pi-goal-x 0.10.1 → 0.11.0

package/README.md

@@ -10,10 +10,10 @@ The extension is designed around one rule: **the user owns intent; the agent exe
 
 All core features of [@capyup/pi-goal](https://github.com/capyup/pi-goal) are preserved. The following changes are specific to pi-goal-x:
 
-### Mid-flight objective updates
+### Goal objective is immutable
 
-- **`update_goal({updatedObjective})`** — the agent can now sync the goal objective mid-flight when user requirements change, *without* completing the goal. This ensures the completion auditor evaluates against the latest requirements. The combined path (`updatedObjective` + `status: "complete"`) applies the update first, then runs the normal completion+audit flow.
-- **`apply_goal_tweak`** remains available for `/goal-tweak` drafting revisions; the new parameter is the lightest possible touch on the existing `update_goal` tool.
+- The goal objective is immutable — the agent **must not** modify it autonomously. Objective changes are only possible through `propose_goal_tweak`, which presents the user with a Confirm / Continue Chatting dialog matching the `propose_goal_draft` confirmation pattern. This prevents the agent from silently changing the goal contract.
+- **`propose_goal_tweak`** is the sole mechanism for updating the objective, available exclusively during a `/goal-tweak` drafting flow. If the user's requirements change, they must run `/goal-tweak` to initiate the revision flow.
 
 ### Deferred archival
 
@@ -23,7 +23,7 @@ All core features of [@capyup/pi-goal](https://github.com/capyup/pi-goal) are pr
 ### E2e test infrastructure
 
 - **Deterministic fork tests using `--mode json`**: the e2e suite spawns a real `pi --fork --mode json` session, parses structured `tool_execution_start`/`tool_execution_end` JSON events for field-level assertions — no free-text AI output parsing. Uses `--append-system-prompt` + `--tools` to force deterministic tool calls.
-- **Full coverage**: 131 tests total — function-level integration tests (12), mock-pi handler tests (4), file-validity checks (6), and real `pi --fork --mode json` tests (3 scenarios: quick-sync, combined sync+complete, deferred archival).
+- **Full coverage**: 143 tests total — function-level integration tests (12), mock-pi handler tests (4), file-validity checks (6), real `pi --fork --mode json` tests (3 scenarios: quick-sync, combined sync+complete, deferred archival), and propose_goal_tweak unit/integration/e2e tests (15).
 
 ### Completion auditor
 
@@ -156,11 +156,12 @@ The extension exposes tools only when they make sense for the current lifecycle
 | `goal_question` | drafting / tweak drafting | Ask one focused user question |
 | `goal_questionnaire` | drafting / tweak drafting | Ask multiple structured questions |
 | `get_goal` | always | Read the focused goal state; mentions other open goals when present |
-| `propose_goal_draft` | registered; accepted only during goal drafting | Submit a concrete draft for user confirmation |
-| `apply_goal_tweak` | tweak drafting only | Submit a revision to an existing goal |
+| `propose_goal_draft` | drafting only (goal creation) | Submit a concrete draft for user confirmation |
+| `propose_goal_tweak` | tweak drafting only | Submit a revision to an existing goal (shows Confirm / Continue Chatting dialog) |
 | `update_goal` | focused active or paused goal | Mark the focused goal complete when all requirements are satisfied. When the auditor is disabled, supply `confirmBypassAuditor: true` after user confirmation to bypass the audit |
 | `pause_goal` | focused active goal | Pause the focused goal because of a real blocker |
 | `abort_goal` | focused active or paused goal | Abort/archive an obsolete, impossible, unsafe, or user-cancelled focused goal |
+| `propose_goal_tweak` | tweak drafting only | Submit a revision to the focused goal (shows Confirm / Continue Chatting dialog) |
 | `step_complete` | hidden / legacy | Compatibility no-op; Sisyphus no longer requires a step counter |
 | `create_goal` | hidden | Direct calls are rejected; normal creation goes through `propose_goal_draft` |
 
@@ -227,7 +228,7 @@ The shipped gates are intentionally small and mechanical.
 | Completion auditor gate | Archiving completion unless an independent pi auditor agent returns `<approved/>` |
 | Abort gate | Aborting missing, stale, completed, or reasonless goals |
 | Direct-create rejection | Hidden `create_goal` calls creating goals without the confirmation flow |
-| Post-stop block | Continuing to call tools after `pause_goal`, `abort_goal`, `update_goal`, or `apply_goal_tweak` stops the turn |
+| Post-stop block | Continuing to call tools after `pause_goal`, `abort_goal`, `update_goal`, or `propose_goal_tweak` stops the turn |
 | Empty-turn guard | Pure chat loops that would keep auto-continuing without meaningful goal work |
 | Abort pause | Active goals staying active after user abort / Ctrl-C |
 | Disk reconciliation | External pause/archive/delete/status changes being ignored or overwritten by stale memory |

package/extensions/goal-draft.ts

@@ -50,6 +50,33 @@ export function buildDraftConfirmationText(args: {
 	return lines.join("\n");
 }
 
