@oh-my-pi/pi-coding-agent 14.0.4 → 14.1.0

package/CHANGELOG.md

@@ -2,6 +2,89 @@
 
 ## [Unreleased]
 
+## [14.1.0] - 2026-04-11
+### Added
+
+- Added richer tool rendering details in session export HTML, including metadata badges, argument formatting, and todo task tree styling for exported tool and workflow messages
+- Added a persistent `js` tool backed by `node:vm`, with cross-session `highway` KV/pubsub, tool calls from inside JS cells, and `$` / `$$` interactive JavaScript execution
+- Added SQLite database read support to the `read` tool for `.sqlite`, `.sqlite3`, `.db`, and `.db3` files with table listing, schema + sample output, row lookup, paginated query filtering, and read-only `q=SELECT` mode
+- Added SQLite mutation support to the `write` tool so `db.sqlite:table` inserts JSON5 rows and `db.sqlite:table:key` updates or deletes rows via row key
+- Added rendering of usage report entries for accounts with no usage limits, including account label and optional plan type with a `-- no limits` indicator
+- Updated account label resolution to fall back to email or accountId so unlabeled unlimited-plan accounts display a meaningful name
+- Added canonical model equivalence and provider coalescing across `models.yml`, `enabledModels`, `--models`, `/model`, and `--list-models`
+- Added `equivalence` overrides/exclusions to `models.yml` and `modelProviderOrder` to `config.yml` for global canonical-provider preference
+
+### Changed
+
+- Enabled `await` and `cancel_job` to be available when `bash.autoBackground.enabled` is set, so auto-backgrounded bash jobs can be awaited or cancelled without enabling `async.enabled`
+- Updated bash auto-background behavior so short commands returned inline output when they completed before the configured threshold, while longer runs moved to background jobs automatically
+- Replaced the LLM-callable Python execution path with JavaScript execution in the shared VM context, including updated renderers, prompts, session messages, and extension events
+- Updated interactive and CLI model listings/selectors to work with canonical model ids while resolving them to concrete provider variants for actual execution
+- Updated role assignment persistence so selected model settings now store the selector used by users, including thinking-level suffixes, while runtime continues to run against the resolved concrete provider model
+- Updated model scope resolution to expand exact canonical model ids into all matching provider variants when filtering supported model sets
+- Changed the agent to avoid giving time estimates or task-duration predictions in user responses, focusing on required work instead
+- Changed generated code guidance to avoid speculative abstractions and extra compatibility scaffolding, favoring direct implementations that match current needs
+- Changed model role resolution so roles can store either canonical model ids or explicit `provider/model` selectors while sessions continue to record the concrete model actually used
+- Updated bash execution to optionally auto-background long-running commands through the existing background-job pipeline, with dedicated settings for enabling the behavior and adjusting the delay
+
+### Fixed
+
+- Fixed session export rendering so JavaScript execution messages now use `jsExecution` labels and content instead of `pythonExecution`, matching current tool behavior
+- Fixed JavaScript cell execution to auto-display returned values once and preserve persistent VM bindings across calls until reset
+- Fixed `.db`/`.db3` reads to verify SQLite file headers and fall back to normal file reading when the extension matches but the content is not a SQLite database
+- Fixed SQLite selector parsing and resolution to correctly route requests to database operations at the file-extension boundary instead of misrouting through plain file/archive handlers
+- Fixed unsupported or unsafe selectors by rejecting missing tables, composite primary keys for row lookups, unknown query parameters, and row operations on non-existent tables
+- Fixed model resolution for commit message generation, title generation, memory consolidation, and image inspection when role strings use canonical ids instead of raw provider/model values
+- Fixed default-model updates so previously configured thinking levels were preserved when reassigning a role
+- Fixed model scope and selection handling in CLI/session startup paths that previously failed to resolve aliases consistently across features
+- Fixed short-lived git subprocesses to disable `core.fsmonitor` and `core.untrackedCache`, avoiding unnecessary repository watchers and cache work during agent git operations
+
+### Security
+
