ultimate-pi 0.10.0 → 0.11.0

package/.pi/lib/harness-run-context.ts

@@ -114,6 +114,15 @@ export function canonicalPlanPath(runId: string, projectRoot: string): string {
 	return join(harnessRunsRoot(projectRoot), runId, "plan-packet.json");
 }
 
+export const PLAN_REVIEW_BASENAME = "plan-review.md";
+
+export function canonicalPlanReviewPath(
+	runId: string,
+	projectRoot: string,
+): string {
+	return join(harnessRunsRoot(projectRoot), runId, PLAN_REVIEW_BASENAME);
+}
+
 const PLAN_PACKET_BASENAME = "plan-packet.json";
 
 const MUTATING_FILE_TOOLS = new Set(["write", "edit"]);
@@ -910,6 +919,9 @@ export function formatPlanContextBlock(
 	];
 	if (ctx.plan_packet_path) {
 		lines.push(`plan_packet_path=${ctx.plan_packet_path}`);
+		lines.push(
+			`plan_review_path=${canonicalPlanReviewPath(ctx.run_id, ctx.project_root)}`,
+		);
 	}
 	if (ctx.task_summary) {
 		lines.push(`task_summary=${ctx.task_summary}`);

package/.pi/prompts/harness-auto.md

@@ -18,7 +18,7 @@ If task missing:
 
 ## Orchestration (required) — same session
 
-1. **Plan** — spawn `harness/planner` → parse JSON → present full plan → `ask_user` Approve/Changes/Cancel → write `plan-packet.json` only on Approve (advances phase via policy-gate).
+1. **Plan** — follow `/harness-plan` parent orchestration (parallel `harness/planning/scout-*`, `decompose`, `hypothesis`, draft PlanPacket, `ask_user` on fork, parallel `plan-adversary` + `hypothesis-eval`, parent `approve_plan` + `create_plan`). Do not spawn `harness/planner`. No second approval pass.
 2. **Execute** — spawn `harness/executor` with `HarnessSpawnContext` (`mode: execute`). Summarize handoff bullets for next spawn (do not paste full subagent log).
 3. **Eval** — spawn `harness/evaluator` (`mode: benchmark`) after parent scripts if needed.
 4. **Review** — spawn `harness/evaluator` (`mode: verdict`) OR rely on eval verdict if policy allows — prefer both when strict gates require.

package/.pi/prompts/harness-plan.md

@@ -1,58 +1,154 @@
 ---
-description: Build a strict read-only PlanPacket before any mutating work.
+description: Transform a vague task into a rigorous hypothesis via decomposition + DARWIN synthesis, then a strict PlanPacket.
 argument-hint: "\"<task>\" [--risk low|med|high] [--budget <amount>] [--quick]"
 ---
 
 # harness-plan
 
-Orchestrator only — spawn `harness/planner` once. The planner runs clarification (`ask_user`), approval (`approve_plan`), and persists the plan (`create_plan`). Do **not** write `plan-packet.json` in this parent session.
+Parent orchestrator — run planning in **this session**. Subagents explore, decompose, hypothesize, and review; you own `ask_user`, `approve_plan`, and `create_plan`. Never `write` or `edit` `plan-packet.json` — use **`create_plan`** only.
+
+Allowed `subagent_type` values (copy exactly):
+
+- `harness/planning/scout-graphify`
+- `harness/planning/scout-structure`
+- `harness/planning/scout-semantic`
+- `harness/planning/decompose`
+- `harness/planning/hypothesis`
+- `harness/planning/plan-adversary`
+- `harness/planning/hypothesis-eval`
+
+Do **not** spawn `harness/planner` or `harness/planning/planner`.
 
 ## Step 0 — Parse arguments
 
 Read `$ARGUMENTS`:
 
-- task statement (required)
+- task statement (required) — **THE QUESTION**
 - optional: `--risk low|med|high`, `--budget <amount>`, `--quick`
 
 If task is missing:
 
 `Usage: /harness-plan "<task>" [--risk low|med|high] [--budget <amount>] [--quick]`
 
-`--quick` narrows planning breadth only — it does **not** skip user approval.
+`--quick` skips `harness/planning/scout-semantic` only — never skip graphify, structure, decompose, hypothesis, or approval.
 
 ## Active plan context
 
-Use injected context only — **do not** read `.pi/harness/specs/*.schema.json` or explore specs with bash.
+Use injected context only — **do not** read `.pi/harness/specs/*.schema.json` from disk.
 
