pi-mono-all 1.0.0

package/node_modules/pi-mono-context-guard/CHANGELOG.md

@@ -0,0 +1,195 @@
+# pi-mono-context-guard
+
+## 1.7.2
+
+### Patch Changes
+
+### Fixed: ask-user-question
+
+- Remove unused `StringEnum` import from `@mariozechner/pi-ai`.
+
+
+## 1.7.1
+
+### Patch Changes
+
+### Fixed: team-mode
+
+- Widget no longer mislabels blocked or approval-pending teams as "running smoothly" — blockers and pending approvals are now detected via team summaries.
+- Preserve in-flight work on re-emitted `session_start` events instead of tearing the runtime down and SIGTERM-ing live teammates.
+- Auto-relaunch leaders for `running` teams after a session reset; surface failures as both a team signal and a UI notification.
+- `createTeam` now defaults `repoRoots` to `[process.cwd()]` when the caller passes an empty array.
+- Archive `process.json` into `history/` before a new task reuses the same role slot, so the prior task's final state is no longer silently clobbered.
+
+### Enhanced: team-mode
+
+- Durable intent queue for subprocess handoff: `team_spawn_teammate` calls made from a teammate subprocess are written to disk and executed by the main session's `LeaderRuntime` instead of spawning orphaned grand-children.
+- New tool `team_task_create_batch` lets the leader emit the full initial task DAG in one call, removing per-task LLM round-trips during bootstrap.
+- `team_create` / `launchLeader` accept an `awaitBootstrap` option so the user sees the task graph before the tool returns; leader launch retries up to 3 times on transient failures.
+- Persist per-turn debug artifacts (prompt, invocation, stderr, raw event stream) for both leader and teammate subprocesses, exposed via `TeammateSummary.debugArtifacts`.
+- Track `exitCode`, `exitSignal`, `terminationReason`, `stderrTail`, `toolExecutions`, `model` and `modelProvider` on every `TeammateProcess` record.
+- Provider detection now consults pi's `settings.json` and `auth.json` in addition to env vars; default model IDs aligned with the provider/model scheme.
+- `collectPiOutput` supports `AbortSignal` cancellation.
+
+### Tests
+
+- New `intent-queue` and `model-config` suites; expanded coverage across `leader-runtime`, `team-manager`, `team-query-tool` and `formatters`.
+
+
+## 1.7.0
+
+### Minor Changes
+
+### Enhanced: status-line
+
+- Improved progress rendering and colors in expert mode
+
+### Enhanced: team-mode
+
+- **LLM-driven leader** — replaced the hardcoded `research → synthesis → implementation → verification` state machine with a pi subprocess coordinator that authors the task graph via tool calls
+- **New tool `team_task_create`** so the leader can author tasks at runtime
+- **New tool `team_handoff`** for explicit teammate → teammate context handoffs (replaces regex-scraping of `Handoffs:` output sections)
+- **File-based teammate specs** — drop `.claude/teammates/<role>.md` frontmatter files (`name`, `description`, `needsWorktree`, `hasMemory`, `modelTier`) to extend or override the seven built-in roles
+- **Event-driven leader wakes** — mailbox messages addressed to the leader (or broadcast) trigger a debounced (~200ms) cycle instead of waiting for the 20s polling tick
+- **Templates accept any string** — `fullstack` / `research` / `refactor` remain as built-ins, but unknown template keys are accepted and no-op gracefully
+- **Provider config per team** — per-team model overrides via `/team models`
+- Reduced leader overhead and parent-session token churn
+- `spawnTeammate` now always appends the full runtime-built context (signals, mailbox, dependencies, team memory) so teammates get the richer snapshot even when the caller's `context` argument is brief
+
+### Breaking changes: team-mode
+
+- Removed `LeaderPhase` enum and `currentPhase` field from `TeamRecord` / `TeamSummary`
+- Removed `parseExplicitHandoffs` export and the legacy `Handoffs:` output parser — peer handoffs must go through the `team_handoff` tool
+- Removed the deterministic auto-spawn loop (`ensureBootstrapTasks`) — all task authoring and teammate spawning is now the LLM leader's responsibility
+- Removed `StringEnum` gate on `team_create`'s `template` parameter (now plain string)
+
+### Fixed: review
+
+- Annotate diff lines so the model picks correct line numbers
+- Fix slice chunk around lines for comments in the reviewer TUI
+
+### Documentation
+
+- Updated root README and sentinel extension README
+- Documented the new file-based teammate spec format and event-driven leader wake in the team-mode README
+
+## 1.6.0
+
+### Minor Changes
+
+### New Extension: sentinel
+
+Replaced the `grep` extension with a new security-focused `sentinel` extension for monitoring and guarding sensitive operations.