+- Blocked destructive SQL execution in read-mode SQLite access by using read-only connections and rejecting bound-parameter raw SQL
+
+## [14.0.5] - 2026-04-11
+### Added
+
+- Added `designer` model role for UI/UX design tasks with Gemini 3.1 Pro as default model
+- Added support for model role fallback lists — roles can now resolve to multiple model patterns with automatic fallback to next available model
+- Added `extractReadableFromHtml` utility function to extract readable content from HTML with Readability article extraction and CSS selector fallback
+- Added support for GFM (GitHub Flavored Markdown) features including tables, strikethrough, and task lists in HTML-to-markdown conversion
+- Added `resolveDiagnosticTargets` utility function to handle glob pattern resolution with fallback to literal file paths for bracket-style paths
+
+### Changed
+
+- Clarified fenced code block editing behavior in markdown — the tool now preserves literal indentation inside fenced blocks, with content written verbatim as supplied
+- Updated guidance for inserting content after markdown section headings to use `after` on the heading chunk rather than `before`/`prepend` on the section itself
+- Reduced default image resize limits to 1568px (from 2000px) and 500KB (from 4.5MB) to match Anthropic's internal downscaling threshold and reduce payload sizes in tool calls
+- Adjusted screenshot compression to use 1024px max dimensions and 150KB budget for more aggressive optimization of browser screenshots in LLM requests
+- Updated JPEG quality defaults from 80 to 75 and refined quality ladder steps (70, 60, 50, 40) for tighter byte budgets
+- Improved image resize fast-path to skip re-encoding when images are already within dimensions and at ≤25% of byte budget, avoiding unnecessary processing of small icons and diagrams
+- Clarified that chunk names are truncated and must be copied from `read` or `?` output rather than constructed from source identifiers
+- Enhanced guidance for editing fenced code blocks in markdown to preserve exact whitespace using `raw` reads, as the tool normalizes tabs to spaces which can damage indentation-sensitive content
+- Updated designer agent to use `pi/designer` role alias instead of explicit model list
+- Refactored model role resolution to support multiple fallback patterns per role, improving model availability handling
+- Replaced regex-based HTML-to-markdown conversion with Turndown library and GFM plugin for more accurate formatting of complex HTML structures
+- Simplified no-changes response to omit redundant response text when chunk content already matches
+- Clarified region suffix behavior on leaf and compound statement chunks — `~` and `^` now fall back to whole-chunk replacement with explicit guidance to supply complete structural content
+- Updated CRC refresh guidance to direct users to use CRCs from edit responses or run `read(path="file", sel="?")`
+- Added clarification that region suffixes fall back to whole-chunk replacement for prose and data formats (markdown, YAML, JSON, fenced code blocks, frontmatter)
+- Documented `L20` shorthand syntax for single-line reads extending to end-of-file, with `L20-L20` for one-line windows
+- Refactored diagnostic target resolution to use new `resolveDiagnosticTargets` function, consolidating glob pattern detection and file matching logic
+- Updated chunk selector syntax from `@region` format to `~` (body) and `^` (head) suffixes for more concise region targeting
+- Simplified chunk edit documentation to use new `~` and `^` region syntax instead of `@head`, `@body`, `@tail`, `@decl` keywords
+- Replaced internal `raceAbort` function with imported `raceWithAbort` utility from pi-utils
+- Refactored cleanup timer to use async iterator pattern with `timers.setInterval` instead of `setInterval`
+- Made `#cleanupIdleSessions` synchronous and moved async cleanup loop logic to new `#runCleanupLoop` method
+- Replaced regex-based `htmlToBasicMarkdown` with a Turndown + GFM plugin pipeline (tables, strikethrough, task lists, nested lists now convert correctly). Added direct `turndown` and `turndown-plugin-gfm` dependencies
+
+### Fixed
+
+- Fixed chunk edit tool to report file-not-found error distinctly when attempting to use chunk selectors on non-existent files, with guidance to use write tool or verify the path
+- Fixed stale child selector reuse to correctly match chunks by checksum when multiple sibling chunks with the same name exist under the same parent
+- Fixed stale diagnostics being reused after unrelated file publishes by clearing cached diagnostics before refreshing file state
+- Fixed Codex search to use streamed answer text when final answer is an image placeholder or empty
+
 ## [14.0.4] - 2026-04-10
 ### Added
 