+export function buildTweakConfirmationText(args: {
+	currentObjective: string;
+	newObjective: string;
+	changeSummary: string;
+	sisyphus: boolean;
+}): string {
+	const lines: string[] = [];
+	const modeLabel = args.sisyphus ? "Sisyphus (prompt/criteria style)" : "Normal goal";
+	lines.push("Goal tweak ready for confirmation.");
+	lines.push("");
+	lines.push("Draft details:");
+	lines.push(`Mode: ${modeLabel}`);
+	lines.push("");
+	lines.push("Change:");
+	lines.push("");
+	lines.push(args.changeSummary);
+	lines.push("");
+	lines.push("Current objective:");
+	lines.push("");
+	lines.push(args.currentObjective);
+	lines.push("");
+	lines.push("Proposed new objective:");
+	lines.push("");
+	lines.push(args.newObjective);
+	return lines.join("\n");
+}
+
 export function evaluateDraftingToolGate(args: {
 	toolName: string;
 	draftingFocus?: GoalDraftingFocus | null;

package/extensions/goal-tool-names.ts

@@ -1,5 +1,5 @@
 export const SISYPHUS_STEP_TOOL_NAME = "step_complete";
-export const TWEAK_APPLY_TOOL_NAME = "apply_goal_tweak";
+export const PROPOSE_TWEAK_TOOL_NAME = "propose_goal_tweak";
 export const PROPOSE_DRAFT_TOOL_NAME = "propose_goal_draft";
 export const CREATE_GOAL_TOOL_NAME = "create_goal";
 export const QUESTION_TOOL_NAME = "goal_question";
@@ -14,7 +14,7 @@ export const GOAL_WORK_TOOL_NAMES = [
 	"update_goal",
 	"pause_goal",
 	ABORT_GOAL_TOOL_NAME,
-	TWEAK_APPLY_TOOL_NAME,
+	PROPOSE_TWEAK_TOOL_NAME,
 	CREATE_GOAL_TOOL_NAME,
 	PROPOSE_DRAFT_TOOL_NAME,
 	QUESTION_TOOL_NAME,
@@ -33,7 +33,6 @@ export const GOAL_PROGRESS_TOOL_NAMES = [
 	"update_goal",
 	"pause_goal",
 	ABORT_GOAL_TOOL_NAME,
-	TWEAK_APPLY_TOOL_NAME,
 	"write",
 	"edit",
 	"bash",

package/extensions/goal.ts

@@ -10,6 +10,7 @@ import {
 } from "./goal-core.ts";
 import {
 	buildDraftConfirmationText,
+	buildTweakConfirmationText,
 	goalDraftingPrompt,
 	validateGoalDraftProposal,
 	type GoalDraftingFocus,
@@ -38,7 +39,7 @@ import {
 	SISYPHUS_STEP_TOOL_NAME,
 	GOAL_PROGRESS_TOOL_NAMES,
 	lifecycleToolNamesForGoalStatus,
-	TWEAK_APPLY_TOOL_NAME,
+	PROPOSE_TWEAK_TOOL_NAME,
 } from "./goal-tool-names.ts";
 import {
 	asRecord,
@@ -106,7 +107,6 @@ import {
 	shouldInjectPostCompactReminder,
 	validateGoalAbort,
 	validateGoalCompletion,
-	validateGoalUpdate,
 	validatePauseGoal,
 	validateResumeGoal,
 } from "./goal-policy.ts";
@@ -128,14 +128,14 @@ const GOAL_PROGRESS_TOOL_SET = new Set<string>(GOAL_PROGRESS_TOOL_NAMES);
 
 /**
  * Tools that are NEVER blocked by the post-stop in-turn block. After pause_goal,
- * abort_goal, update_goal=complete, or apply_goal_tweak fires, the agent should
+ * abort_goal, or update_goal=complete fires, the agent should
  * yield the turn; we block all subsequent tool calls except these read-only inspections.
  */
 const POST_STOP_ALLOWED_TOOL_SET = new Set<string>(POST_STOP_ALLOWED_TOOLS);
 
 /**
  * When non-null, /goal-tweak drafting is in progress for this goal id and the
- * agent is allowed to call apply_goal_tweak. Cleared after the tweak is applied
+ * agent is allowed to call propose_goal_tweak. Cleared after the tweak is applied
  * or when a user-driven turn arrives without a tweak follow-through. This is
  * the schema-level affordance gate that prevents the agent from "tweaking" via
  * arbitrary write/edit calls.
@@ -376,7 +376,7 @@ export default function goalExtension(pi: ExtensionAPI): void {
 	// Per-turn flags reset in turn_start (#4, C9 fix).
 	// goalWorkToolCalledThisTurn: tracks whether a real goal-work tool was called.
 	//   If false at turn_end, we don't queue another autoContinue (empty chat turn).
-	// turnStoppedFor: set by pause_goal / update_goal(complete) / apply_goal_tweak
+	// turnStoppedFor: set by pause_goal / update_goal(complete) / propose_goal_tweak
 	//   after their successful execute. Once set, pi.on("tool_call") blocks all
 	//   subsequent in-turn tool calls except POST_STOP_ALLOWED_TOOLS. This is the
 	//   schema fix for "agent keeps writing files after pause_goal".
@@ -393,18 +393,6 @@ export default function goalExtension(pi: ExtensionAPI): void {
 		lastAccountedAt: null as number | null,
 	};
 