+
+### Enhanced: team-mode
+
+- Added comprehensive test suite with integration tests
+- New mock helpers for subprocess testing
+- Improved signal manager with better error handling
+- Leader runtime refactoring for stability
+- Team query tool with dedicated tests
+
+### Enhanced: status-line
+
+- Added basic and expert mode displays
+- Improved index.ts with better state management
+
+### Enhanced: clear
+
+- Updated keyboard shortcut to `Ctrl+Shift+L`
+- Better busy-state handling for shortcuts
+- Added warning/cancel handling and error notifications
+
+### Enhanced: context-guard
+
+- Improved read deduplication across sessions
+- Added `context-guard:file-modified` event for cache eviction
+
+### Documentation
+
+- Added dedicated README for `clear` extension
+- Added dedicated README for `context-guard` extension
+- Updated main README with improved extension descriptions
+
+## 1.5.0
+
+### Minor Changes
+
+- ### `multi-edit` — diverge from upstream fork
+
+  The extension was originally derived from [mitsuhiko/agent-stuff](https://github.com/mitsuhiko/agent-stuff)'s `pi-extensions/multi-edit.ts`. This release rewrites the largest unmodified subsystems so the implementation is structurally distinct from upstream while keeping the public contract intact.
+
+  - **Modularized layout** — the 953-line `index.ts` is split into purpose-scoped modules: `types.ts`, `workspace.ts`, `classic.ts`, `patch.ts`, `diff.ts`, and a slim `index.ts` (~180 lines of registration + dispatch wiring).
+  - **New patch engine** — `patch.ts` is now a recursive-descent parser over a `LineCursor` class with `indexOf`-based hunk anchoring. Hunks are stored as `{ oldBlock, newBlock }` raw strings (previously `{ oldLines[], newLines[] }` arrays), letting the applier splice content directly instead of reconstructing line arrays per apply.
+  - **Two-pass diff renderer** — `diff.ts` now walks `diffLines` parts into a typed `Entry[]` stream and makes all gutter / context-collapse decisions in a second pass, replacing the prior single-loop state-flag design.
+  - **Polished classic edits** — extracted `groupEditsByPath`, `sortGroupByPosition`, `applyGroupToContent`, and `rollbackSnapshots` helpers; formalized the quote-fallback as an ordered `MATCH_PASSES` array so new normalizers (dashes, NBSP, etc.) can be added by appending one entry.
+  - **First contract test suite** — 34 tests under `__tests__/` cover classic edits (positional reordering, redundant-pair skip, quote fallback, atomic rollback, read-only preflight), patch operations (Add/Delete/Update round-trips, move-rejection, multi-op batches), and diff rendering (line-number gutter, context collapse, add/remove-only cases). Runs via `npm test` (`tsx --test`).
+  - **Dropped Codex apply_patch edge cases** (documented in `README.md` → "Codex apply_patch compatibility"): `*** End of File` sentinel hunks, 4-pass fuzzy `seekSequence` matching, implicit first hunk without `@@`, whitespace-tolerant anchoring. Common paths (Add/Delete/Update-single-chunk, Update with multiple hunks, Add+Update+Delete batches) are fully tested and preserved.
+  - **README attribution** — new "Origins" section crediting `mitsuhiko/agent-stuff` as the original source.
+
+## 1.4.0
+
+### Minor Changes
+
+- Add teammate progress heartbeats and widget refresh improvements to team mode.