 If `[HarnessActivePlan]` is present:
 
-- Treat task as **revise/amend** unless `/harness-new-run` was used.
-- Pass `mode: revise` using the `HarnessSpawnContext` JSON in `[HarnessRunContext]`.
+- Treat as **revise/amend** unless `/harness-new-run` was used.
+- Set `mode: revise` in `HarnessSpawnContext` from `[HarnessRunContext]`.
+- **Preserve `plan_id` and `task_id`** from the existing packet when amending.
+- Scouts focus on delta vs existing `plan_packet_path`; full re-scout only if scope changed materially.
+
+Otherwise use `HarnessSpawnContext` from `[HarnessRunContext]` with `mode: create`.
+
+## Phase 1 — Parallel scouts (required)
+
+1. Copy `HarnessSpawnContext` from `[HarnessRunContext]` (adjust `risk_level`, `quick`, `mode` from `$ARGUMENTS`).
+2. Spawn scouts with **`inherit_context: false`**. Prefer parallel: `run_in_background: true` on each `Agent` call, then `get_subagent_result` for all.
+
+```
+Agent({ subagent_type: "harness/planning/scout-graphify", prompt: "<task + HarnessSpawnContext + scout JSON schema>", run_in_background: true })
+Agent({ subagent_type: "harness/planning/scout-structure", prompt: "…", run_in_background: true })
+```
+
+Skip `harness/planning/scout-semantic` when `--quick` or `quick: true`.
 
-Otherwise use `HarnessSpawnContext` from `[HarnessRunContext]` for greenfield `mode: create`.
+3. Parse each scout’s fenced `json` (`lane`, `status`, `findings`, `key_paths`, `open_questions`).
+4. **Partial failure:** require successful **graphify + structure** lanes. Semantic is optional. If a required lane fails, continue with `plan_status: partial` and document gaps in `assumptions`.
+5. If JSON parse fails for a lane, summarize free-text output and add an assumption that the lane was unstructured.
 
-## Orchestration (required)
+## Phase 2 — Decompose (DeepMind-style)
 
-1. Copy the `HarnessSpawnContext=…` JSON from `[HarnessRunContext]` into the spawn prompt (adjust `risk_level`, `quick`, `mode` from `$ARGUMENTS` if needed). Do **not** add “call ask_user for approval” in the `Agent` prompt — the planner agent instructions already define `approve_plan` / `create_plan`.
-2. Spawn **once** with **`inherit_context: false`**:
+1. Spawn once with merged scout JSON:
 
 ```
-Agent({ subagent_type: "harness/planner", prompt: "<task + HarnessSpawnContext JSON + output schema>" })
+Agent({ subagent_type: "harness/planning/decompose", prompt: "<HarnessSpawnContext + task + all scout lane JSON>", inherit_context: false })
 ```
 