-	const draftingHiddenWorkTools = [
-		"bash",
-		"read",
-		"write",
-		"edit",
-		"grep",
-		"find",
-		"ls",
-		SISYPHUS_STEP_TOOL_NAME,
-		TWEAK_APPLY_TOOL_NAME,
-		CREATE_GOAL_TOOL_NAME,
-	] as const;
 	const goalExecutionWorkTools = ["read", "bash", "edit", "write"] as const;
 
 	function syncGoalTools(): void {
@@ -421,15 +409,16 @@ export default function goalExtension(pi: ExtensionAPI): void {
 			// mechanism. Keep step_complete registered for legacy transcripts, but do
 			// not expose it as an active work tool.
 			active.delete(SISYPHUS_STEP_TOOL_NAME);
-			// apply_goal_tweak is only available during a /goal-tweak drafting flow.
+			// propose_goal_tweak is only available during a /goal-tweak drafting flow.
 			// Note: tweak drafting can run against active OR paused goals.
 			if (state.goal && tweakDraftingFor === state.goal.id) {
-				active.add(TWEAK_APPLY_TOOL_NAME);
+				active.add(PROPOSE_TWEAK_TOOL_NAME);
 				active.add(QUESTION_TOOL_NAME);
 				active.add(QUESTIONNAIRE_TOOL_NAME);
 			} else {
-				active.delete(TWEAK_APPLY_TOOL_NAME);
+				active.delete(PROPOSE_TWEAK_TOOL_NAME);
 			}
+	
 			// Keep the commit tool available and let its validator enforce that a
 			// drafting flow is active. This avoids fragile hidden-tool drift after
 			// question turns, compaction, or active-tool resync.
@@ -1054,11 +1043,11 @@ export default function goalExtension(pi: ExtensionAPI): void {
 		if (!focused) return;
 		const sisyphusOn = focused.sisyphus;
 		const label = sisyphusOn ? "Sisyphus tweak drafting" : "Goal tweak drafting";
-		// Activate the tweak edit-gate so apply_goal_tweak is callable.
+		// Activate the tweak edit-gate so propose_goal_tweak is callable.
 		tweakDraftingFor = focused.id;
 		syncGoalTools();
 		ctx.ui.notify(
-			`${label} started${trimmed ? `: ${truncateText(trimmed, 60)}` : ""}. The agent will interview you and then call apply_goal_tweak.`,
+			`${label} started${trimmed ? `: ${truncateText(trimmed, 60)}` : ""}. The agent will interview you and then propose the revision for you to Confirm.`,
 			"info",
 		);
 		const draftId = `tweak-${focused.id}-${Date.now().toString(36)}`;
@@ -1696,84 +1685,182 @@ export default function goalExtension(pi: ExtensionAPI): void {
 	}));
 
