the-high-council 0.6.0 → 0.7.5

package/README.md

@@ -139,7 +139,31 @@ Fill in `.env` with only the keys your chosen chain needs. Supported providers:
 `OPENROUTER_API_KEY` · `ZAI_API_KEY` · `XAI_API_KEY`
 
 The CLI refuses to start if any seat in the chosen chain is missing its key, rather than failing
-halfway through a paid run.
+halfway through a paid run. `ollama` (below) is the one exception - it needs no key at all.
+
+### Local models (Ollama, LM Studio, ...)
+
+A seat can run on a model you're hosting yourself instead of a paid API. Set its provider to
+`ollama`:
+
+```json
+{ "provider": "ollama", "model": "llama3.1" }
+```
+
+This talks to Ollama's own OpenAI-compatibility endpoint (`http://localhost:11434/v1` by
+default) - no `OLLAMA_API_KEY` needed; Ollama's docs describe a key as "required, but unused."
+If your local runner listens somewhere else (LM Studio, a different port, a remote box), add
+`baseUrl` to that seat:
+
+```json
+{ "provider": "ollama", "model": "mistral", "baseUrl": "http://localhost:1234/v1" }
+```
+
+A local seat prices at $0 in every estimate and run report - `dry_run` shows it as priced, not
+"unpriced," and it can never trip the spend cap. See `chains/local-ollama.json` for a full
+example chain with every seat local. As with every other seat in this project: BYOK is really
+BYO-compute here, and no efficacy claim is made about local models versus hosted ones - nothing
+has been measured.
 
 ## Quick start (CLI)
 
@@ -280,6 +304,7 @@ Then drive it with these tools:
 | `run_status` | Stage reached, panel verdicts, scoreboard, cost, budget remaining |
 | `spend_report` | What every run has cost across a window of days, plus `session_cost_today` (calendar-day, per-model breakdown), derived from disk |
 | `verdict_stats` | How the debate mechanism itself is doing, per chain and per lab: sign-off rate, rounds, objections, dropouts, cost/wall time, largest prompt files - derived from disk |
+| `metrics_report` | Descriptive telemetry only (never an evaluation or baseline): amendment rate, withdrawal rate, objection-follow-through rate, tool-call usage, consensus-induced-regression count - derived from disk. See [`docs/paid-comparison-run.md`](docs/paid-comparison-run.md) for what a *real* comparison against a single model would need - not built, not this metric |
 | `external_prompt` | The prompt a paused external stage is waiting on - **read it with this, not by opening the file**; the prompts are routinely tens of thousands of tokens and most clients silently truncate a file read |
 | `prepare_stage_prompt` | Write a self-contained bundle for a paused stage, so a driving session doing other work at the same time can dispatch it to a fresh subagent instead of authoring it inline - see [`docs/dispatch-pattern.md`](docs/dispatch-pattern.md) |
 | `submit_stage` | Answer a paused stage and resume. Optional `claimed_by` records who's answering; a second submission for an already-answered stage no longer errors - it's kept as `<stage>.late.md` with a warning instead of being rejected |
@@ -417,6 +442,16 @@ claiming 29 chains and 12 tools when there were 30 and 11.
 
 No claim is made here that any of this produces better plans. It has not been measured.
 
+## Skills
+
+`skills/` carries two Claude Code skill files shipped alongside this harness, not part of it:
+`backend-developer` and `frontend-developer` - general backend and frontend engineering rules
+(contract-first design, save/persistence integrity, accessibility, verification discipline) that
+apply to any codebase, not just this one. Same MIT/BYOK ethos as the harness itself. They don't
+claim to make code better; they describe what each rule guards against and why. Point Claude Code
+at this repo's `skills/` directory (or copy the two folders into your own project's
+`.claude/skills/`) to use them.
+
 ## Known limits, stated plainly
 