package/package.json

@@ -1,7 +1,7 @@
 {
 	"type": "module",
 	"name": "@oh-my-pi/pi-coding-agent",
-	"version": "14.0.4",
+	"version": "14.1.0",
 	"description": "Coding agent CLI with read, bash, edit, write tools and session management",
 	"homepage": "https://github.com/can1357/oh-my-pi",
 	"author": "Can Boluk",
@@ -46,12 +46,12 @@
 	"dependencies": {
 		"@agentclientprotocol/sdk": "0.16.1",
 		"@mozilla/readability": "^0.6",
-		"@oh-my-pi/omp-stats": "14.0.4",
-		"@oh-my-pi/pi-agent-core": "14.0.4",
-		"@oh-my-pi/pi-ai": "14.0.4",
-		"@oh-my-pi/pi-natives": "14.0.4",
-		"@oh-my-pi/pi-tui": "14.0.4",
-		"@oh-my-pi/pi-utils": "14.0.4",
+		"@oh-my-pi/omp-stats": "14.1.0",
+		"@oh-my-pi/pi-agent-core": "14.1.0",
+		"@oh-my-pi/pi-ai": "14.1.0",
+		"@oh-my-pi/pi-natives": "14.1.0",
+		"@oh-my-pi/pi-tui": "14.1.0",
+		"@oh-my-pi/pi-utils": "14.1.0",
 		"@sinclair/typebox": "^0.34",
 		"@xterm/headless": "^6.0",
 		"ajv": "^8.18",
@@ -63,10 +63,13 @@
 		"lru-cache": "11.3.1",
 		"markit-ai": "0.5.0",
 		"puppeteer": "^24.37",
+		"turndown": "7.2.4",
+		"turndown-plugin-gfm": "1.0.2",
 		"zod": "4.3.6"
 	},
 	"devDependencies": {
-		"@types/bun": "^1.3"
+		"@types/bun": "^1.3",
+		"@types/turndown": "5.0.6"
 	},
 	"engines": {
 		"bun": ">=1.3.7"

package/src/async/index.ts

@@ -1 +1,2 @@
 export * from "./job-manager";
+export * from "./support";

package/src/async/support.ts

@@ -0,0 +1,5 @@
+import type { Settings } from "../config/settings";
+
+export function isBackgroundJobSupportEnabled(settings: Pick<Settings, "get">): boolean {
+	return settings.get("async.enabled") || settings.get("bash.autoBackground.enabled");
+}

package/src/cli/list-models.ts

@@ -6,6 +6,45 @@ import { formatNumber } from "@oh-my-pi/pi-utils";
 import type { ModelRegistry } from "../config/model-registry";
 import { fuzzyFilter } from "../utils/fuzzy";
 
+interface ProviderRow {
+	provider: string;
+	model: string;
+	context: string;
+	maxOut: string;
+	thinking: string;
+	images: string;
+}
+
+interface CanonicalRow {
+	canonical: string;
+	selected: string;
+	variants: string;
+	context: string;
+	maxOut: string;
+}
+
+function writeLine(line = ""): void {
+	process.stdout.write(`${line}\n`);
+}
+
+function renderTable<T extends Record<string, string>>(rows: T[], headers: T): void {
+	const widths = Object.fromEntries(
+		Object.keys(headers).map(key => [key, Math.max(headers[key]!.length, ...rows.map(row => row[key]!.length))]),
+	) as Record<keyof T, number>;
+
+	const headerLine = Object.keys(headers)
+		.map(key => headers[key as keyof T]!.padEnd(widths[key as keyof T]))
+		.join("  ");
+	writeLine(headerLine);
+
+	for (const row of rows) {
+		const line = Object.keys(headers)
+			.map(key => row[key as keyof T]!.padEnd(widths[key as keyof T]))
+			.join("  ");
+		writeLine(line);
+	}
+}
+
 /**
  * List available models, optionally filtered by search pattern
  */
@@ -13,77 +52,77 @@ export async function listModels(modelRegistry: ModelRegistry, searchPattern?: s
 	const models = modelRegistry.getAvailable();
 
 	if (models.length === 0) {
-		console.log("No models available. Set API keys in environment variables.");
+		writeLine("No models available. Set API keys in environment variables.");
 		return;
 	}
 