+
+## 1.3.0
+
+## 1.2.0
+
+### Minor Changes
+
+- ### `multi-edit` — robustness improvements
+
+  - **No-op write guard**: skip file write and `context-guard:file-modified` event when new content is identical to what was last read — prevents unnecessary watcher churn
+  - **Early write-access check**: virtual workspace `checkWriteAccess` now validates real-filesystem permissions during the preflight pass so read-only files fail fast before any real file is touched
+  - **Curly-quote normalization**: new `findActualString` helper falls back to normalized quote matching (`"` / `'` ↔ `"` / `'`) when exact `oldText` search fails — the most common class of preflight mismatch
+  - **Atomic batch rollback**: `applyClassicEdits` gains a `rollbackOnError` option that restores all successfully written files when a later edit in the same batch fails
+
+  ### `ask-user-question` — UX fixes
+
+  - **Reliable text capture on submit**: answer is read directly from the editor before it clears itself, fixing a race where the stored value was always empty
+  - **Unified advance logic**: `advanceTab()` and `saveOtherModeText()` helpers replace scattered single-question fast-paths — behaviour is now consistent regardless of form length
+  - **Auto-advance on Enter / Tab**: pressing Enter or Tab in any question (text, radio with "Other", checkbox with "Other") advances to the next tab without requiring a separate click
+
+  ### `team-mode` — stability fixes
+
+  - **Infinite retry loop eliminated**: subprocess guard (`PI_TEAM_SUBPROCESS=1`) prevents spawned pi subprocesses from launching a ghost `LeaderRuntime` that immediately marks in-progress tasks as stalled
+  - **Stall detection grace period**: tasks updated within the last 2 × `LEADER_POLL_MS` (10 s) are skipped by `detectStalledTasks` — eliminates false positives on the spawning cycle
+  - **Circuit breaker**: tasks that stall more than `MAX_TASK_RETRIES` (3) times are permanently cancelled with a clear error signal instead of being silently re-queued
+  - **Concurrent cycle guard**: `runLeaderCycle` returns early if a cycle is already in-flight for the same team, preventing overlapping read-modify-write from the poll interval and completion handlers
+  - **Widget cleanup**: cancelled and completed teams are no longer shown in the team widget — only `initializing | running | paused | failed` states are displayed
+  - **Shorter auto-generated names**: `objectiveToName` now splits on non-alphanumeric characters (handles path separators), filters stopwords and extreme-length tokens, and hard-caps at 32 characters
+
+## 1.1.1
+
+### Patch Changes
+
+- chore: update all packages for consistency and include team-mode fixes
+
+## 1.1.0
+
+### Minor Changes
+
+- Add context-guard and grep extensions; improve multi-edit with dedup
+
+  **New: `pi-mono-context-guard`**
+  Extension that keeps the LLM context window lean with three guards:
+
+  - `read` without `limit` → auto-injects `limit=120`
+  - Read dedup → mtime-based stub for unchanged files (~20 tokens vs full content re-send)
+  - `bash` with unbounded `rg` → appends `| head -60`
+
+  Listens to `context-guard:file-modified` events to invalidate the dedup cache after edits.
+  `/context-guard` command to inspect and toggle guards at runtime.
+
+  **New: `pi-mono-grep`**
+  Dedicated ripgrep wrapper tool. Replaces raw `rg` in bash with a structured tool that has
+  `head_limit=60` built into the schema, `output_mode` (files_with_matches / content / count),
+  pagination via `offset`, and automatic VCS directory exclusions.
+  Prompt guidelines instruct the model to always use `grep` instead of bash+rg.
+
+  **Updated: `pi-mono-multi-edit`**
+
+  - Per-call read cache in `createRealWorkspace` deduplicates disk reads within a single `execute()` invocation (preflight + real-apply)
+  - Emits `context-guard:file-modified` event after every real `writeText` and `deleteFile` so context-guard can evict stale dedup cache entries

package/node_modules/pi-mono-context-guard/README.md

@@ -0,0 +1,81 @@
+# context-guard extension
+
+Keeps pi sessions lean by preventing common context-window waste during tool use.
+
+## What it does
+
+This extension intercepts tool calls and applies three safeguards:
+
+1. **Auto-limit `read` calls**  
+   If the model calls `read` without a `limit`, `context-guard` injects a default limit of `120` lines and shows a notification. The model can continue with `offset` to paginate.
+
+2. **Deduplicate unchanged `read` calls**  
+   If the same file is read again with the same `offset` and `limit`, and the file has not changed on disk, the extension blocks the duplicate read and returns a short stub instead of sending the file content again.
+
+3. **Bound raw `rg` output in `bash`**  
+   If a `bash` command uses `rg` without an output-bounding operator such as `head`, `tail`, or `wc`, the extension appends `| head -60` automatically.