 - The spend cap is enforced against a **worst case**, not a prediction: the whole prompt billed as
@@ -425,7 +460,9 @@ No claim is made here that any of this produces better plans. It has not been me
   left. That is the intended trade - resume it with a higher ceiling.
 - A seat whose model has no entry in `src/pricing.json` is **unpriced, and therefore uncapped**.
   It contributes $0 to the running total no matter what it really costs. Check `dry_run` output
-  for `unpriced` before trusting a ceiling.
+  for `unpriced` before trusting a ceiling. `ollama` seats are the one deliberate exception: they
+  price at an explicit $0 for any model name (they are genuinely free to run), so they never show
+  up in that `unpriced` list and never need a pricing.json entry.
 - Prices in `src/pricing.json` are hand-maintained list prices, last verified 2026-09-06. They are
   estimates, not invoices. Your provider's bill is the real number.
 - Chains with many labs and high round caps get expensive quickly. `plan-unanimous` at three rounds

package/TROUBLESHOOTING.md

@@ -43,3 +43,16 @@ lab actually returned. The run log also says why it was unreadable: truncated
 (hit its token cap), malformed JSON, or the provider itself returned an error
 mid-generation - each needs a different fix (a higher `maxTokens`, a smaller
 prompt, or just retrying).
+
+## COUNCIL-E005 - Policy refusal (fatal)
+
+Either `policy.json` itself isn't valid JSON, or the chain you're about to run
+violates one of its limits (an unlisted provider or region, over the per-run or
+per-month spend cap, a missing required tag, or an unpriced seat when
+`refuse_unpriced_seats` is set). A policy file is a locked, operator-set set of
+limits checked before any provider is called - a run that would violate it never
+starts, so nothing was spent.
+
+**Fix:** if the file itself won't parse, fix its JSON syntax. Otherwise, either
+change the chain so it no longer violates the listed limit(s), or change
+`policy.json` if the limit itself was wrong. See [docs/policy.md](docs/policy.md).

package/chains/eu-only.json

@@ -0,0 +1,44 @@
+{
+  "name": "eu-only",
+  "description": "Compliance chain (v7.x, procurement readiness): touches only anthropic, openai, mistral - none of the non-EU labs (deepseek, zai) the source procurement report flags. Region: EU. Two labs' worth of critique, same shape as verify.",
+  "maxRounds": 2,
+  "stopOnPass": true,
+  "compliance": {
+    "regions": ["EU"],
+    "providers": ["anthropic", "openai", "mistral"]
+  },
+  "estimate": {
+    "promptTokens": 4000,
+    "draftTokens": 6000,
+    "critiqueTokens": 1200
+  },
+  "seats": {
+    "criteria": {
+      "provider": "anthropic",
+      "model": "claude-opus-5",
+      "maxTokens": 2000
+    },
+    "builder": {
+      "provider": "anthropic",
+      "model": "claude-opus-5",
+      "maxTokens": 16000
+    },
+    "reviser": {
+      "provider": "anthropic",
+      "model": "claude-opus-5",
+      "maxTokens": 16000
+    },
+    "critics": [
+      {
+        "provider": "openai",
+        "model": "gpt-5",
+        "maxTokens": 20000
+      },
+      {
+        "provider": "mistral",
+        "model": "mistral-large-latest",
+        "maxTokens": 20000
+      }
+    ]
+  }
+}

package/chains/local-ollama.json

@@ -0,0 +1,43 @@
+{
+  "name": "local-ollama",
+  "description": "BYO-hardware: every seat runs on a local Ollama instance. No API key, no spend - `dry_run` prices this chain at $0. Swap the model names for whatever you've pulled (`ollama pull <model>`); swap `baseUrl` if Ollama (or an LM Studio / other OpenAI-compatible local runner) listens somewhere other than the default `http://localhost:11434/v1`.",
+  "maxRounds": 2,
+  "stopOnPass": true,
+  "estimate": {
+    "promptTokens": 4000,
+    "draftTokens": 6000,
+    "critiqueTokens": 1200
+  },
+  "seats": {
+    "criteria": {
+      "provider": "ollama",
+      "model": "llama3.1",
+      "maxTokens": 2000
+    },
+    "builder": {
+      "provider": "ollama",
+      "model": "llama3.1",
+      "maxTokens": 16000
+    },
+    "reviser": {
+      "provider": "ollama",
+      "model": "llama3.1",
+      "maxTokens": 16000
+    },
+    "critics": [
+      {
+        "provider": "ollama",
+        "model": "qwen2.5",
+        "maxTokens": 8000,
+        "lab": "ollama-qwen"
+      },
+      {
+        "provider": "ollama",
+        "model": "mistral",
+        "maxTokens": 8000,
+        "baseUrl": "http://localhost:1234/v1",
+        "lab": "lmstudio-mistral"
+      }
+    ]
+  }
+}