-	// Apply fuzzy filter if search pattern provided
 	let filteredModels: Model<Api>[] = models;
 	if (searchPattern) {
-		filteredModels = fuzzyFilter(models, searchPattern, m => `${m.provider} ${m.id}`);
+		filteredModels = fuzzyFilter(models, searchPattern, model => `${model.provider} ${model.id}`);
 	}
 
-	if (filteredModels.length === 0) {
-		console.log(`No models matching "${searchPattern}"`);
+	const filteredCanonical = modelRegistry
+		.getCanonicalModels({ availableOnly: true, candidates: filteredModels })
+		.map(record => {
+			const selected = modelRegistry.resolveCanonicalModel(record.id, {
+				availableOnly: true,
+				candidates: filteredModels,
+			});
+			if (!selected) return undefined;
+			return {
+				canonical: record.id,
+				selected: `${selected.provider}/${selected.id}`,
+				variants: String(record.variants.length),
+				context: formatNumber(selected.contextWindow),
+				maxOut: formatNumber(selected.maxTokens),
+			} satisfies CanonicalRow;
+		})
+		.filter((row): row is CanonicalRow => row !== undefined)
+		.sort((left, right) => left.canonical.localeCompare(right.canonical));
+
+	if (filteredModels.length === 0 && filteredCanonical.length === 0) {
+		writeLine(`No models matching "${searchPattern}"`);
 		return;
 	}
 
-	// Sort by provider, then by model id
-	filteredModels.sort((a, b) => {
-		const providerCmp = a.provider.localeCompare(b.provider);
+	filteredModels.sort((left, right) => {
+		const providerCmp = left.provider.localeCompare(right.provider);
 		if (providerCmp !== 0) return providerCmp;
-		return a.id.localeCompare(b.id);
+		return left.id.localeCompare(right.id);
 	});
 
-	// Calculate column widths
-	const rows = filteredModels.map(m => ({
-		provider: m.provider,
-		model: m.id,
-		context: formatNumber(m.contextWindow),
-		maxOut: formatNumber(m.maxTokens),
-		thinking: m.thinking ? getSupportedEfforts(m).join(",") : m.reasoning ? "yes" : "-",
-		images: m.input.includes("image") ? "yes" : "no",
-	}));
+	const providerRows = filteredModels.map(model => ({
+		provider: model.provider,
+		model: model.id,
+		context: formatNumber(model.contextWindow),
+		maxOut: formatNumber(model.maxTokens),
+		thinking: model.thinking ? getSupportedEfforts(model).join(",") : model.reasoning ? "yes" : "-",
+		images: model.input.includes("image") ? "yes" : "no",
+	})) satisfies ProviderRow[];
 
-	const headers = {
-		provider: "provider",
-		model: "model",
-		context: "context",
-		maxOut: "max-out",
-		thinking: "thinking",
-		images: "images",
-	};
-
-	const widths = {
-		provider: Math.max(headers.provider.length, ...rows.map(r => r.provider.length)),
-		model: Math.max(headers.model.length, ...rows.map(r => r.model.length)),
-		context: Math.max(headers.context.length, ...rows.map(r => r.context.length)),
-		maxOut: Math.max(headers.maxOut.length, ...rows.map(r => r.maxOut.length)),
-		thinking: Math.max(headers.thinking.length, ...rows.map(r => r.thinking.length)),
-		images: Math.max(headers.images.length, ...rows.map(r => r.images.length)),
-	};
-
-	// Print header
-	const headerLine = [
-		headers.provider.padEnd(widths.provider),
-		headers.model.padEnd(widths.model),
-		headers.context.padEnd(widths.context),
-		headers.maxOut.padEnd(widths.maxOut),
-		headers.thinking.padEnd(widths.thinking),
-		headers.images.padEnd(widths.images),
-	].join("  ");
-	console.log(headerLine);
+	if (filteredCanonical.length > 0) {
+		writeLine("Canonical models");
+		renderTable(filteredCanonical, {
+			canonical: "canonical",
+			selected: "selected",
+			variants: "variants",
+			context: "context",
+			maxOut: "max-out",
+		});
+		if (providerRows.length > 0) {
+			writeLine();
+		}
+	}
 
