pi-advisor-flow 0.2.4 → 0.2.7

package/CHANGELOG.md

@@ -0,0 +1,232 @@
+# Changelog
+
+All notable changes to this project are documented here.
+
+The format follows [Keep a Changelog](https://keepachangelog.com/en/1.1.0/), and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
+
+## 0.2.7
+
+### Security
+
+- Escaped every untrusted Advisor prompt region and hardened automatic decision parsing against malformed fenced blocks ([#2](https://github.com/philipbrembeck/pi-advisor/issues/2)).
+- Kept Advisor prompts, models, gates, budgets, disclosure, redaction, integrations, and consent global by no longer applying repository-controlled project `advisor.json` files ([#3](https://github.com/philipbrembeck/pi-advisor/issues/3)).
+- Redacted unterminated oversized PEM blocks before bounded preference or untracked-file content can leave the process ([#4](https://github.com/philipbrembeck/pi-advisor/issues/4)).
+- Bounded and redacted Herdr blocked-state metadata while reliably clearing previously reported labels ([#5](https://github.com/philipbrembeck/pi-advisor/issues/5)).
+- Bounded untracked-file Git probes and switched to NUL-delimited path handling for non-ASCII filenames ([#6](https://github.com/philipbrembeck/pi-advisor/issues/6)).
+
+### Changed
+
+- Advisor settings now load and save globally. Move any intended values from project `.pi/advisor.json` files into the Pi agent directory's global `advisor.json`.
+- Pi's bundled modules (`@earendil-works/pi-ai`, `@earendil-works/pi-coding-agent`, `@earendil-works/pi-tui`, and `typebox`) are declared as optional peer dependencies. Installing the extension no longer installs a copy of them; Pi supplies them at runtime. Version ranges continue to document the supported Pi API.
+- Published packages now include `CHANGELOG.md`.
+- Release workflows pin GitHub Actions to commit SHAs, kept current by Dependabot.
+
+## 0.2.6
+
+### Added
+
+- Draft-aware `ask_advisor` reviews with opaque advice IDs, explicit outcome reporting, trusted-project preferences, and opt-in explicit untracked-file context.
+- Global-only, privacy-minimal outcome JSONL logging with salted advice digests and no raw advice, prompts, paths, repository data, or session identifiers.
+- `.pi/advisor-preferences.md` support for trusted projects; preferences remain untrusted, redacted, capped, and never auto-written.
+
+## 0.2.5
+
+This version was never published to npm; its changes shipped in 0.2.6.
+
+### Added
+
+- Repository change context for consultations, controlled by `advisorGitContext` (`off`, `summary`, `full`, default `summary`) and capped by `advisorGitContextMaxChars`. `summary` discloses changed file names, change status, and line counts; `full` adds the patch. Untracked files are always reported by name only.
+- An optional `gitContext` argument on `ask_advisor` so the Executor can request less, or request more up to the configured allowance. A request above the allowance is narrowed to it and the Advisor is told that a fuller view was withheld.
+- Repository context is escaped and capped as a single labelled untrusted region, so a crafted path cannot end the region early and have following text read as instructions.
+
+### Fixed
+
+- Restored keyboard selection for the Context window slider in advanced `/advisor-settings` mode, matching its position at the top of the screen.
+
+## 0.2.4
+
+### Added
+
+- Simple mode for voluntary `ask_advisor` and `/advisor-manual` consultations without automatic gates, blocks, budgets, or session summaries; privacy and context controls remain active.
+- Persistent `alwaysOn` Advisor-flow activation, including Executor restoration and, while the flow is active, adoption of an explicit `/model` selection as the Executor. `/advisor-off` also turns persistent activation off.
+- Simple-mode settings for voluntary Advisor use and persistent activation.
+- Static `◆ ADVISOR · SOUND` rendering for ordinary Advisor replies beginning with `Verdict: sound`, for both `ask_advisor` results and `/advisor-manual` responses.
+
+### Changed
+
+- Session Advisor Summary now defaults to off.
+
+### Fixed
+
+- `/advisor contextMaxChars=N` now persists, so the supplied limit applies to later consultations instead of reverting at the next tool call.
+- A zero context limit with no targeted focus no longer sends an empty Advisor request that some providers reject.
+- An Advisor gate response that quotes a decision line inside a fenced example is no longer rejected as a duplicate or contradictory decision.