+
+## Why it helps
+
+These guards reduce unnecessary token usage and make it harder for long sessions to burn context on repeated or unbounded file output.
+
+## Current defaults
+
+- `read` auto-limit: `120`
+- `rg` head limit: `60`
+- read guard: enabled
+- read dedup guard: enabled
+- raw `rg` guard: enabled
+
+## Read dedup behavior
+
+The dedup cache is scoped to the current session.
+
+A cached `read` entry is only reused when all of the following are true:
+
+- the same file path is requested
+- the same `offset` is requested
+- the same `limit` is requested
+- the file's modification time has not changed
+
+When a file changes, the cache entry is invalidated. The extension also listens for the `context-guard:file-modified` event so companion extensions such as `multi-edit` can evict stale cache entries immediately after writes.
+
+## Example behaviors
+
+### `read` without a limit
+
+Input:
+
+```json
+{ "path": "src/index.ts" }
+```
+
+Effective call:
+
+```json
+{ "path": "src/index.ts", "limit": 120 }
+```
+
+### Duplicate `read` of an unchanged file
+
+Instead of re-sending the file contents, the extension returns a short message telling the agent to reuse the earlier `read` result.
+
+### Raw `rg` inside `bash`
+
+Input:
+
+```bash
+rg "TODO" src
+```
+
+Effective command:
+
+```bash
+rg "TODO" src | head -60
+```
+
+## Files
+
+- `index.ts` — extension entry point
+- `package.json` — package metadata
+- `CHANGELOG.md` — release history

package/node_modules/pi-mono-context-guard/index.ts

@@ -0,0 +1,212 @@
+/**
+ * context-guard — keep the LLM context window lean.
+ *
+ * Intercepts tool calls before they execute and applies three guards:
+ *
+ * 1. `read` without `limit`
+ *    → auto-injects `limit: DEFAULT_READ_LIMIT` and notifies the user.
+ *      The model can paginate with `offset` if it needs more.
+ *
+ * 2. `read` for a file already seen this session (mtime unchanged)
+ *    → blocks the call and returns a stub:
+ *      "File unchanged since last read — refer to the earlier result."
+ *      (~20 tokens vs re-sending the full content). Evicted when
+ *      multi-edit emits `context-guard:file-modified`.
+ *
+ * 3. `bash` using `rg` without any output-bounding operator
+ *    → appends `| head -N` so grep dumps don't fill the context window.
+ *
+ * All guards are enabled by default.
+ */
+
+import { stat } from "node:fs/promises";
+import { resolve } from "node:path";
+
+import type { ExtensionAPI } from "@mariozechner/pi-coding-agent";
+import { isToolCallEventType } from "@mariozechner/pi-coding-agent";
+
+// ---------------------------------------------------------------------------
+// Defaults
+// ---------------------------------------------------------------------------
+
+const DEFAULTS = {
+	readLimit: 120,
+	rgHeadLimit: 60,
+	readGuard: true,
+	dedupGuard: true,
+	rgGuard: true,
+};
+
+type Config = typeof DEFAULTS;
+
+// ---------------------------------------------------------------------------
+// Read dedup cache
+// ---------------------------------------------------------------------------
+
+/** What we remember about a past read. */
+type ReadEntry = {
+	/** mtime in milliseconds at the time of the read. */
+	mtimeMs: number;
+	/** offset used (undefined = start of file). */
+	offset: number | undefined;
+	/** limit used (undefined = whole file). */
+	limit: number | undefined;
+};
+
+const FILE_UNCHANGED_STUB =
+	"File unchanged since last read. The content from the earlier Read " +
+	"tool_result in this conversation is still current — refer to that " +
+	"instead of re-reading.";
+
+// ---------------------------------------------------------------------------
+// Helpers
+// ---------------------------------------------------------------------------
+