 	pi.registerTool(defineTool({
-		name: "update_goal",
-		label: "Update Goal",
-		description: "Mark the current active or paused pi goal complete when the objective is actually achieved.",
-		promptSnippet: "Mark the active or paused pi goal complete when the objective is achieved.",
+		name: PROPOSE_TWEAK_TOOL_NAME,
+		label: "Propose Goal Tweak",
+		description: "During a /goal-tweak drafting flow, present the revised goal to the user for confirmation. The user sees a Confirm / Continue Chatting dialog — only on Confirm is the goal updated.",
+		promptSnippet: "Propose the revised goal to the user with a Confirm / Continue Chatting dialog.",
 		promptGuidelines: [
-			"Use update_goal with status=complete only when the pi goal objective has actually been achieved and no required work remains.",
-			"Before calling update_goal, summarize the evidence you believe proves completion; the tool will launch an independent pi auditor agent to inspect the workspace and judge the claim.",
-			"The auditor is authoritative: completion is archived only if the auditor report ends with <approved/>. If it ends with <disapproved/> or no approval marker, update_goal is rejected and the goal remains open.",
-			"Do not call update_goal merely because work is stopping, substantial progress was made, or tests passed without covering every requirement.",
-			"Do not use update_goal=complete as an escape hatch when you are blocked. If you are blocked, call pause_goal({reason, suggestedAction?}) instead so the user can intervene.",
-			"For sisyphus goals, do not mark complete until every numbered step has been executed and individually verified against its done criterion.",
-			"If the user gives requirements, feedback, or corrections that differ from the goal objective, the goal is stale. Use update_goal with updatedObjective to sync the objective before continuing work or before marking the goal complete. This ensures the auditor evaluates against the latest requirements.",
-			"If you have just run the test suite successfully and the tests all pass, include a testResults object with the exit code (0) and relevant output. The auditor will see this evidence and can skip re-running the tests.",
+			"Only call propose_goal_tweak inside a /goal-tweak drafting flow (the prompt makes that explicit). It is rejected at any other time.",
+			"newObjective must be the FULL revised objective text, formatted the same way as the original (=== Goal === or === Sisyphus Goal === block). Do NOT pass a diff or partial patch; pass the whole new objective.",
+			"For Sisyphus goals: preserve the Sisyphus style and ordered-plan wording unless the user explicitly asks to remove it.",
+			"changeSummary is a one-sentence description of WHAT changed (for the confirmation dialog, activity log, and tweak log).",
+			"The user will see a full plain-text report plus a [Confirm] / [Continue Chatting] choice. Confirm applies the tweak; Continue Chatting returns control to you to ask follow-up questions.",
+			"If the tool returns 'continue chatting', ask the user what they want changed. Do NOT re-propose the same content immediately; iterate based on their feedback first.",
+			"Do NOT use write/edit/bash to modify the active goal file directly. propose_goal_tweak is the only sanctioned channel.",
 		],
 		parameters: Type.Object({
-			status: Type.Optional(StringEnum([COMPLETE_STATUS] as const, { description: "Set to complete only when the objective is achieved." })),
-			completionSummary: Type.Optional(Type.String({ description: "Concise completion claim and evidence summary passed to the independent auditor agent." })),
-			confirmBypassAuditor: Type.Optional(Type.Boolean({ description: "Set to true to confirm bypassing the independent auditor when it is disabled in settings." })),
-			updatedObjective: Type.Optional(Type.String({ description: "Revised goal objective. Use when the user's requirements have changed mid-flight. The goal remains active so the agent can continue working toward the new objective. Can be combined with status=complete to update the objective before the completion audit." })),
-			testResults: Type.Optional(Type.Object({
-				exitCode: Type.Number({ description: "Exit code of the test run (0 = success)" }),
-				suiteName: Type.Optional(Type.String({ description: "Test suite name, e.g. 'npm test'" })),
-				output: Type.Optional(Type.String({ description: "Last lines of test output showing results" })),
-				timestamp: Type.Optional(Type.String({ description: "ISO timestamp of when tests were run" })),
-			}, { description: "Structured test evidence passed to the auditor so it can skip redundant test re-runs. If you have just run the test suite successfully, include this so the auditor accepts the results without re-running." })),
+			newObjective: Type.String({ description: "The complete revised objective text. For Sisyphus goals, preserve the Sisyphus style unless the user explicitly changes it." }),
+			changeSummary: Type.String({ description: "One-sentence description of what was changed (used in confirmation dialog and tweak log)." }),
 		}),
 		executionMode: "sequential",
-		async execute(_toolCallId, params, signal, _onUpdate, ctx) {
+		async execute(_toolCallId, params, _signal, _onUpdate, ctx) {
 			reconcileFocusedGoalFromDisk(ctx);
+			if (!state.goal) {
+				return {
+					content: [{ type: "text", text: "No goal is set; propose_goal_tweak is a no-op." }],
+					details: goalDetails(state.goal),
+				};
+			}
+			if (tweakDraftingFor !== state.goal.id) {
+				return {
+					content: [{
+						type: "text",
+						text: "propose_goal_tweak REJECTED: no /goal-tweak drafting flow is active for this goal. " +
+							"This tool can only be called during a /goal-tweak drafting interview that the user initiated. " +
+							"If you want to change the goal, ask the user to run /goal-tweak.",
+					}],
+					details: goalDetails(state.goal),
+				};
+			}
+			if (state.goal.status !== "active" && state.goal.status !== "paused") {
+				return {
+					content: [{ type: "text", text: `Goal is ${statusLabel(state.goal)}; cannot apply a tweak.` }],
+					details: goalDetails(state.goal),
+				};
+			}
+			const newObjective = params.newObjective.trim();
+			if (!newObjective) throw new Error("propose_goal_tweak requires a non-empty newObjective.");
+			const changeSummary = params.changeSummary.trim();
+			if (!changeSummary) throw new Error("propose_goal_tweak requires a non-empty changeSummary.");
+
+			// Build the confirmation dialog text.
+			const draftSummary = buildTweakConfirmationText({
+				currentObjective: state.goal.objective,
+				newObjective,
+				changeSummary,
+				sisyphus: !!state.goal.sisyphus,
+			});
+
+			const headless = shouldAutoConfirmProposal({ hasUI: ctx.hasUI, autoConfirmEnv: process.env.PI_GOAL_AUTO_CONFIRM });
 