+- Reconstructed Advisor context is assembled without re-joining the accumulated branch on every entry, which removes a slowdown on long sessions with large context limits.
+
+## 0.2.3
+
+### Added
+
+- Optional local secret redaction for reconstructed Advisor context, including user messages, assistant text and tool arguments, compaction summaries, and full tool results.
+- Exact-name Advisor tool disclosure policies: `full` includes call arguments and capped output; `summary` retains only result status and size metadata; `exclude` omits call details and output. Tools without a policy remain `full` for compatibility.
+- `/advisor-settings` controls for secret redaction and inline JSON editing of tool disclosure policies, with validation errors that keep invalid input open for correction.
+- Configuration validation, loading, and persistence for `advisorRedactSecrets` and `advisorToolPolicies`.
+
+### Changed
+
+- Redact eligible tool output before applying line and byte limits so truncated context cannot retain the beginning or end of a matched secret.
+- Reorganize the README around installation, first use, commands, automatic gate behavior, configuration, privacy boundaries, development, and releases.
+
+## 0.2.2
+
+### Fixed
+
+- Reopening `/advisor-settings` now displays values saved earlier in the same Pi session without requiring `/reload`.
+
+## 0.2.1
+
+### Added
+
+- Ultracite lint commands and a Husky pre-commit hook that formats and lints staged TypeScript and JSON files.
+
+### Changed
+
+- Resolved the existing Ultracite lint violations through structural refactors and stronger type boundaries without changing Advisor-flow behavior.
+
+### Fixed
+
+- Empty persisted Executor, Advisor, and reasoning-effort settings now retain their configured defaults.
+- Keep a session blocked after a critical automatic-gate decision or session-blocking gate failure.
+- Preserve custom Advisor context and reasoning settings when saving unrelated changes.
+- Persist an unlimited Advisor-call budget correctly after removing a prior finite limit.
+- Enforce configured Advisor tool-result byte and line limits, including for long Unicode lines.
+- Render automatic and manual Advisor failures in the transcript.
+- Avoid false loop detection for semantic field names such as `update`.
+- Make release automation skip unchanged versions while explicitly dispatching publication after an Action-created tag.
+
+## 0.2.0
+
+### Added
+
+- Separate Markdown consultations from strict automatic loop-gate decisions.
+- Typed gate parsing for `proceed`, `revise`, and `blocked`, including safe failure classification.
+
+### Changed
+
+- Normal Advisor and Executor-requested consultations preserve raw Markdown and no longer fabricate or enforce a structured verdict.
+- Automatic gate decisions render separately from their Markdown explanation.
+- Advisor calls now use one shared per-session budget with explicit used/remaining accounting.
+- Gate failures support `block-session`, `block-tool`, and `warn-and-continue`; Herdr failures also show sanitized `notification.show` toasts when integration is enabled.
+- Advisor settings validate values at startup, preserve unknown fields on save, and expose Herdr integration plus tool-result limits.
+- Advisor context keeps complete semantic entries and caps oversized tool results using Pi-compatible defaults while preserving head/tail sections.
+- Tool-result limits are configurable by line and byte count, with explicit omission markers that never split semantic entries.
+- Loop detection now uses explainable normalized tool signatures with allowlisted volatile-field and shell-whitespace normalization.
+- Local ephemeral summaries distinguish Markdown advice from automatic gate decisions and include triggers, models, usage/cost when available, budget, failures, and execution effects.
+
+## [0.1.9]
+
+### Fixed
+
+- Keep manual and automatic Advisor responses human-readable. Manual consultations return direct Markdown; automatic loop reviews use a concise Markdown `Decision:` line for machine-readable gating without exposing a JSON protocol.
+
+## [0.1.8]
+
+### Added
+
+- Structured Advisor verdicts: `proceed`, `revise`, `insufficient-evidence`, and critical `blocked` responses, with findings, required verification, and a smallest next step.
+- Critical-block handling: optionally abort the active run, mark the session blocked, and report the blocked state to Herdr.