+function usesUnboundedRg(cmd: string): boolean {
+	if (!/(?:^|[|;&\s])rg\s/.test(cmd)) return false;
+	if (/\|\s*(?:head|tail|wc|less|more|grep\s+-c)/.test(cmd)) return false;
+	if (/\brg\b[^|]*\s(?:-l|--files-with-matches|-c|--count|--json)\b/.test(cmd)) return false;
+	return true;
+}
+
+function appendHead(cmd: string, n: number): string {
+	return `${cmd.trimEnd().replace(/;+$/, "").trimEnd()} | head -${n}`;
+}
+
+// ---------------------------------------------------------------------------
+// Extension
+// ---------------------------------------------------------------------------
+
+export default function (pi: ExtensionAPI): void {
+	const cfg: Config = { ...DEFAULTS };
+
+	/** Session-scoped read cache: absolute path → last-seen read metadata. */
+	const readCache = new Map<string, ReadEntry>();
+
+	// -------------------------------------------------------------------------
+	// Cache invalidation — fired by multi-edit after every real file write
+	// -------------------------------------------------------------------------
+	pi.events.on("context-guard:file-modified", (data: { path: string }) => {
+		if (data?.path) {
+			readCache.delete(resolve(data.path));
+		}
+	});
+
+	// -------------------------------------------------------------------------
+	// Reset cache on new session
+	// -------------------------------------------------------------------------
+	pi.on("session_start", async () => {
+		readCache.clear();
+	});
+
+	// -------------------------------------------------------------------------
+	// Guard 1 + 2: read — limit injection + dedup
+	// -------------------------------------------------------------------------
+	pi.on("tool_call", async (event, ctx) => {
+		if (!isToolCallEventType("read", event)) return;
+
+		// Guard 1: inject limit if missing
+		if (cfg.readGuard && event.input.limit === undefined) {
+			event.input.limit = cfg.readLimit;
+			ctx.ui.notify(
+				`[context-guard] read: auto-limit=${cfg.readLimit} (use offset to paginate)`,
+				"info",
+			);
+		}
+
+		// Guard 2: dedup — block if file is unchanged since last read
+		if (!cfg.dedupGuard) return;
+
+		const rawPath = event.input.path;
+		if (!rawPath) return;
+
+		// Normalise path the same way pi does (strip leading @, resolve relative)
+		const absolutePath = resolve(
+			ctx.cwd,
+			rawPath.startsWith("@") ? rawPath.slice(1) : rawPath,
+		);
+
+		const entry = readCache.get(absolutePath);
+		if (!entry) return;
+
+		// Only dedup exact range matches
+		const sameOffset = entry.offset === (event.input.offset ?? undefined);
+		const sameLimit  = entry.limit  === (event.input.limit  ?? undefined);
+		if (!sameOffset || !sameLimit) return;
+
+		// Check mtime — if the file changed on disk, let it through
+		try {
+			const { mtimeMs } = await stat(absolutePath);
+			if (mtimeMs !== entry.mtimeMs) {
+				readCache.delete(absolutePath);
+				return;
+			}
+		} catch {
+			// stat failed (file deleted, permission error, etc.) — let tool handle it
+			readCache.delete(absolutePath);
+			return;
+		}
+
+		// File is unchanged — block the call and return the stub
+		ctx.ui.notify(`[context-guard] read dedup: ${rawPath} unchanged`, "info");
+		return {
+			block: true,
+			reason: FILE_UNCHANGED_STUB,
+		};
+	});
+
+	// -------------------------------------------------------------------------
+	// Populate cache after a successful read
+	// -------------------------------------------------------------------------
+	pi.on("tool_result", async (event) => {
+		if (!cfg.dedupGuard) return;
+		if (event.toolName !== "read") return;
+		if (event.isError) return;
+
+		const rawPath = (event.input as { path?: string }).path;
+		if (!rawPath) return;
+
+		// Only cache full successful text reads (not images, PDFs, etc.)