-			// -- Phase 1: Objective update (quick sync) --
-			// Apply updatedObjective before any completion logic so the completion
-			// flow (if status=complete is also set) reads the latest objective.
-			if (params.updatedObjective !== undefined) {
-				const newObjective = params.updatedObjective.trim();
-				if (!newObjective) throw new Error("update_goal requires a non-empty updatedObjective.");
-				const updateGate = validateGoalUpdate({ goal: state.goal });
-				if (!updateGate.ok) {
+			let decision: "confirm" | "continue";
+			if (headless) {
+				decision = "confirm";
+			} else {
+				try {
+					decision = await showProposalDialog(ctx, draftSummary, state.goal.sisyphus ? "sisyphus" : "goal");
+				} catch (err) {
+					const message = proposalDialogFailureMessage(err);
+					ctx.ui.notify(message, "error");
 					return {
-						content: [{ type: "text", text: updateGate.message }],
+						content: [{ type: "text", text: message }],
 						details: goalDetails(state.goal),
 					};
 				}
-				if (!state.goal) throw new Error("Goal disappeared during objective update.");
+			}
+
+			if (decision === "confirm") {
+				// Apply the tweak: write the new objective to disk authoritatively.
 				const next: GoalRecord = {
 					...state.goal,
 					objective: newObjective,
 					updatedAt: nowIso(),
+					// Clear any prior agent pause reason — the user has redefined the work.
+					pauseReason: undefined,
+					pauseSuggestedAction: undefined,
 				};
+				// IMPORTANT: bypass setGoal() / persist() here. persist() calls
+				// syncGoalPromptFromDisk() which would RE-READ the stale objective
+				// from the still-old goal file on disk and clobber our new objective
+				// before writing. propose_goal_tweak is the authoritative source for
+				// objective changes — the disk is downstream, not upstream. Do the
+				// minimal state update manually:
+				//   1) write the new record to disk authoritatively
+				//   2) update in-memory `goal` to the canonical post-write record
+				//   3) append the state entry and re-sync tools
+				//   4) clear the tweak drafting gate so propose_goal_tweak can't be re-used
 				state.goal = writeActiveGoalFile(ctx, next);
 				pi.appendEntry(STATE_ENTRY, goalDetails(state.goal));
+				tweakDraftingFor = null;
+				// Reset autoContinue counter — plan changed, agent gets a fresh chain.
+				resetGetGoalNudgeState(state.goal.id);
+				// Mark turn-stopped so subsequent in-turn tool calls are blocked.
+				turnStoppedFor = state.goal.id;
+				syncGoalTools();
+				updateUI(ctx);
+				ctx.ui.notify(`Goal tweaked: ${truncateText(changeSummary, 160)}`, "info");
+				// Append ledger event for tweak
 				try {
 					appendGoalEvent(ctx, {
 						type: "goal_tweaked",
 						goalId: state.goal.id,
-						changeSummary: "Objective updated via update_goal",
+						changeSummary,
 						at: state.goal.updatedAt,
 					});
 				} catch {
-					// Ledger append failure should not block update
-				}
-				updateUI(ctx);
-
-				// Quick sync only (no status=complete) — return without terminating
-				if (params.status !== COMPLETE_STATUS) {
-					return {
-						content: [{ type: "text", text: `Goal objective updated.` }],
-						details: goalDetails(state.goal),
-					};
+					// Ledger append failure should not crash tweak
 				}
-				// Fall through: status=complete also set, proceed with completion below
+				return {
+					content: [{
+						type: "text",
+						text: `Goal tweak applied. ${changeSummary}\nStop now; the next continuation will arrive automatically if the goal is active.`,
+					}],
+					details: goalDetails(state.goal),
+					terminate: true,
+				};
 			}
 