+- Automatic loop gate that consults the Advisor after three equivalent tool calls. A `proceed` verdict resumes execution; `revise` and `insufficient-evidence` block only the repeated action; critical verdicts, failed reviews, and exhausted budgets block the session and report Herdr state.
+- Per-session Advisor-call limit, with an Executor prompt hint only when a finite limit is configured.
+- Local, in-memory-only `[Session Advisor Summary]` after a non-blocked settled run; no summary data is persisted or sent to Herdr.
+- `/advisor-settings` controls for critical blocking, enabling/disabling the automatic loop gate, loop threshold, max Advisor calls per session, and the Session Advisor Summary.
+- Session-state tests covering loop detection, Advisor-call budgets, and summary generation.
+- Research note covering evidence-backed Advisor-flow improvements.
+
+### Changed
+
+- Advisor responses now require validated JSON and safely fall back to `insufficient-evidence` when the response is malformed.
+- Manual, Executor-requested, and automatic Advisor consultations share the configured session call limit.
+- Herdr activity and blocked state use separate extension metadata sources so clearing one does not clear the other.
+
+
+## [0.1.7]
+
+### Added
+
+- Herdr integration: Advisor consultations display as `seeking advice` while active when Pi runs in a Herdr-managed pane.
+
+## [0.1.6]
+
+### Added
+
+- `/advisor-manual [focus]` to start an Advisor consultation in parallel without interrupting the Executor's active tool work; the completed advice is delivered before the Executor's next model call.
+- Immediate transcript entries and rendered Advisor responses for manual consultations.
+
+### Changed
+
+- Reuse the Advisor call UI for manual consultations and cancel an earlier manual request when a newer one starts or the session shuts down.
+
+## [0.1.5]
+
+### Added
+
+- `/advisor-settings`: one keyboard-navigable screen for Advisor context size, reasoning effort, invocation gates, response collapsing, and a custom invocation rule.
+- Claude Code-style Advisor context selector with `0`, `10k`, `25k`, `100k`, `200k`, and `ALL` presets.
+- Individually configurable plan, repeated-failure, and completion-review Advisor gates.
+- Optional collapsed Advisor responses that expand with `Ctrl+O`.
+- Inline custom invocation-rule editing in Advisor settings.
+
+### Changed
+
+- General `ask_advisor({})` consultations now send conversation context without an invented request or question; targeted questions remain optional.
+- Advisor instructions explicitly tell the Executor not to invent a question for a normal review and tell the Advisor to make a best-effort contextual review without requesting more input.
+- Preserve unknown fields when saving `advisor.json`.
+- Support `0` as a no-history context setting and `Number.MAX_SAFE_INTEGER` as the ALL-context sentinel.
+
+### Fixed
+
+- Ignore persisted Advisor configuration files with invalid field types instead of crashing during model resolution.
+- Restore interactive Advisor settings arrow-key navigation using Pi TUI key matching.
+
+## [0.1.4]
+
+### Added
+
+- Configurable reconstructed-conversation limit via `contextMaxChars` in `advisor.json` or `/advisor contextMaxChars=N` (default: 15,000; maximum: 1,000,000).
+
+### Changed
+
+- Clarified that the Executor may call `ask_advisor({})` without a question for a general review.
+- Removed the extra no-question “General task review” text from the Advisor call UI.
+- Reframed Advisor guidance as a brief second opinion that stress-tests the Executor's own candidate direction rather than taking over planning.
+
+## [0.1.3]
+
+### Documentation
+
+- Changed publication flow, no code changes
+
+## [0.1.2]
+
+### Added
+
+- General contextual Advisor reviews: the Executor can call `ask_advisor({})` without a specific question.
+- A skill-style Advisor invocation row that distinguishes an Executor request from an Advisor response.
+- Markdown rendering support for the Advisor response, including code blocks and inline code.
+
+### Changed
+
+- Advisor responses display the advising model and advice separately from the tool-result payload.
+- The Advisor spinner is shown only while a response is streaming and is cleared when the response completes.
+
+## [0.1.1]
+
+### Documentation
+
+- Fixed documentation link
+
+## [0.1.0]
+
+### Added
+
+- Initial npm and git package release.

