Titlewise

titlewise

Generate rich, well-formatted titles (plus a description and tags) for the CURRENT conversation, so it is easy to find later by date and topic. It is interactive: it asks the user how they want the title, then produces matching variants and marks one recommended pick. Suggestion only - it proposes titles and never changes the session or its stored title.

When to use

Trigger when the user asks to name, title, or label the current conversation - for example "baslik oner", "konusmaya isim ver", "sohbeti adlandir", "label this chat", "name this conversation", or the command /titlewise.

Why suggestion-only

The host app owns the conversation title (for example, Claude Code stores and regenerates it automatically). A skill cannot reliably set it, and editing the live transcript is fragile. So this skill only PROPOSES titles: you copy the one you like and set it as the conversation title yourself, using your client's rename/title option. The skill never edits the session, the transcript, or any file.

Workflow

Follow these steps in order when invoked.

1. Read the conversation

Read the current conversation from your context: main topic/project, tools, servers, people, databases, files, and what was actually done and decided. Use only what appears in the conversation - never invent details. If the conversation is empty, too short, or has no clear subject to title, say so and ask the user for a one-line hint instead of guessing.

2. Collect preferences (Q&A)

If the user did NOT already state preferences in their invocation, ask them in ONE AskUserQuestion call with these four dimensions. All four are multi-select.

AskUserQuestion accepts at most 4 questions, and at most 4 options per question. Each dimension below therefore offers exactly 4 options or fewer. Do not add a fifth option to any of them - the tool call will be rejected. The six catalogued styles stay reachable because Compact expands into three variants.

• Language (multi-select): User language (the language of the conversation - auto-detect) / English (EN) / Chinese (CHN)
• Length (multi-select): Long / Medium / Short
• Date (multi-select): Prefix (YYYY-MM-DD - ...) / Suffix (... - YYYY-MM-DD) / None / Month (YYYY-MM)
• Style (multi-select):
  • Descriptive (+ joined)
  • Compact - produces all three compact variants: Slug (kebab-case), Keyword (middot ·), Bracket ([date][tool])
  • Outcome-focused
  • Emoji

If the user already specified preferences in the invocation (e.g. /titlewise short english slug), skip those questions and go straight to generation. A named compact style in the invocation (slug, keyword, or bracket) selects only that one variant, not the whole Compact group. If the user gives no answer or dismisses, default to: user language + all three lengths + date prefix + descriptive.

3. Generate the title variant(s)

Produce the variant(s) that match the chosen dimensions, drawing from the catalog below. If multiple options were picked in a dimension, produce one variant per combination. Rules:

• Each variant is a single, copy-ready line.
• Fill the <...> placeholders from the conversation.
• Use today's date (from your context) as an absolute YYYY-MM-DD.
• Produce each variant in each selected language. Match the target language's proper characters/diacritics; when the language is Turkish, use ç, ğ, ı, ö, ş, ü.
• Never use an em dash - use a normal hyphen.

Cap the output at 12 variants per language. Every dimension is multi-select, so a full cross-product can reach dozens of lines and stops being useful - the user has to scan them all to pick one. If the combinations exceed 12, keep the recommended pick plus the most distinct variants (one per style, then vary length, then vary date position), and end the list with a single line stating how many combinations were suppressed and which dimension to narrow to see them.

Then pick ONE recommended variant: the best default among those produced for scanning a history list - prefer dated + medium length + descriptive in the user's language when those were selected, and keep it short enough to read at a glance (about 70 characters or fewer).

Catalog (vocabulary)

A) Length x date

• Long + dated: YYYY-MM-DD - <Main Topic> (<tool/project>): <sub1> + <sub2> + ... (<people/note>)
• Long + undated: same, without the date prefix
• Medium + dated: YYYY-MM-DD - <Main Topic> (<tool>): <2-3 key subtopics>
• Medium + undated: <Main Topic> (<tool>): <2-3 key subtopics>
• Short + dated: YYYY-MM-DD - <3-6 word main topic>
• Short + undated: <3-6 word main topic>

B) Compact / search-friendly - the Compact style option produces the first three of these

• Slug (kebab): <tool>-<topic>-<topic>-YYYY-MM-DD (filename-safe: only lowercase letters, digits, hyphens)
• Keyword (middot): <tool> · <topic1> · <topic2> · <topic3>
• Bracket tag: [YYYY-MM-DD][<tool>] <topic1> + <topic2>
• Date suffix: <Main Topic> (<tool>) - YYYY-MM-DD (from the Date dimension, not the Style dimension)
• Month: YYYY-MM - <tool> <main topic> (from the Date dimension)

C) Style

• Outcome-focused: YYYY-MM-DD - <tool> <outcome>: <main-work1> + <main-work2>
• Emoji + short: <emoji> <tool> <main topic> (YYYY-MM-DD) - pick the emoji from the dominant work type: feature ✨ · new/launch 🎉 · bugfix 🐛 · fix/patch 🔧 · refactor ♻️ · docs 📝 · perf ⚡ · security 🔒 · release/ship 🚀 · config ⚙️ · tests ✅ · data/db 🗄️ · infra 🏗️ · design/ui 🎨 · research 🔎

For more options, cross-product the dimensions: language x length x date-position x style, up to the 12-variant cap. The Slug, Keyword, Bracket, and Month styles carry their own built-in date handling (or none); the Date dimension governs the descriptive (length) and outcome-focused variants.

4. Description

Write a 3-6 sentence summary paragraph (what was done, the outcome), in the selected language.

5. Tags

Give 5-8 comma-separated lowercase kebab tags: date + entities (tool, server/IP, people, database) + topics.

6. Present the results

Lead with the recommended title in a fenced code block (so it is one-click copy-ready). Then show the full title variants as a clearly labeled, copy-ready list grouped by category, followed by the description and the tags. End with a one-line note: copy the variant you like and set it as the conversation title using your client's rename/title option. Do not run or emit any command to apply the title, and do not edit any file.

Generic example (illustrative only)

For a session that fixed a payment timeout bug:

• Recommended: 2025-03-14 - Payment API Debug (payments-service): timeout fix + retries
• Long + dated: 2025-03-14 - Payment API Debug (payments-service): timeout fix + retry logic + integration tests
• Short + dated: 2025-03-14 - payment timeout fix
• Slug: payments-service-timeout-fix-2025-03-14
• Keyword: payments-service · timeout · retry · integration-tests
• Bracket: [2025-03-14][payments-service] timeout fix + retry logic
• Emoji: 🐛 payments timeout fix (2025-03-14)
• Tags: 2025-03-14, payments-service, timeout, retry, backoff, integration-tests, bugfix

Rules

• The Q&A is one AskUserQuestion call: at most 4 questions, at most 4 options each. Never add a fifth option to a dimension - reach new variants by expanding an existing option (as Compact does), not by adding one.
• Cap the produced variants at 12 per language; state how many were suppressed rather than printing the full cross-product.
• Never use an em dash; use a normal hyphen.
• Match the output language's proper characters; when Turkish, use ç, ğ, ı, ö, ş, ü.
• Each title variant is one line, copy-ready, with no trailing punctuation.
• The slug variant is filename-safe: only lowercase letters, digits, and hyphens.
• Never invent details - use only what the conversation contains; if there is nothing to title, ask for a hint.
• Suggestion only: never edit the transcript or the app's title store, and never emit a command to apply the title.