+		const resultText = event.content
+			.filter((b): b is { type: "text"; text: string } => b.type === "text")
+			.map((b) => b.text)
+			.join("");
+
+		// Skip stubs injected by us — don't overwrite the real entry
+		if (resultText === FILE_UNCHANGED_STUB) return;
+
+		const absolutePath = resolve(
+			(event.input as { path?: string; cwd?: string }).cwd ?? "",
+			rawPath.startsWith("@") ? rawPath.slice(1) : rawPath,
+		);
+
+		try {
+			const { mtimeMs } = await stat(absolutePath);
+			readCache.set(absolutePath, {
+				mtimeMs,
+				offset: (event.input as { offset?: number }).offset ?? undefined,
+				limit:  (event.input as { limit?: number }).limit   ?? undefined,
+			});
+		} catch {
+			// best-effort only
+		}
+	});
+
+	// -------------------------------------------------------------------------
+	// Guard 3: bash — rg without head/tail/wc
+	// -------------------------------------------------------------------------
+	pi.on("tool_call", async (event, ctx) => {
+		if (!cfg.rgGuard) return;
+		if (!isToolCallEventType("bash", event)) return;
+
+		const cmd = event.input.command ?? "";
+		if (usesUnboundedRg(cmd)) {
+			event.input.command = appendHead(cmd, cfg.rgHeadLimit);
+			ctx.ui.notify(
+				`[context-guard] bash: appended | head -${cfg.rgHeadLimit} to rg`,
+				"info",
+			);
+		}
+	});
+
+}

package/node_modules/pi-mono-context-guard/package.json

@@ -0,0 +1,23 @@
+{
+  "name": "pi-mono-context-guard",
+  "version": "1.7.2",
+  "description": "Pi extension that guards context window growth by auto-limiting read and rg output",
+  "keywords": [
+    "pi-package",
+    "pi-extension"
+  ],
+  "peerDependencies": {
+    "@mariozechner/pi-coding-agent": "*",
+    "@sinclair/typebox": "*"
+  },
+  "pi": {
+    "extensions": [
+      "./index.ts"
+    ]
+  },
+  "repository": {
+    "type": "git",
+    "url": "git+https://github.com/emanuelcasco/pi-mono-extensions.git",
+    "directory": "extensions/context-guard"
+  }
+}

package/node_modules/pi-mono-figma/CHANGELOG.md

@@ -0,0 +1,59 @@
+# pi-mono-figma
+
+## 0.2.1
+
+### Patch Changes
+
+- Add the all-in-one pi package and bundle the shared pi-common workspace package into distributed packages.
+
+## 0.2.0
+
+### Minor Changes
+
+### Added: LLM-ready Figma workflow
+
+- Added `figma_get_node_summary`, `figma_extract_text`, `figma_explain_node`, and `figma_get_implementation_context` processed tools.
+- Added compact node summarization that ignores hidden nodes, vector internals, and component instance internals by default.
+- Changed `figma_get_design_context` to return compact top-level file context or target-node summaries instead of full recursive file structure.
+- Updated tool descriptions, README, and skill guidance to prefer processed tools and keep raw JSON tools as debugging escape hatches.
+- Added response caps, truncation metadata, and next-step suggestions for summarized node output.
+
+### Added: Dev Mode parity helpers
+
+- Added golden fixture tests for Figma summarization and design-to-code helper behavior.
+- Added `figma_find_nodes_by_name` and `figma_find_nodes_by_text` for compact path-aware layer/text search.
+- Enriched `figma_get_implementation_context` with CSS layout, responsive, accessibility, design token, framework, and starter snippet hints.
+- Added `figma_extract_assets` for SVG/icon exports, node renders, image fill downloads, hashes, byte sizes, suggested names, and node-path manifests.
+- Added `figma_find_code_connect_mapping` for bounded local Code Connect/Figma reference discovery.
+- Added `figma_get_component_implementation_hints` for higher-level component implementation guidance.
+- Documented live selection parity as future plugin/bridge work rather than a REST baseline feature.
+
+## 0.1.2
+
+### Patch Changes
+
+### Fixed: package extension entrypoint
+
+- Added the package root `index.ts` extension entrypoint so pi can load the Figma tools from the published package manifest.
+- Documented the extension's benefits over Figma MCP in the package README.
+
+## 0.1.1
+
+### Patch Changes
+
+### Enhanced: auth setup
+
+- Added `/figma-auth` command and `figma_configure_auth` tool for masked local token capture.
+- Figma tools now prompt for a fresh token and retry once when auth is missing, invalid, or expired.
+- Token setup writes to `~/.pi/agent/auth.json` without returning the token to the model.
+
+## 0.1.0
+
+### Minor Changes
+
+### New Extension: figma
+
+- Added native Figma REST API tools for file, node, styles, variables, components, component sets, component search, design-context, URL parsing, metadata, and node rendering workflows.
+- Added bundled `figma` skill that explains when and how to use the native `figma_*` tools.
+- Added authentication support via `FIGMA_TOKEN` and `~/.pi/agent/auth.json` at `.figma.token`.
+- Added README documentation for token creation, required Figma file-content read scope, and setup/troubleshooting.