package/chains/plan-debate-roles-c1.json

@@ -1,6 +1,6 @@
 {
   "name": "plan-debate-roles-c1",
-  "description": "v6 role-seat plan, phase 5 real-run candidate (docs/v6-decisions.md 'Phase 5 - role debate chain'). Same shape as plan-debate-open-c2, minus Kimi (five seats, not six, to match the plan's five lenses/five personas one-to-one). BUILD-FIX 2026-09-14: originally written against phase 1/2's seats.critics-does-double-duty shape, before the full v6 release integration split seats.proposers (proposes and debates, role-bearing) from seats.critics (panel/critique stage, role-free) as its own array - chain-lint's role-on-non-proposer-seat check (v6 phase 7) is correct that a role on seats.critics alone is a no-op under the current, intentional contract. Fixed by declaring both arrays explicitly: seats.proposers carries the five role-bearing debate seats; seats.critics is the same five labs for the later panel/critique stage, with role stripped (the panel stage must never see it, per v6 phase 2's stage-isolation guarantee - role appearing there at all would itself be a lint/architecture violation, not just redundant). SUBSTITUTION 2026-09-14: the gemini/adversary/parzival seat is now minimax (MiniMax M3, via openrouter) - Google AI Studio's prepayment credits were depleted mid-run (real HTTP 429, not a code bug); Muad chose swapping the lab over topping up credit so the run could happen the same night. Only this one seat changed; the other four labs and their pairings are unaffected. Reasoning for the replacement pick is in docs/v6-decisions.md, not repeated here. Pairings and their reasoning are in docs/v6-decisions.md, not repeated here so there is one place to keep them current.",
+  "description": "v6 role-seat plan, phase 5 real-run candidate (docs/v6-decisions.md 'Phase 5 - role debate chain'). Same shape as plan-debate-open-c2, minus Kimi (five critic seats, not six, to match the plan's five lenses/five personas one-to-one) - each critic seat now carries a role: {lens, persona} pair (v6 §1/§2, applied only to the debate-stage system prompt; the panel stage and every pipeline seat below are unaffected by construction). Pipeline seats (criteria/skeleton/builder/reviser/handoff) are untouched external Claude-Code seats, exactly as in the source chain - roles only ever apply to the critics array's debate-stage prompt. Pairings and their reasoning are in docs/v6-decisions.md, not repeated here so there is one place to keep them current.",
   "maxRounds": 3,
   "signoff": "unanimous",
   "proposals": {
@@ -35,7 +35,7 @@
       "provider": "external",
       "model": "claude-code-session"
     },
-    "proposers": [
+    "critics": [
       {
         "provider": "together",
         "model": "deepseek-ai/DeepSeek-V4-Flash-0731",
@@ -80,69 +80,12 @@
         "role": { "lens": "user-advocate", "persona": "van-gogh" }
       },
       {
-        "provider": "openrouter",
-        "model": "minimax/minimax-m3",
+        "provider": "google",
+        "model": "gemini-3.6-flash",
         "maxTokens": 20000,
-        "extra": {
-          "reasoning": {
-            "enabled": false
-          }
-        },
-        "lab": "minimax",
+        "lab": "gemini",
         "role": { "lens": "adversary", "persona": "parzival" }
       }
-    ],
-    "critics": [
-      {
-        "provider": "together",
-        "model": "deepseek-ai/DeepSeek-V4-Flash-0731",
-        "maxTokens": 20000,
-        "extra": {
-          "chat_template_kwargs": {
-            "thinking": false
-          }
-        },
-        "lab": "deepseek"
-      },
-      {
-        "provider": "openrouter",
-        "model": "qwen/qwen3.5-plus-20260420",
-        "maxTokens": 20000,
-        "extra": {
-          "reasoning": {
-            "enabled": false
-          }
-        },
-        "lab": "qwen"
-      },
-      {
-        "provider": "openrouter",
-        "model": "z-ai/glm-5.3-flash",
-        "maxTokens": 20000,
-        "lab": "glm",
-        "extra": {
-          "reasoning": {
-            "effort": "low"
-          }
-        }
-      },
-      {
-        "provider": "openrouter",
-        "model": "mistralai/mistral-large-2512",
-        "maxTokens": 20000,
-        "lab": "mistral"
-      },
-      {
-        "provider": "openrouter",
-        "model": "minimax/minimax-m3",
-        "maxTokens": 20000,
-        "extra": {
-          "reasoning": {
-            "enabled": false
-          }
-        },
-        "lab": "minimax"
-      }
     ]
   },
   "scope": "open"

