pi-bar 0.3.27 → 0.3.29

package/README.md

@@ -9,9 +9,9 @@
 claude-opus-4.7  ❯  think:med  ❯  2.6% / 1.0M  ❯  Inspecting package structure  ❯  plan:active
 ```
 
-![pi-bar with low context usage](https://cdn.jsdelivr.net/npm/pi-bar@0.3.27/assets/screenshot-green.png)
-![pi-bar with medium context usage](https://cdn.jsdelivr.net/npm/pi-bar@0.3.27/assets/screenshot-yellow.png)
-![pi-bar with high context usage](https://cdn.jsdelivr.net/npm/pi-bar@0.3.27/assets/screenshot-red.png)
+![pi-bar with low context usage](https://cdn.jsdelivr.net/npm/pi-bar@0.3.29/assets/screenshot-green.png)
+![pi-bar with medium context usage](https://cdn.jsdelivr.net/npm/pi-bar@0.3.29/assets/screenshot-yellow.png)
+![pi-bar with high context usage](https://cdn.jsdelivr.net/npm/pi-bar@0.3.29/assets/screenshot-red.png)
 
 ## Why use it?
 

package/extensions/status-footer.ts

@@ -1008,21 +1008,139 @@ class FooterTldrEngine {
 	}
 }
 
+// First-word allow/ban lists are duplicated between the prompt and the
+// sanitizer fallback. They live here so both stay aligned automatically when
+// new verbs get added or moved.
+const ALLOWED_FIRST_WORDS_PROGRESSIVE = [
+	"Reviewing",
+	"Investigating",
+	"Exploring",
+	"Updating",
+	"Refining",
+	"Fixing",
+	"Implementing",
+	"Wrapping up",
+	"Bumping",
+	"Releasing",
+	"Preparing",
+	"Drafting",
+	"Resuming",
+	"Pulling",
+	"Surveying",
+	"Recording",
+] as const;
+const ALLOWED_FIRST_WORDS_PAST = [
+	"Reviewed",
+	"Investigated",
+	"Explored",
+	"Updated",
+	"Refined",
+	"Fixed",
+	"Implemented",
+	"Wrapped up",
+	"Bumped",
+	"Released",
+	"Prepared",
+	"Drafted",
+	"Resumed",
+	"Pulled",
+	"Surveyed",
+	"Recorded",
+] as const;
+const BANNED_FIRST_WORDS = [
+	"Read",
+	"Reading",
+	"Grep",
+	"Grepping",
+	"Listing",
+	"List",
+	"Counting",
+	"Counted",
+	"Extracting",
+	"Extracted",
+	"Displaying",
+	"Displayed",
+	"Editing",
+	"Edited",
+	"Writing",
+	"Wrote",
+	"Running",
+	"Ran",
+	"Publishing",
+	"Published",
+	"Capturing",
+	"Captured",
+	"Verifying",
+	"Verified",
+	"Verify",
+	"Validating",
+	"Validated",
+	"Validate",
+	"Checking",
+	"Checked",
+	"Check",
+	"Confirming",
+	"Confirmed",
+	"Confirm",
+	"Searching",
+	"Searched",
+	"Search",
+	"Finding",
+	"Found",
+	"Find",
+] as const;
+
 function checkpointSystemPrompt(job: TldrCheckpointJob): string {
-	const tenseInstruction =
-		job.displayPriority === "final"
-			? "Start with a past-tense verb."
-			: "Start with a present-tense -ing verb.";
+	// Tense + exemplars vary by priority. Final TLDRs describe completed turns;
+	// using -ing examples for final caused all final TLDRs to slip back into
+	// present-progressive even though the literal instruction said past tense.
+	// Immediate TLDRs cover a brand-new user request; they need a rephrase
+	// directive instead of generic filler like "Continuing with task".