-3. `get_subagent_result` — parse final JSON (`status`, `plan_packet`, `human_summary`, `clarification`) via fenced `json` block.
-4. If `status === "ready"` and parent `harness-run-context` shows `plan_ready: true` (planner called `create_plan`), confirm `plan_packet_path` exists — do **not** write the file yourself.
-5. If `needs_clarification`, tell the user the planner is waiting — do **not** re-spawn; user should answer in the subagent or re-run `/harness-plan`.
-6. Do **not** call `ask_user`, `approve_plan`, or `create_plan` in this parent session.
+2. Parse `PlanDecompositionBrief` JSON (`problem_restatement`, `core_tension`, `tensions`, `prior_art`, etc.).
+3. On parse failure: one retry with “output valid JSON only”; if still failing, abort with `plan_status: needs_clarification`.
+
+## Phase 3 — Hypothesis (DARWIN)
+
+1. Spawn once:
+
+```
+Agent({ subagent_type: "harness/planning/hypothesis", prompt: "<HarnessSpawnContext + task + PlanDecompositionBrief + scout summaries>", inherit_context: false })
+```
+
+2. Parse `PlanHypothesisBrief` JSON (`primary`, `dialectical_fork`, `alternatives`, `recommended_next_steps`).
+3. **Revision cap:** at most **one** re-spawn of `hypothesis` if Phase 6 eval requests revision (see below).
+
+## Phase 4 — Draft PlanPacket + fork clarification (parent)
+
+Map hypothesis → [`PlanPacket`](.pi/harness/specs/plan-packet.schema.json):
+
+| Field | Source |
+|-------|--------|
+| `scope` | `problem_restatement` (narrowed) + `primary.claim` + `primary.mechanism` (implementation-ready) |
+| `assumptions` | `core_tension`, `prior_art.dead_ends`, scout `open_questions`, chosen fork path (if any) |
+| `acceptance_checks` | Each `primary.prediction` and `primary.experiment` as verifiable checklist items (min 1) |
+| `risk_level` | From `$ARGUMENTS` or infer from fork uncertainty / blast radius |
+
+Build complete draft: `plan_id`, `task_id`, `scope`, `assumptions`, `risk_level`, `acceptance_checks`, `rollback_plan` (`revert_commit_ready: true`, artifacts filled).
+
+Call **`ask_user`** when `dialectical_fork` is material (Path A vs B materially different) **before** Phase 5 reviews.
+
+Assemble `research_brief` for approval:
+
+```json
+{
+  "decomposition": { /* PlanDecompositionBrief */ },
+  "hypothesis": { /* PlanHypothesisBrief */ },
+  "eval": null
+}
+```
+
+## Phase 5 — Parallel reviews
+
+Spawn in parallel (`run_in_background: true`):
+
+```
+Agent({ subagent_type: "harness/planning/plan-adversary", prompt: "<HarnessSpawnContext + draft PlanPacket + scout summaries + decomposition human_summary>", inherit_context: false })
+Agent({ subagent_type: "harness/planning/hypothesis-eval", prompt: "<original task ONLY + PlanHypothesisBrief JSON — no decomposition, no PlanPacket>", inherit_context: false })
+```
+
+1. Parse `PlanAdversaryBrief` — merge `mitigations` into scope, assumptions, or `acceptance_checks`.
+2. Parse `PlanHypothesisEval` — set `research_brief.eval`.
+3. If `revision_recommended` or testability &lt; 70 or `relevance.passes` is false: re-spawn `hypothesis` once with eval rationale, update PlanPacket + `research_brief.hypothesis`, then re-run **hypothesis-eval** only (not adversary unless PlanPacket changed materially).
+
+Cap: **at most 2** plan-adversary spawns and **at most 2** `approve_plan` rounds per invocation.
+
+## Phase 6 — Approval + persistence (parent)
+
+1. Call **`approve_plan`** with `plan_packet`, `human_summary` (primary claim + fork if any), and `research_brief`.
+2. On **Approve** only, call **`create_plan`** with the **same** `plan_packet`.
+3. If `create_plan` fails, tell the user to fix validation errors or run `/harness-plan-commit` after approval is recorded.
+4. Confirm `[HarnessRunContext]` `plan_ready: true` before handoff.
+
+On **Cancel** or Esc: `plan_status: needs_clarification`; do **not** call `create_plan`.
+
+On **Request changes**: revise draft and re-run phases 4–6 only (re-scout/decompose/hypothesis only if scope changed).
+
+## Recovery and ownership
+
+- Plan only in the **owner** session (`owner_pi_session_id` on run context); otherwise `/harness-use-run`.
+- `/harness-plan-commit` only after parent `approve_plan` (Approve) is in the transcript.
+- If `plan_ready: true` already, stop — summarize and set `next_command: /harness-run`.
 
 ## Parent rules
 
-- Do not mutate project source files in the plan phase.
-- Do not embed `plan_id=` in prompts for policy sync.
-- Optional recovery: `/harness-plan-commit` only if the planner approved but `create_plan` failed.
+- Do not mutate project source in plan phase.
+- Subagents never call `ask_user`, `approve_plan`, or `create_plan`.
+- Do not embed `plan_id=` in spawn prompts for policy sync.
 
 ## Completion
 
-- `plan_status`: `ready` or `needs_clarification`
+- `plan_status`: `ready`, `partial`, or `needs_clarification`
 - `risk_level` used
+- `plan_review_path` shown for editor review
 - `next_command`: `/harness-run` when `ready` (never `/harness-run --plan …`)

package/.pi/prompts/harness-setup.md

@@ -387,7 +387,7 @@ Manual override: **`/router profile auto`** anytime after reload if they changed
 
 **Slash commands are orchestrators:** `/harness-plan`, `/harness-run`, etc. spawn `harness/*` agents via the `Agent` tool — bootstrap stays **script-first**; only optionally spawn `harness/sentrux-bootstrap` for Sentrux (see Step 4.2).
 
-Optional per-repo overrides: place `.md` files at the **same relative path** (e.g. `.pi/agents/harness/planner.md` overrides the package planner).
+Optional per-repo overrides: place `.md` files at the **same relative path** (e.g. `.pi/agents/harness/planning/scout-graphify.md` overrides the package scout). Deprecated: `harness/planner.md` — use `harness/planning/` agents instead.
 