package/chains/plan-two-strong.json

@@ -0,0 +1,74 @@
+{
+  "name": "plan-two-strong",
+  "description": "Roster inversion (relay/runs/2026-09-14T14-56-18-834Z/deliverable.md, item 6). Two strong seats from different labs propose and debate one round; one strong adjudicator seat signs off. No cheap-tier critic roster at all. Additive-only - no existing chain references this file.",
+  "maxRounds": 1,
+  "signoff": "unanimous",
+  "roster": {
+    "minimal_two_strong": true
+  },
+  "proposals": {
+    "parts": 3,
+    "maxTokens": 1500
+  },
+  "debate": true,
+  "handoff": true,
+  "estimate": {
+    "promptTokens": 12000,
+    "draftTokens": 16000,
+    "critiqueTokens": 2000
+  },
+  "seats": {
+    "criteria": {
+      "provider": "anthropic",
+      "model": "claude-opus-5",
+      "maxTokens": 2000,
+      "lab": "anthropic"
+    },
+    "skeleton": {
+      "provider": "anthropic",
+      "model": "claude-opus-5",
+      "maxTokens": 6000,
+      "lab": "anthropic"
+    },
+    "builder": {
+      "provider": "anthropic",
+      "model": "claude-opus-5",
+      "maxTokens": 40000,
+      "lab": "anthropic"
+    },
+    "reviser": {
+      "provider": "anthropic",
+      "model": "claude-opus-5",
+      "maxTokens": 40000,
+      "lab": "anthropic"
+    },
+    "handoff": {
+      "provider": "anthropic",
+      "model": "claude-opus-5",
+      "maxTokens": 4000,
+      "lab": "anthropic"
+    },
+    "proposers": [
+      {
+        "provider": "anthropic",
+        "model": "claude-opus-5",
+        "maxTokens": 20000,
+        "lab": "anthropic"
+      },
+      {
+        "provider": "openai",
+        "model": "gpt-5",
+        "maxTokens": 20000,
+        "lab": "openai"
+      }
+    ],
+    "critics": [
+      {
+        "provider": "google",
+        "model": "gemini-2.5-pro",
+        "maxTokens": 20000,
+        "lab": "google"
+      }
+    ]
+  }
+}