+			// "continue" — user wants to keep chatting. Drafting state stays armed.
+			return {
+				content: [{
+					type: "text",
+					text: "Goal tweak refinement requested (Continue Chatting). The tweak was not applied — drafting remains active. Ask the user what they want changed about the revision, then revise and call propose_goal_tweak again. Do not re-propose the same content — wait for the user's input first.",
+				}],
+				details: goalDetails(state.goal),
+			};
+		},
+		renderCall(args, theme) {
+			const summary = typeof args?.changeSummary === "string" ? truncateText(args.changeSummary, 80) : "";
+			return new Text(theme.fg("toolTitle", "propose_goal_tweak ") + theme.fg("muted", summary), 0, 0);
+		},
+		renderResult(result, _options, theme) {
+			return renderGoalResult(result, theme);
+		},
+	}));
+
+	pi.registerTool(defineTool({
+		name: "update_goal",
+		label: "Update Goal",
+		description: "Mark the current active or paused pi goal complete when the objective is actually achieved.",
+		promptSnippet: "Mark the active or paused pi goal complete when the objective is achieved.",
+		promptGuidelines: [
+			"Use update_goal with status=complete only when the pi goal objective has actually been achieved and no required work remains.",
+			"Before calling update_goal, summarize the evidence you believe proves completion; the tool will launch an independent pi auditor agent to inspect the workspace and judge the claim.",
+			"The auditor is authoritative: completion is archived only if the auditor report ends with <approved/>. If it ends with <disapproved/> or no approval marker, update_goal is rejected and the goal remains open.",
+			"Do not call update_goal merely because work is stopping, substantial progress was made, or tests passed without covering every requirement.",
+			"Do not use update_goal=complete as an escape hatch when you are blocked. If you are blocked, call pause_goal({reason, suggestedAction?}) instead so the user can intervene.",
+			"For sisyphus goals, do not mark complete until every numbered step has been executed and individually verified against its done criterion.",
+			"The goal objective is immutable. The agent MUST NOT modify the goal objective on its own initiative. If the user gives requirements, feedback, or corrections that differ from the goal objective, ask the user to run /goal-tweak to revise the goal. Use goal_question to confirm when the change is ambiguous.",
+			"If you have just run the test suite successfully and the tests all pass, include a testResults object with the exit code (0) and relevant output. The auditor will see this evidence and can skip re-running the tests.",
+		],
+		parameters: Type.Object({
+			status: Type.Optional(StringEnum([COMPLETE_STATUS] as const, { description: "Set to complete only when the objective is achieved." })),
+			completionSummary: Type.Optional(Type.String({ description: "Concise completion claim and evidence summary passed to the independent auditor agent." })),
+			confirmBypassAuditor: Type.Optional(Type.Boolean({ description: "Set to true to confirm bypassing the independent auditor when it is disabled in settings." })),
+
+			testResults: Type.Optional(Type.Object({
+				exitCode: Type.Number({ description: "Exit code of the test run (0 = success)" }),
+				suiteName: Type.Optional(Type.String({ description: "Test suite name, e.g. 'npm test'" })),
+				output: Type.Optional(Type.String({ description: "Last lines of test output showing results" })),
+				timestamp: Type.Optional(Type.String({ description: "ISO timestamp of when tests were run" })),
+			}, { description: "Structured test evidence passed to the auditor so it can skip redundant test re-runs. If you have just run the test suite successfully, include this so the auditor accepts the results without re-running." })),
+		}, { additionalProperties: false }),
+		executionMode: "sequential",
+		async execute(_toolCallId, params, signal, _onUpdate, ctx) {
+			reconcileFocusedGoalFromDisk(ctx);
+
 			// -- Phase 2: Status validation --
 			if (params.status !== COMPLETE_STATUS) {
-				if (params.updatedObjective === undefined) {
-					throw new Error("update_goal requires either status=complete or updatedObjective.");
-				}
 				throw new Error("update_goal requires status=complete when marking a goal complete.");
 			}
 
@@ -2074,7 +2161,7 @@ export default function goalExtension(pi: ExtensionAPI): void {
 			};
 		},
 		renderCall(args, theme) {
-			const label = args?.status ?? args?.updatedObjective ? "sync" : "";
+			const label = args?.status ?? "";
 			return new Text(theme.fg("toolTitle", "update_goal ") + theme.fg("success", label), 0, 0);
 		},
 		renderResult(result, _options, theme) {
@@ -2247,110 +2334,6 @@ export default function goalExtension(pi: ExtensionAPI): void {
 		},
 	}));
 