 Verify manifest drift after `pi update ultimate-pi`:
 

package/.pi/scripts/harness-resolve-up-pkg.mjs

@@ -30,7 +30,20 @@ function hasHarnessScripts(root) {
 	return existsSync(join(root, ".pi", "scripts", "harness-cli-verify.sh"));
 }
 
+function isSourceCheckout(root) {
+	try {
+		const pkg = requireFromCwd.resolve("./package.json");
+		return dirname(pkg) === root;
+	} catch {
+		return false;
+	}
+}
+
 function tryResolveUltimatePi() {
+	if (hasHarnessScripts(process.cwd()) && isSourceCheckout(process.cwd())) {
+		return process.cwd();
+	}
+
 	if (process.env.ULTIMATE_PI_PKG) {
 		const envRoot = process.env.ULTIMATE_PI_PKG;
 		if (hasHarnessScripts(envRoot)) return envRoot;

package/CHANGELOG.md

@@ -4,6 +4,24 @@ All notable changes to this project are documented in this file.
 
 ## [Unreleased]
 
+## [v0.11.0] — 2026-05-17
+
+### ✨ Features
+
+- **Harness Darwin plan pipeline:** decomposition and hypothesis agents with plan-adversary, scouts, and structured plan brief schemas (ADR 0034).
+- **Harness plan review:** `plan-review.md` for editor review; extension load guard.
+- **Cursor Pi experts:** cursor-pi domain expert agents.
+
+### 🔄 CI/CD
+
+- **Biome:** ignore harness runtime JSON; format committed harness plan pipeline sources.
+
+## [v0.10.1] — 2026-05-17
+
+### 🐛 Fixes
+
+- **Harness plan TUI:** agents widget stacks above harness-live; full-height bottom `approve_plan` overlay above the harness band; block duplicate parent `approve_plan` / plan `ask_user` after planner subagent approval.
+
 ## [v0.10.0] — 2026-05-17
 
 ### ✨ Features

package/biome.json

@@ -12,7 +12,10 @@
 			"**/*.{ts,tsx,js,jsx,json,jsonc,css}",
 			"!graphify-out/**/*",
 			"!graphify-books-out/**/*",
-			"!vendor/**/*"
+			"!vendor/**/*",
+			"!.pi/harness/active-run.json",
+			"!.pi/harness/runs/**/run-context.json",
+			"!.pi/harness/runs/**/plan-packet.json"
 		]
 	},
 	"formatter": {

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "ultimate-pi",
-	"version": "0.10.0",
+	"version": "0.11.0",
 	"description": "Ultimate AI coding harness for pi.dev — extensible skills, Obsidian wiki knowledge layer, compressed context, deterministic output",
 	"keywords": [
 		"pi-package",
@@ -82,7 +82,7 @@
 		"format": "biome format --write",
 		"format:check": "biome format",
 		"prepare": "lefthook install",
-		"test": "node --test test/harness-verify.test.mjs test/harness-ask-user.test.mjs test/harness-subagents-loader.test.mjs test/harness-subagents-import-path.test.mjs test/sentrux-rules-sync.test.mjs test/harness-budget-guard.test.mjs && npx -y tsx --test test/harness-vcc-settings.test.ts test/harness-plan-phase-policy.test.mjs test/harness-subagent-policy.test.mjs test/harness-turn-routing.test.mjs test/plan-approval-format.test.mjs test/plan-approval-sync.test.mjs test/plan-create-plan.test.mjs",
+		"test": "node --test test/harness-verify.test.mjs test/harness-ask-user.test.mjs test/harness-subagents-loader.test.mjs test/harness-subagents-import-path.test.mjs test/sentrux-rules-sync.test.mjs test/harness-budget-guard.test.mjs && npx -y tsx --test test/harness-vcc-settings.test.ts test/harness-plan-phase-policy.test.mjs test/harness-subagent-policy.test.mjs test/harness-turn-routing.test.mjs test/plan-approval-format.test.mjs test/plan-approval-dialog.test.mjs test/plan-approval-sync.test.mjs test/plan-create-plan.test.mjs test/plan-review-format.test.mjs",
 		"test:vcc": "npx -y tsx --test vendor/pi-vcc/tests/*.test.ts",
 		"harness:sentrux-bootstrap": "node .pi/scripts/harness-sentrux-bootstrap.mjs",
 		"harness:sentrux-sync": "node .pi/scripts/sentrux-rules-sync.mjs --force",