package/README.md

@@ -1,22 +1,23 @@
 # pi-advisor
 
 <div align="center">
-  <img src="https://raw.githubusercontent.com/philipbrembeck/pi-advisor/refs/heads/main/assets/screenshot.png" alt="Pi Advisor consultation in the terminal" width="760">
 
-A configurable second-opinion workflow for <a href="https://github.com/earendil-works/pi">Pi</a> coding agents.
+![Pi Advisor consultation in the terminal](https://raw.githubusercontent.com/philipbrembeck/pi-advisor/refs/heads/main/assets/screenshot.png)
 
-</div>
+A configurable second-opinion workflow for <a href="https://github.com/earendil-works/pi">Pi</a> coding agents, inspired by the ["Steering Black-Box LLMs with Advisor Models" paper](https://arxiv.org/abs/2510.02453) and Claude's [Advisor](https://code.claude.com/docs/en/advisor) feature.
 
-Keep the work to the executor model, while the advisor steers it.
+</div>
 
-This extension introduces a strategic "Executor/Advisor" workflow, inspired by Claudes [Advisor](https://code.claude.com/docs/en/advisor).
+This extension introduces a strategic "Executor/Advisor" workflow.
 
 `pi-advisor-flow` keeps one model focused on execution and makes a second, smarter model available for consequential decisions, stalled work, and final reviews. The Executor still owns the work. The Advisor provides a concise review, answers questions and can provide help; it does not take over planning or run tools.
 
-[Read more about Advisors here](https://philipbrembeck.com/writings/2026/07/only-as-much-intelligence-as-you-need).
+The concept is simple, keep implementation on a fast model, borrow frontier reasoning only when decisions actually matter. [Read more about Advisors here](https://philipbrembeck.com/writings/2026/07/only-as-much-intelligence-as-you-need).
 
