathers cross-repo git movement, ledger/decision deltas, health sweep, GPU state since last marker. daily_ops_workflow.md is the canonical 6-step procedure (assess -> qubo -> act/delegate -> newsletter -> publish -> mark) with public-disclosure safety gates. Inaugural post live on news.genesisconductor.io. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
|
1 participant
Summary
Bootstraps the Diamondnode Goal Conductor control plane under
goal-conductor/, per the canonical first-prompt (evt-diamondnode-qubo-first-prompt).8 artifacts, conservative real values, no fabricated QUBO/VPD numbers — the
diamondnode-qubo-economics-strategistskill is absent so VPD fields arenullrather than invented (respects the no-unbenchmarked-claims safety gate).Files
goal_conductor_live_artifact.yaml— G1–G5 goals, topology, routing matrix, hardware probebudget_policy.yaml— daily $25/$50, monthly $500/$1000, Haiku per-task capsopenclaw_dispatch_policy.yaml— routing table + arbiter/opus gateshaiku_worker_prompt.md— bounded worker templatehermes_openclaw_bridge.yaml— ingress→dispatch design (~5%, not deployed)utilization_ledger.jsonl/session_registry.jsonl/decision_log.jsonlSync status
Not included
QUBO portfolio computation — blocked on the missing strategist skill (authored separately).