package/chains/single-vendor-anthropic.json

@@ -0,0 +1,43 @@
+{
+  "name": "single-vendor-anthropic",
+  "description": "Compliance chain (v7.x, procurement readiness): every seat routes through anthropic (models claude-opus-5, claude-sonnet-5, claude-haiku-4-5-20251001), pinned to exact model versions in src/pricing.json - a single vendor account and a single provider key for the whole run.",
+  "maxRounds": 2,
+  "stopOnPass": true,
+  "compliance": {
+    "providers": ["anthropic"]
+  },
+  "estimate": {
+    "promptTokens": 4000,
+    "draftTokens": 6000,
+    "critiqueTokens": 1200
+  },
+  "seats": {
+    "criteria": {
+      "provider": "anthropic",
+      "model": "claude-opus-5",
+      "maxTokens": 2000
+    },
+    "builder": {
+      "provider": "anthropic",
+      "model": "claude-opus-5",
+      "maxTokens": 16000
+    },
+    "reviser": {
+      "provider": "anthropic",
+      "model": "claude-opus-5",
+      "maxTokens": 16000
+    },
+    "critics": [
+      {
+        "provider": "anthropic",
+        "model": "claude-sonnet-5",
+        "maxTokens": 20000
+      },
+      {
+        "provider": "anthropic",
+        "model": "claude-haiku-4-5-20251001",
+        "maxTokens": 20000
+      }
+    ]
+  }
+}

package/chains/single-vendor-openai.json

@@ -0,0 +1,43 @@
+{
+  "name": "single-vendor-openai",
+  "description": "Compliance chain (v7.x, procurement readiness): every seat routes through openai (models gpt-5, gpt-5-mini, gpt-5-nano), pinned to exact model versions in src/pricing.json - a single vendor account and a single provider key for the whole run.",
+  "maxRounds": 2,
+  "stopOnPass": true,
+  "compliance": {
+    "providers": ["openai"]
+  },
+  "estimate": {
+    "promptTokens": 4000,
+    "draftTokens": 6000,
+    "critiqueTokens": 1200
+  },
+  "seats": {
+    "criteria": {
+      "provider": "openai",
+      "model": "gpt-5",
+      "maxTokens": 2000
+    },
+    "builder": {
+      "provider": "openai",
+      "model": "gpt-5",
+      "maxTokens": 16000
+    },
+    "reviser": {
+      "provider": "openai",
+      "model": "gpt-5",
+      "maxTokens": 16000
+    },
+    "critics": [
+      {
+        "provider": "openai",
+        "model": "gpt-5-mini",
+        "maxTokens": 20000
+      },
+      {
+        "provider": "openai",
+        "model": "gpt-5-nano",
+        "maxTokens": 20000
+      }
+    ]
+  }
+}

package/chains/us-only.json

@@ -0,0 +1,49 @@
+{
+  "name": "us-only",
+  "description": "Compliance chain (v7.x, procurement readiness): touches only anthropic, openai, xai, google - all US-based labs. Region: US. Two labs' worth of critique, same shape as verify.",
+  "maxRounds": 2,
+  "stopOnPass": true,
+  "compliance": {
+    "regions": ["US"],
+    "providers": ["anthropic", "openai", "xai", "google"]
+  },
+  "estimate": {
+    "promptTokens": 4000,
+    "draftTokens": 6000,
+    "critiqueTokens": 1200
+  },
+  "seats": {
+    "criteria": {
+      "provider": "anthropic",
+      "model": "claude-opus-5",
+      "maxTokens": 2000
+    },
+    "builder": {
+      "provider": "anthropic",
+      "model": "claude-opus-5",
+      "maxTokens": 16000
+    },
+    "reviser": {
+      "provider": "anthropic",
+      "model": "claude-opus-5",
+      "maxTokens": 16000
+    },
+    "critics": [
+      {
+        "provider": "openai",
+        "model": "gpt-5",
+        "maxTokens": 20000
+      },
+      {
+        "provider": "xai",
+        "model": "grok-4",
+        "maxTokens": 20000
+      },
+      {
+        "provider": "google",
+        "model": "gemini-2.5-pro",
+        "maxTokens": 20000
+      }
+    ]
+  }
+}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "the-high-council",
-  "version": "0.6.0",
+  "version": "0.7.5",
   "description": "The High Council: a multi-model planning harness. One builder, independent critics from other labs, checkable acceptance criteria, a debate on the record, and a hard stop.",
   "type": "module",
   "license": "MIT",