 ## Install
 
+Requires Pi 0.80.7 or later. The extension installs no dependencies of its own; Pi supplies the modules it uses at runtime.
+
 Install into your Pi agent environment:
 
 ```bash
@@ -58,7 +59,9 @@ Enable with models in one command when preferred:
 
 ### `ask_advisor`
 
-The Executor calls `ask_advisor({})` for a general review of the current task and reconstructed conversation. It can pass a `question` for a targeted review.
+The Executor calls `ask_advisor({})` for a general review of the current task and reconstructed conversation. It can pass a `question` for a targeted review, or a concise `draft` for plan and completion reviews. A draft should name proposed work, validation, and remaining risks; it is an unverified claim, not evidence.
+
+Successful calls return an opaque `adviceId`. When global outcome logging is enabled, the Executor may voluntarily call `record_advisor_outcome` once with that ID, an adoption value, and a final validation status.
 
 Use the Advisor after the Executor has investigated and formed a candidate direction. It is intended to challenge assumptions, expose risks, and confirm the next verification step—not to replace the Executor's work.
 
@@ -101,7 +104,7 @@ Malformed, missing, duplicate, or contradictory decisions are gate failures. The
 
 ## Settings and configuration
 
-`/advisor-models` and `/advisor-settings` save to `advisor.json` in the Pi agent directory. If a trusted project already has its own configuration, Pi uses that file instead.
+`/advisor-models` and `/advisor-settings` save to global `advisor.json` in the Pi agent directory. Repository-controlled project `advisor.json` files are not applied; models, prompts, gates, budgets, disclosure, redaction, integrations, and consent remain under the user's global configuration.
 
 All fields are optional. This example shows the available settings and their normal defaults:
 
@@ -126,6 +129,8 @@ All fields are optional. This example shows the available settings and their nor
   "gateFailureMode": "block-session",
 
   "advisorSessionSummary": false,
+  "advisorGitContext": "summary",
+  "advisorGitContextMaxChars": 20000,
   "simpleMode": false,
   "alwaysOn": false,
   "advisorHerdrIntegration": true,
@@ -133,6 +138,8 @@ All fields are optional. This example shows the available settings and their nor
   "advisorToolResultMaxBytes": 51200,
 
   "advisorRedactSecrets": false,
+  "advisorUntrackedContent": false,
+  "advisorOutcomeLogging": false,
   "advisorToolPolicies": {
     "bash": "summary",
     "deploy": "exclude"
@@ -146,6 +153,27 @@ All fields are optional. This example shows the available settings and their nor
 - `alwaysOn` defaults to `false`. When enabled, Pi restores the configured Executor and activates `ask_advisor` for new, resumed, forked, and reloaded sessions. While the Advisor flow is active, an explicit `/model` selection becomes the persisted Executor for the next activation; a model restored with a session does not change the saved Executor. `/advisor-off` turns `alwaysOn` off so the flow stays disabled in later sessions.
 - In Simple mode, settings keeps the Context window/history slider alongside Simple mode and Always on; advanced values remain saved and take effect when Simple mode is disabled.
 
+### Repository context
+
+- `advisorGitContext` defaults to `summary`. It controls how much of the working tree reaches the Advisor:
+  - `off` sends no repository information.
+  - `summary` sends changed file names, change status, and line counts. It never sends file contents.
+  - `full` additionally sends the patch.
+- Changes are measured against the last commit and cover staged and unstaged work. Untracked files are always listed by name only; their contents are never sent by `gitContext: full`.
+- `advisorUntrackedContent` defaults to false. When enabled, `includeUntracked` can attach only exact named, repository-relative, untracked regular files. Files are redacted and capped before egress; sibling files remain withheld.
+- `advisorGitContextMaxChars` defaults to `20000`. Repository context may claim its own cap or half of `contextMaxChars`, whichever is smaller, so it cannot crowd out the conversation.
+- The Executor may pass `gitContext` to `ask_advisor` as `none`, `summary`, or `full`. `advisorGitContext` is the ceiling: a larger request is narrowed to the configured level and the Advisor is told that a fuller view was withheld, so it does not claim verification it could not perform.