+	let tenseInstruction: string;
+	let goodExamples: string;
+	let allowedFirstWords: readonly string[];
+	if (job.displayPriority === "final") {
+		tenseInstruction = "Start with a past-tense verb describing what was completed.";
+		goodExamples = [
+			"- Updated footer summary behavior",
+			"- Investigated live TLDR regressions",
+			"- Refined sanitizer for stray prefixes",
+			"- Wrapped up extension release",
+		].join("\n");
+		allowedFirstWords = ALLOWED_FIRST_WORDS_PAST;
+	} else if (job.displayPriority === "immediate") {
+		tenseInstruction =
+			"Rephrase the user's new request as a concise present-progressive task clause. If the request is opaque (e.g. \"continue\", \"go\", \"ok\"), name the carry-over task with a noun (e.g. \"Resuming refactor work\"), not generic filler.";
+		goodExamples = [
+			"- Reviewing footer summary behavior",
+			"- Investigating live TLDR regressions",
+			"- Refining sanitizer for stray prefixes",
+			"- Preparing extension release",
+			"- Resuming refactor work",
+		].join("\n");
+		allowedFirstWords = ALLOWED_FIRST_WORDS_PROGRESSIVE;
+	} else {
+		tenseInstruction = "Start with a present-tense -ing verb describing current work.";
+		goodExamples = [
+			"- Reviewing footer summary behavior",
+			"- Investigating live TLDR regressions",
+			"- Refining sanitizer for stray prefixes",
+			"- Wrapping up extension release",
+		].join("\n");
+		allowedFirstWords = ALLOWED_FIRST_WORDS_PROGRESSIVE;
+	}
+
 	// Frame the model as a human developer describing visible work, not an agent
 	// summarizing its own mechanics. The banned phrases and few-shot examples
 	// fix concrete regressions observed in backtests: tool-name verbs, success
-	// suffixes, file path leaks, and markdown formatting.
+	// suffixes, file path leaks, and markdown formatting. The HARD CONSTRAINTS
+	// block sits at the tail of the prompt because instruction-following models
+	// give the most weight to the most-recent directives (recency bias).
 	return `Write one plain-English TLDR for a Pi coding agent.
 Describe the work progress as if a human developer were doing it.
 Focus on the task activity and current outcome, not agent mechanics.
 Do not mention tools, tool calls, prompts, messages, model output, or implementation details.
-Do not start with tool-name verbs such as Read, Reading, Grep, Listing, Counting, Extracting, Displaying, Editing, Writing, Running, Publishing.
-Use human-developer verbs instead: Reviewing, Investigating, Exploring, Updating, Refining, Fixing, Implementing, Wrapping up.
+Use human-developer verbs instead of tool-narration verbs.
 Do not use file paths, file extensions, code identifiers, package names, or version strings.
 Do not use backticks, asterisks, underscores, quotes, or any markdown formatting.
 Do not append filler suffixes such as "with success", "successfully", or "completed successfully".
@@ -1039,18 +1157,20 @@ Output only the status fragment itself. No prefixes, labels, bullets, or quotes.
 Plain text only; no markdown, JSON, code, file paths, or tool names.
 
 Good examples:
-- Reviewing footer summary behavior
-- Investigating live TLDR regressions
-- Refining sanitizer for stray prefixes
-- Wrapping up extension release
+${goodExamples}
 
 Bad examples:
 - Editing extensions/status-footer.ts with success.
 - Reading status-footer file completed successfully.
 - Publishing \`pi-bar@0.3.3\` to npm.
 - Grepping for sanitizeTldrText callers.
+- Verifying repository status after commit.
+- Investigating user input responses.
 
-${tenseInstruction}`;
+HARD CONSTRAINTS (apply last; override anything above that conflicts):
+- First word MUST be one of: ${allowedFirstWords.join(", ")}.
+- First word MUST NOT be: ${BANNED_FIRST_WORDS.join(", ")}.
+- ${tenseInstruction}`;
 }
 
 function extractStopReason(message: unknown): string | undefined {
@@ -1236,6 +1356,95 @@ function stripIdentifierLeaks(text: string): string {
 		.trim();
 }
 
+// After identifier removal the sentence often has a dangling preposition like
+// "Bumping version to for publishing" (was "to 0.3.2 for publishing") or a
+// trailing "to"/"version". Run in a fix-point loop so chained leftovers (e.g.
+// "to version") collapse cleanly.
+const DANGLING_TRAILING_PREP_PATTERN =
+	/\s+(?:to|at|as|of|by|for|in|on|with|from|version|v)\s*[.!?,;:]?\s*$/i;
+const DANGLING_PREP_CHAIN_PATTERN =
+	/\b(to|at|as|of|by|from|version|v)\s+(for|in|on|with|from|after|before|during|to|at|as|of|by|and|but|or)\b/gi;
+// File-path / identifier strip can leave a verb directly followed by a
+// preposition: "Reviewing README for image update" -> after path strip ->
+// "Reviewing for image update". Collapse the bare verb+prep into just the verb.