-	pi.registerTool(defineTool({
-		name: TWEAK_APPLY_TOOL_NAME,
-		label: "Apply Goal Tweak",
-		description: "Atomically apply a /goal-tweak revision to the active goal. The ONLY way to modify an active goal's objective. Only available during a /goal-tweak drafting flow.",
-		promptSnippet: "Apply the revised goal objective produced by a /goal-tweak drafting interview.",
-		promptGuidelines: [
-			"Only call apply_goal_tweak inside a /goal-tweak drafting flow (the prompt makes that explicit). It is rejected at any other time.",
-			"newObjective must be the FULL revised objective text, formatted the same way as the original (=== Goal === or === Sisyphus Goal === block). Do NOT pass a diff or partial patch; pass the whole new objective.",
-			"For Sisyphus goals: preserve the Sisyphus style and ordered-plan wording unless the user explicitly asks to remove it.",
-			"changeSummary is a one-sentence description of WHAT changed (for the activity log and pause messages).",
-			"Do NOT use write/edit/bash to modify the active goal file directly. apply_goal_tweak is the only sanctioned channel.",
-			"After apply_goal_tweak returns, stop. Do not begin new task work in the same turn. The system will queue the next continuation.",
-		],
-		parameters: Type.Object({
-			newObjective: Type.String({ description: "The complete revised objective text. For Sisyphus goals, preserve the Sisyphus style unless the user explicitly changes it." }),
-			changeSummary: Type.String({ description: "One-sentence description of what was changed (used in UI notification and tweak log)." }),
-		}),
-		executionMode: "sequential",
-		async execute(_toolCallId, params, _signal, _onUpdate, ctx) {
-			reconcileFocusedGoalFromDisk(ctx);
-			if (!state.goal) {
-				return {
-					content: [{ type: "text", text: "No goal is set; apply_goal_tweak is a no-op." }],
-					details: goalDetails(state.goal),
-				};
-			}
-			if (tweakDraftingFor !== state.goal.id) {
-				return {
-					content: [{
-						type: "text",
-						text: "apply_goal_tweak REJECTED: no /goal-tweak drafting flow is active for this goal. " +
-							"This tool can only be called during a /goal-tweak drafting interview that the user initiated. " +
-							"If you want to change the goal, ask the user to run /goal-tweak.",
-					}],
-					details: goalDetails(state.goal),
-				};
-			}
-			if (state.goal.status !== "active" && state.goal.status !== "paused") {
-				return {
-					content: [{ type: "text", text: `Goal is ${statusLabel(state.goal)}; cannot apply a tweak.` }],
-					details: goalDetails(state.goal),
-				};
-			}
-			const newObjective = params.newObjective.trim();
-			if (!newObjective) throw new Error("apply_goal_tweak requires a non-empty newObjective.");
-			const changeSummary = params.changeSummary.trim();
-			if (!changeSummary) throw new Error("apply_goal_tweak requires a non-empty changeSummary.");
-			const next: GoalRecord = {
-				...state.goal,
-				objective: newObjective,
-				updatedAt: nowIso(),
-				// Clear any prior agent pause reason — the user has redefined the work.
-				pauseReason: undefined,
-				pauseSuggestedAction: undefined,
-			};
-			// IMPORTANT: bypass setGoal() / persist() here. persist() calls
-			// syncGoalPromptFromDisk() which would RE-READ the stale objective
-			// from the still-old goal file on disk and clobber our new objective
-			// before writing. apply_goal_tweak is the authoritative source for
-			// objective changes — the disk is downstream, not upstream. Do the
-			// minimal state update manually:
-			//   1) write the new record to disk authoritatively
-			//   2) update in-memory `goal` to the canonical post-write record
-			//   3) append the state entry and re-sync tools
-			//   4) clear the tweak drafting gate so apply_goal_tweak can't be re-used
-			state.goal = writeActiveGoalFile(ctx, next);
-			pi.appendEntry(STATE_ENTRY, goalDetails(state.goal));
-			tweakDraftingFor = null;
-			// Reset autoContinue counter — plan changed, agent gets a fresh chain.
-			resetGetGoalNudgeState(state.goal.id);
-			// C9 fix: mark turn-stopped so subsequent in-turn tool calls are blocked.
-			turnStoppedFor = state.goal.id;
-			syncGoalTools();
-			updateUI(ctx);
-			ctx.ui.notify(`Goal tweaked: ${truncateText(changeSummary, 160)}`, "info");
-			// Append ledger event for tweak
-			try {
-				appendGoalEvent(ctx, {
-					type: "goal_tweaked",
-					goalId: state.goal.id,
-					changeSummary,
-					at: state.goal.updatedAt,
-				});
-			} catch {
-				// Ledger append failure should not crash tweak
-			}
-			return {
-				content: [{
-					type: "text",
-					text: `Goal tweak applied. ${changeSummary}\nStop now; the next continuation will arrive automatically if the goal is active.`,
-				}],
-				details: goalDetails(state.goal),
-				terminate: true,
-			};
-		},
-		renderCall(args, theme) {
-			const summary = typeof args?.changeSummary === "string" ? truncateText(args.changeSummary, 80) : "";
-			return new Text(theme.fg("toolTitle", "apply_goal_tweak ") + theme.fg("muted", summary), 0, 0);
-		},
-		renderResult(result, _options, theme) {
-			return renderGoalResult(result, theme);
-		},
-	}));
-
 	syncGoalTools();
 
 	pi.on("context", async (event): Promise<{ messages: typeof event.messages } | undefined> => {
@@ -2400,7 +2383,7 @@ export default function goalExtension(pi: ExtensionAPI): void {
 	// #4 + C9 fix + Phase 5 C3: gate in-turn tool calls based on lifecycle state.
 	pi.on("tool_call", async (event, ctx) => {
 		// Post-stop in-turn block (C9 0ad8 fix): after pause_goal / abort_goal /
-		// update_goal=complete / apply_goal_tweak fires in this turn, block all subsequent tool calls except
+		// update_goal=complete / propose_goal_tweak fires in this turn, block all subsequent tool calls except
 		// read-only inspection. Forces the agent to yield the turn instead of "fixing"
 		// the situation by creating extra files etc.
 		if (turnStoppedFor !== null && !POST_STOP_ALLOWED_TOOL_SET.has(event.toolName)) {

package/extensions/prompts/goal-prompts.ts

@@ -48,7 +48,7 @@ If the user explicitly asks to abandon/cancel this goal, or the objective is obs
 
 Do NOT silently invent workarounds, fake completion, or quietly redefine the objective. Do NOT call update_goal=complete to escape a blocker.
 