+- `summary` deliberately excludes diff hunk headers. Git derives those from surrounding file content, so a hunk header can reproduce a line the change never touched, including a credential.
+- Redaction runs before the region is capped, and repository content is labelled as untrusted data in the request. Paths and patch text are escaped so a crafted path cannot close the region early and have the remainder read as instructions.
+- File names themselves can be sensitive. `summary` withholds file contents, not file names; use `off` when names must not leave the machine.
+- Collection shares a single overall time budget across its git commands and degrades to a stated failure rather than implying a clean tree.
+
+### Project preferences and outcomes
+
+In a trusted project only, `.pi/advisor-preferences.md` may provide a short local brief. It is never written by pi-advisor, is treated as lower-priority untrusted text, and is redacted/capped before egress. Symlinks, unreadable files, and paths outside the project are ignored.
+
+`advisorOutcomeLogging` defaults to false and is global-only: a project config cannot enable it. When enabled, `~/.pi/agent/advisor-outcomes.jsonl` stores bounded rotating JSONL records with only a version, timestamp, salted truncated advice digest, trigger, adoption, and validation status. It stores no prompt, advice, paths, tool output, repository data, session ID, or advice ID.
+
 ### Context and limits
 
 - `contextMaxChars` defaults to `15000`. It preserves complete semantic entries and adds an omission marker rather than splitting a message.
@@ -173,7 +201,7 @@ The optional Session Advisor Summary defaults to off. When enabled, it is local
 
 It distinguishes regular Markdown advice from gate decisions and records the trigger, model, usage/cost when available, failures, budget, and execution effect.
 
-[Herdr](https://github.com/ogulcancelik/herdr) integration is enabled by default. It reports Advisor activity and blocked state through Herdr's metadata paths; disable it with `advisorHerdrIntegration`.
+[Herdr](https://github.com/ogulcancelik/herdr) integration is enabled by default. It reports Advisor activity and a bounded, redacted blocked-state summary through Herdr's metadata paths; disable it with `advisorHerdrIntegration`. Previously reported state is still cleared when integration is disabled.
 
 ## Development
 

package/package.json

@@ -1,4 +1,12 @@
 {
+  "name": "pi-advisor-flow",
+  "version": "0.2.7",
+  "author": "Philip Brembeck",
+  "repository": {
+    "type": "git",
+    "url": "https://github.com/philipbrembeck/pi-advisor.git"
+  },
+  "main": "extensions/index.ts",
   "devDependencies": {
     "@biomejs/biome": "2.5.3",
     "@earendil-works/pi-ai": "^0.80.7",
@@ -12,37 +20,35 @@
     "typescript": "^5.3.3",
     "ultracite": "7.9.4"
   },
-  "name": "pi-advisor-flow",
   "peerDependencies": {
     "@earendil-works/pi-ai": "^0.80.7",
     "@earendil-works/pi-coding-agent": "^0.80.7",
     "@earendil-works/pi-tui": "^0.80.7",
     "typebox": "^1.1.38"
   },
-  "scripts": {
-    "test": "bun test",
-    "typecheck": "tsc --noEmit",
-    "format": "bunx ultracite fix --linter-enabled=false",
-    "lint": "bunx ultracite check",
-    "lint:fix": "bunx ultracite fix",
-    "package:check": "npm pack --dry-run --json >/dev/null",
-    "prepare": "husky"
-  },
-  "lint-staged": {
-    "*.{json,jsonc,ts}": "bun run lint:fix --"
-  },
-  "type": "module",
-  "version": "0.2.4",
-  "license": "MIT",
-  "repository": {
-    "type": "git",
-    "url": "https://github.com/philipbrembeck/pi-advisor.git"
+  "peerDependenciesMeta": {
+    "@earendil-works/pi-ai": {
+      "optional": true
+    },
+    "@earendil-works/pi-coding-agent": {
+      "optional": true
+    },
+    "@earendil-works/pi-tui": {
+      "optional": true
+    },
+    "typebox": {
+      "optional": true
+    }
   },
-  "homepage": "https://github.com/philipbrembeck/pi-advisor",
-  "author": "Philip Brembeck",
   "description": "Advanced Executor/Advisor flow for Pi, fully configurable and extendable.",
-  "main": "extensions/index.ts",
-  "types": "extensions/index.ts",
+  "files": [
+    "extensions",
+    "src",
+    "CHANGELOG.md",
+    "LICENSE",
+    "README.md"
+  ],
+  "homepage": "https://github.com/philipbrembeck/pi-advisor",
   "keywords": [