+const VERB_BARE_PREP_PATTERN =
+	/\b(Reviewing|Investigating|Updating|Refining|Exploring|Fixing|Implementing|Bumping|Releasing|Preparing|Drafting|Resuming|Pulling|Surveying|Recording)\s+(?:for|in|on|with|after|before|to|at|as|of|by|from)\s+/gi;
+
+function stripDanglingPrepositions(text: string): string {
+	let cleaned = text;
+	let previous: string;
+	do {
+		previous = cleaned;
+		cleaned = cleaned.replace(VERB_BARE_PREP_PATTERN, "$1 ");
+		cleaned = cleaned.replace(DANGLING_PREP_CHAIN_PATTERN, "$2");
+		cleaned = cleaned.replace(DANGLING_TRAILING_PREP_PATTERN, "").trim();
+	} while (cleaned !== previous && cleaned.length > 0);
+	return cleaned;
+}
+
+// Last-resort fallback when the model ignored the allow-list and started the
+// TLDR with a banned tool-narration verb. The map preserves tense (present
+// vs past). Case of the first letter follows the original word.
+const BANNED_FIRST_WORD_REWRITES: Record<string, string> = {
+	Reading: "Reviewing",
+	Read: "Reviewed",
+	Grepping: "Investigating",
+	Grep: "Investigated",
+	Listing: "Reviewing",
+	List: "Reviewed",
+	Counting: "Surveying",
+	Counted: "Surveyed",
+	Extracting: "Pulling",
+	Extracted: "Pulled",
+	Displaying: "Reviewing",
+	Displayed: "Reviewed",
+	Editing: "Updating",
+	Edited: "Updated",
+	Writing: "Drafting",
+	Wrote: "Drafted",
+	Running: "Working on",
+	Ran: "Worked on",
+	Publishing: "Releasing",
+	Published: "Released",
+	Capturing: "Recording",
+	Captured: "Recorded",
+	Verifying: "Reviewing",
+	Verified: "Reviewed",
+	Verify: "Review",
+	Validating: "Reviewing",
+	Validated: "Reviewed",
+	Validate: "Review",
+	Checking: "Reviewing",
+	Checked: "Reviewed",
+	Check: "Review",
+	Confirming: "Reviewing",
+	Confirmed: "Reviewed",
+	Confirm: "Review",
+	Searching: "Investigating",
+	Searched: "Investigated",
+	Search: "Investigate",
+	Finding: "Investigating",
+	Found: "Investigated",
+	Find: "Investigate",
+};
+const BANNED_FIRST_WORD_PATTERN = new RegExp(
+	`^(${Object.keys(BANNED_FIRST_WORD_REWRITES).join("|")})\\b`,
+);
+
+function rewriteBannedFirstWord(text: string): string {
+	const match = BANNED_FIRST_WORD_PATTERN.exec(text);
+	if (!match) return text;
+	const original = match[1];
+	const replacement = BANNED_FIRST_WORD_REWRITES[original];
+	if (!replacement) return text;
+	// Preserve lowercase-start TLDRs (rare but possible) by matching original case.
+	const cased =
+		original[0] === original[0].toLowerCase()
+			? replacement[0].toLowerCase() + replacement.slice(1)
+			: replacement;
+	return cased + text.slice(original.length);
+}
+
 // Trailing filler suffixes observed in backtests:
 //   "... with success.", "... completed successfully.", "... successfully."
 const SUCCESS_SUFFIX_PATTERN =
@@ -1265,8 +1474,10 @@ function sanitizeTldrText(
 	const withoutMarkdown = stripMarkdownFormatting(stripped);
 	const withoutScaffolding = stripLeakedScaffolding(withoutMarkdown) || withoutMarkdown;
 	const withoutLeaks = stripIdentifierLeaks(withoutScaffolding) || withoutScaffolding;
-	const withoutSuccess = stripSuccessSuffix(withoutLeaks) || withoutLeaks;
-	return truncateText(withoutSuccess, maxChars);
+	const withoutDangling = stripDanglingPrepositions(withoutLeaks) || withoutLeaks;
+	const withoutSuccess = stripSuccessSuffix(withoutDangling) || withoutDangling;
+	const rewritten = rewriteBannedFirstWord(withoutSuccess);
+	return truncateText(rewritten, maxChars);
 }
 
 function shouldShowStatus(key: string, filter: StatusFilter): boolean {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "pi-bar",
-  "version": "0.3.27",
+  "version": "0.3.29",
   "description": "Pi footer/statusline showing active model, thinking level, context pressure, live TLDR, and extension statuses.",
   "type": "module",
   "keywords": [
@@ -51,6 +51,6 @@
     "extensions": [
       "./extensions/status-footer.ts"
     ],
-    "image": "https://cdn.jsdelivr.net/npm/pi-bar@0.3.27/assets/screenshot-yellow.png"
+    "image": "https://cdn.jsdelivr.net/npm/pi-bar@0.3.29/assets/screenshot-yellow.png"
   }
 }