-	// Print rows
-	for (const row of rows) {
-		const line = [
-			row.provider.padEnd(widths.provider),
-			row.model.padEnd(widths.model),
-			row.context.padEnd(widths.context),
-			row.maxOut.padEnd(widths.maxOut),
-			row.thinking.padEnd(widths.thinking),
-			row.images.padEnd(widths.images),
-		].join("  ");
-		console.log(line);
+	if (providerRows.length > 0) {
+		writeLine("Provider models");
+		renderTable(providerRows, {
+			provider: "provider",
+			model: "model",
+			context: "context",
+			maxOut: "max-out",
+			thinking: "thinking",
+			images: "images",
+		});
 	}
 }

package/src/commit/model-selection.ts

@@ -1,7 +1,12 @@
 import type { ThinkingLevel } from "@oh-my-pi/pi-agent-core";
 import type { Api, Model } from "@oh-my-pi/pi-ai";
 import { MODEL_ROLE_IDS } from "../config/model-registry";
-import { parseModelPattern, resolveModelRoleValue, resolveRoleSelection } from "../config/model-resolver";
+import {
+	type ModelLookupRegistry,
+	parseModelPattern,
+	resolveModelRoleValue,
+	resolveRoleSelection,
+} from "../config/model-resolver";
 import type { Settings } from "../config/settings";
 import MODEL_PRIO from "../priority.json" with { type: "json" };
 
@@ -11,19 +16,20 @@ export interface ResolvedCommitModel {
 	thinkingLevel?: ThinkingLevel;
 }
 
+type CommitModelRegistry = ModelLookupRegistry & {
+	getApiKey: (model: Model<Api>) => Promise<string | undefined>;
+};
+
 export async function resolvePrimaryModel(
 	override: string | undefined,
 	settings: Settings,
-	modelRegistry: {
-		getAvailable: () => Model<Api>[];
-		getApiKey: (model: Model<Api>) => Promise<string | undefined>;
-	},
+	modelRegistry: CommitModelRegistry,
 ): Promise<ResolvedCommitModel> {
 	const available = modelRegistry.getAvailable();
 	const matchPreferences = { usageOrder: settings.getStorage()?.getModelUsageOrder() };
 	const resolved = override
-		? resolveModelRoleValue(override, available, { settings, matchPreferences })
-		: resolveRoleSelection(["commit", "smol", ...MODEL_ROLE_IDS], settings, available);
+		? resolveModelRoleValue(override, available, { settings, matchPreferences, modelRegistry })
+		: resolveRoleSelection(["commit", "smol", ...MODEL_ROLE_IDS], settings, available, modelRegistry);
 	const model = resolved?.model;
 	if (!model) {
 		throw new Error("No model available for commit generation");
@@ -37,15 +43,12 @@ export async function resolvePrimaryModel(
 
 export async function resolveSmolModel(
 	settings: Settings,
-	modelRegistry: {
-		getAvailable: () => Model<Api>[];
-		getApiKey: (model: Model<Api>) => Promise<string | undefined>;
-	},
+	modelRegistry: CommitModelRegistry,
 	fallbackModel: Model<Api>,
 	fallbackApiKey: string,
 ): Promise<ResolvedCommitModel> {
 	const available = modelRegistry.getAvailable();
-	const resolvedSmol = resolveRoleSelection(["smol"], settings, available);
+	const resolvedSmol = resolveRoleSelection(["smol"], settings, available, modelRegistry);
 	if (resolvedSmol?.model) {
 		const apiKey = await modelRegistry.getApiKey(resolvedSmol.model);
 		if (apiKey) return { model: resolvedSmol.model, apiKey, thinkingLevel: resolvedSmol.thinkingLevel };
@@ -53,7 +56,7 @@ export async function resolveSmolModel(
 
 	const matchPreferences = { usageOrder: settings.getStorage()?.getModelUsageOrder() };
 	for (const pattern of MODEL_PRIO.smol) {
-		const candidate = parseModelPattern(pattern, available, matchPreferences).model;
+		const candidate = parseModelPattern(pattern, available, matchPreferences, { modelRegistry }).model;
 		if (!candidate) continue;
 		const apiKey = await modelRegistry.getApiKey(candidate);
 		if (apiKey) return { model: candidate, apiKey };