     "pi-package",
     "pi-extension",
@@ -51,15 +57,28 @@
     "pi-advisor",
     "herdr"
   ],
-  "files": [
-    "extensions",
-    "src",
-    "README.md"
-  ],
+  "license": "MIT",
+  "lint-staged": {
+    "*.{json,jsonc,ts}": "bun run lint:fix --"
+  },
+  "overrides": {
+    "brace-expansion": "5.0.8"
+  },
   "pi": {
     "extensions": [
       "./extensions/index.ts"
     ],
     "image": "https://raw.githubusercontent.com/philipbrembeck/pi-advisor/refs/heads/main/assets/hero.png"
-  }
+  },
+  "scripts": {
+    "test": "bun test",
+    "typecheck": "tsc --noEmit",
+    "format": "bunx ultracite fix --linter-enabled=false",
+    "lint": "bunx ultracite check",
+    "lint:fix": "bunx ultracite fix",
+    "package:check": "npm pack --dry-run --json >/dev/null",
+    "prepare": "husky"
+  },
+  "type": "module",
+  "types": "extensions/index.ts"
 }

package/src/commands.ts

@@ -16,6 +16,7 @@ import {
   loadConfig,
   parseArgs,
   saveConfig,
+  saveGlobalOutcomeLogging,
   setAdvisorAutoLoopGateRef,
   setAdvisorBlockOnBlockedRef,
   setAdvisorCollapseResponsesRef,
@@ -24,9 +25,12 @@ import {
   setAdvisorEffortRef,
   setAdvisorFailureGateRef,
   setAdvisorFailureModeRef,
+  setAdvisorGitContextMaxCharsRef,
+  setAdvisorGitContextRef,
   setAdvisorHerdrIntegrationRef,
   setAdvisorLoopThresholdRef,
   setAdvisorMaxCallsPerSessionRef,
+  setAdvisorOutcomeLoggingRef,
   setAdvisorPlanGateRef,
   setAdvisorRedactSecretsRef,
   setAdvisorRef,
@@ -34,6 +38,7 @@ import {
   setAdvisorToolPoliciesRef,
   setAdvisorToolResultMaxBytesRef,
   setAdvisorToolResultMaxLinesRef,
+  setAdvisorUntrackedContentRef,
   setAlwaysOnRef,
   setContextMaxCharsRef,
   setExecutorEffortRef,
@@ -108,7 +113,12 @@ type ManualConsult = (
   ctx: ExtensionContext,
   question?: string,
   signal?: AbortSignal
-) => Promise<{ markdown: string; thinkingText: string }>;
+) => Promise<{
+  markdown: string;
+  thinkingText: string;
+  draftBytes?: number;
+  preferenceBytes?: number;
+}>;
 type ThinkingLevel = Parameters<ExtensionAPI["setThinkingLevel"]>[0];
 
 const notify = (
@@ -248,7 +258,11 @@ export const registerCommands = (
       pi.setThinkingLevel(executorEffortRef as ThinkingLevel);
     }
     if (!flowEnabled()) {
-      pi.setActiveTools([...pi.getActiveTools(), "ask_advisor"]);
+      pi.setActiveTools([
+        ...pi.getActiveTools(),
+        "ask_advisor",
+        "record_advisor_outcome",
+      ]);
     }
     if (announce) {
       notify(
@@ -459,6 +473,7 @@ export const registerCommands = (
 
   pi.registerCommand("advisor-settings", {
     description: "Configure Advisor context and reasoning effort",
+    // biome-ignore lint/complexity/noExcessiveCognitiveComplexity: one settings form maps every persisted control.
     handler: async (_args, ctx) => {
       loadConfig(ctx);
       if (!ctx.hasUI) {
@@ -505,9 +520,19 @@ export const registerCommands = (
       setAdvisorToolResultMaxLinesRef(settings.toolResultMaxLines ?? 2000);
       setAdvisorToolResultMaxBytesRef(settings.toolResultMaxBytes ?? 50 * 1024);
       setAdvisorRedactSecretsRef(settings.redactSecrets ?? false);
+      setAdvisorGitContextRef(settings.gitContext ?? "summary");
+      setAdvisorGitContextMaxCharsRef(settings.gitContextMaxChars ?? 20_000);
       setAdvisorToolPoliciesRef(settings.toolPolicies ?? {});
+      setAdvisorUntrackedContentRef(settings.untrackedContent ?? false);
+      setAdvisorOutcomeLoggingRef(settings.outcomeLogging ?? false);
       const path = saveConfig(ctx);
-      ctx.ui.notify(`Saved Advisor settings to ${path}`, "info");
+      const globalPath = saveGlobalOutcomeLogging(
+        settings.outcomeLogging ?? false
+      );
+      ctx.ui.notify(
+        `Saved Advisor settings to ${path}; outcome logging globally to ${globalPath}`,
+        "info"
+      );
     },
   });
 
@@ -515,7 +540,12 @@ export const registerCommands = (
     description: "Disable on-demand Advisor calls; keep the current model",
     handler: (_args, ctx) => {
       pi.setActiveTools(
-        pi.getActiveTools().filter((name) => name !== "ask_advisor")
+        pi
+          .getActiveTools()
+          .filter(
+            (name) =>
+              name !== "ask_advisor" && name !== "record_advisor_outcome"
+          )
       );
       // Leaving alwaysOn set would silently reactivate the flow next session.
       const wasAlwaysOn = alwaysOnRef;