-Goal evolution: if the user gives requirements, feedback, or corrections that differ from the goal objective, the goal is stale. Propose the updated objective concisely and wait for the user to confirm before continuing. Use update_goal with updatedObjective for narrow focus-area changes, or suggest /goal-tweak for broader revisions (boundaries, constraints, multiple sections). Do NOT mark the goal complete with a stale objective.${sisyphusDisciplineBlock(goal) ? `\n${sisyphusDisciplineBlock(goal)}` : ""}`;
+Goal evolution: if the user gives requirements, feedback, or corrections that differ from the goal objective, the goal is stale. The goal objective is immutable — the agent must NOT modify it autonomously. Propose the updated objective concisely and ask the user to run /goal-tweak to revise it. Do NOT mark the goal complete with a stale objective.${sisyphusDisciplineBlock(goal) ? `\n${sisyphusDisciplineBlock(goal)}` : ""}`;
 }
 
 export function continuationPrompt(goal: GoalRecord): string {
@@ -83,7 +83,7 @@ export function continuationPrompt(goal: GoalRecord): string {
 		"Do not call update_goal unless the goal is complete enough to survive independent semantic auditing. Do not mark a goal complete merely because work is stopping.",
 		"Do not ask the user for confirmation unless there is a real blocker.",
 		"",
-		"Goal evolution: if the user gives requirements, feedback, or corrections that differ from the goal objective, the goal is stale. Propose the updated objective concisely and wait for the user to confirm before continuing. Use update_goal with updatedObjective for narrow focus-area changes, or suggest /goal-tweak for broader revisions (boundaries, constraints, multiple sections). Do NOT mark the goal complete with a stale objective.",
+		"Goal evolution: if the user gives requirements, feedback, or corrections that differ from the goal objective, the goal is stale. The goal objective is immutable — the agent must NOT modify it autonomously. Propose the updated objective concisely and ask the user to run /goal-tweak to revise it. Do NOT mark the goal complete with a stale objective.",
 		"",
 		"If you hit a real blocker (missing credentials, contradictory spec, file/permission you cannot access, dangerous operation pending user approval, or an unclear Sisyphus-style ordered plan), call pause_goal({reason, suggestedAction?}) and stop. If the user explicitly asks to abandon/cancel, or the objective is obsolete, impossible, or unsafe to continue, call abort_goal({reason}) and stop. Do not silently invent workarounds. Do not fake completion. pause_goal and abort_goal are structured lifecycle exits; update_goal=complete is not an escape hatch for blockers.",
 		...(goal.sisyphus ? ["", sisyphusDisciplineBlock(goal)] : []),
@@ -139,17 +139,19 @@ export function goalTweakDraftingPrompt(current: GoalRecord, hint: string): stri
 		...focusItems,
 		"",
 		"When the revision is clear:",
-		"1. Call apply_goal_tweak with:",
+		"1. Call propose_goal_tweak with:",
 		"   - newObjective: the FULL revised objective text, formatted the same way as the original" + (sisyphusOn
 			? " === Sisyphus Goal === block (Objective / Success criteria / Boundaries / Constraints / If blocked / Sisyphus reminder)."
 			: " === Goal === block (Objective / Success criteria / Boundaries / Constraints / If blocked)."),
 		"   - changeSummary: one sentence describing what changed.",
-		"2. apply_goal_tweak is the ONLY sanctioned way to change an active goal's objective. It atomically updates the goal record and the on-disk file. Do not attempt to bypass it.",
-		"3. After apply_goal_tweak returns, stop. If the goal is active, the next continuation will arrive automatically. If the goal is paused, the user will resume it explicitly. Either way, do not begin task work in this same turn.",
+		"2. propose_goal_tweak opens the user's Confirm / Continue Chatting dialog.",
+		"   - Confirm applies the tweak. Stop; the next continuation will arrive automatically if the goal is active.",
+		"   - Continue Chatting means the drafting stays active — ask the user what they want changed, then revise and call propose_goal_tweak again.",
+		"3. propose_goal_tweak is the ONLY sanctioned way to change an active goal's objective. It atomically updates the goal record and the on-disk file. Do not attempt to bypass it.",
 		"",
 		"Edge cases:",
-		"- If you decide no change is actually needed, say so clearly in one sentence and stop without calling apply_goal_tweak.",
-		"- If the hint conflicts with the existing goal in a major way, propose two or three concrete alternative revisions and let the user pick before calling apply_goal_tweak.",
+		"- If you decide no change is actually needed, say so clearly in one sentence and stop without calling propose_goal_tweak.",
+		"- If the hint conflicts with the existing goal in a major way, propose two or three concrete alternative revisions and let the user pick before calling propose_goal_tweak.",
 	].join("\n");
 }
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "pi-goal-x",
-  "version": "0.10.1",
+  "version": "0.11.0",
   "description": "Goal mode extension for pi: persistent long-running objectives, /goal-set drafting, Sisyphus prompt style, autoContinue, and an above-editor status overlay. Fork of @capyup/pi-goal.",
   "license": "MIT",
   "author": "pi-goal-x contributors",