wiki-graph-core 0.3.1

package/data/editor/clue_reviewer_generator.jinja

@@ -0,0 +1,428 @@
+You are creating a tiered verification strategy for reviewing compressed text. Your job is to analyze the chunks in this narrative clue and generate a structured checklist with THREE verification levels based on each chunk's retention level.
+
+**CRITICAL Language Requirement:**
+
+Write your entire verification strategy in the **SAME LANGUAGE as the source text** provided below. This ensures:
+- Terminology and key terms remain consistent with the original text without translation loss
+- The reviewer can directly compare against source text without language barriers
+- Literary/technical expressions are preserved accurately
+
+**Your output (rubric, core facts, key points, all examples) must be in the source text's original language.**
+
+**Note:** The source text may contain multiple languages (e.g., an English book quoting French passages). Use the predominant language of the text for your rubric, but preserve any multilingual quotes exactly as they appear.
+
+---
+
+## Input Data
+
+The user will provide a narrative clue presented as **continuous book-like text** with chunk boundaries marked using XML-style tags.
+
+**XML Markup Format:**
+```xml
+Regular unmarked text...
+
+<chunk label="Chunk label" retention="detailed">
+Text belonging to this chunk...
+</chunk>
+
+More unmarked text...
+
+<chunk label="Another chunk" retention="focused">
+More chunk text...
+</chunk>
+```
+
+**Chunk Attributes:**
+- `label`: Semantic label (required)
+- `retention`: verbatim/detailed/focused/relevant (required)
+
+**Reading Tips:**
+- Read the text naturally like a book
+- Unmarked text provides context (surrounding narrative)
+- Marked chunks (`<chunk>`) are the focus points for verification
+- If chunks overlap (same text, different attributes), the highest-priority attributes are shown
+
+---
+
+## Simplified Three-Level System
+
+Both retention levels (for user-focused chunks) and importance levels (for book-coherence chunks) are mapped to a unified three-tier verification system:
+
+| retention | importance | Priority Level | Completely Missing | Mentioned but Incomplete |
+|-----------|-----------|---------------|-------------------|-------------------------|
+| verbatim | — | **high (TIER 1a)** | CRITICAL | MAJOR (completeness check) |
+| detailed | — | **high (TIER 1a)** | CRITICAL | MAJOR (completeness check) |
+| — | critical | **high (TIER 1b)** | CRITICAL | MAJOR |
+| focused | — | **medium** | MAJOR | MINOR |
+| — | important | **medium** | MAJOR | MINOR |
+| relevant | — | **low** | MINOR | MINOR |
+| — | helpful | **low** | MINOR | MINOR |
+
+**Merging rule**: When a chunk has both retention and importance attributes, use the **higher priority level** for verification (TIER 1a > TIER 1b > TIER 2 > TIER 3).
+
+**Important principle**: Every chunk must be able to generate at least a MINOR issue. This ensures no chunk is effectively ignored during verification.
+
+**Key differences between TIER 1a and TIER 1b:**
+- **TIER 1a (verbatim/detailed)**: Core facts + Completeness check against source text. Student can achieve 100% by copying original text.
+- **TIER 1b (critical)**: Core facts + Minor details (at least one). More flexible compression allowed.
+
+---
+
+## Your Task
+
+Analyze each marked `<chunk>` and assign it to one of THREE tiers based on its attributes:
+
+### TIER 1 (high priority): verbatim OR detailed OR critical
+
+**Two sub-types with different verification strategies:**
+
+#### TIER 1a: verbatim OR detailed chunks
+**Verification Condition:** "Core Facts + Completeness Check"
+- Must include ALL items from the **core facts list**
+- THEN compare with source text for information completeness
+
+**Issue reporting:**
+- **If completely missing:** CRITICAL issue
+- **If core facts incomplete:** CRITICAL issue
+- **If core facts satisfied but information incomplete compared to source:** MAJOR issue (guaranteed minimum)
+
+**Strategy:**
+- Extract the original chunk text from the markup (will be provided to reviewer)
+- **FIRST, detect if this chunk contains literary/metaphorical content** (see guidelines below)
+- Identify core facts (essential information, usually 2-5 items)
+- Add completeness check instruction: reviewer must compare compressed text with source text
+- The student CAN achieve 100% satisfaction by copying the original text (at the cost of compression ratio)
+
+#### TIER 1b: critical importance chunks
+**Verification Condition:** "Core + Minor"
+- Must include ALL items from the **core facts list**
+- PLUS at least ONE item from the **minor details list**
+
+**Issue reporting:**
+- **If completely missing:** CRITICAL issue
+- **If mentioned but core incomplete or no minor details:** MAJOR issue
+
+**Strategy:**
+- Extract the original chunk text from the markup
+- **FIRST, detect if this chunk contains literary/metaphorical content** (see guidelines below)
+- Identify core facts (essential information that must be present)
+- Identify minor details (supporting details, at least one required)
+- Generate precise requirements
+
+**Literary/Metaphorical Content Detection:**
+
+Before extracting core facts, analyze the chunk text for these characteristics:
+- **Metaphorical expressions**: 比喻性语言 (e.g., "掘墓人", "最后一根稻草", "凤凰涅槃")
+- **Prophetic/foreshadowing tone**: 预示性叙述 (e.g., "命运为...准备了", "历史在等待", "注定要...")
+- **Symbolic language**: 象征性表达 (e.g., "黑云压城", "星火燎原")
+- **Authorial commentary**: 作者评论 (e.g., "这标志着...", "这意味着...")
+- **Dramatic framing**: 戏剧性框架 (e.g., "浴火重生", "命运转折")
+
+If the chunk contains ANY of these, mark it as **[LITERARY]** and follow special extraction rules:
+
+**Special extraction rules for literary chunks:**
+
+1. **Preserve metaphorical expressions in core facts**:
+   - DON'T: "某人出生"
+   - DO: "命运为王朝准备了'终结者'" (preserve the metaphorical framing)
+
+2. **Capture the tone/function, not just facts**:
+   - DON'T: "某事件引发起义"
+   - DO: "某事件成为'压垮帝国的最后一根稻草'" (preserve judgmental/conclusive tone)
+
+3. **Include literary language in both core facts AND minor details**:
+   - Core facts should capture essential metaphorical meanings
+   - Minor details should include vivid metaphors, dramatic descriptions
+
+4. **Add explicit instructions in core facts**:
+   - Use quotation marks around key metaphorical terms to signal they must be preserved
+   - Example: Core fact: "使用'掘墓人'或类似metaphorical language描述人物的历史角色"
+
+**Example (TIER 1a - detailed/verbatim chunk, factual):**
+```
+**地主拒绝帮助**
+
+Source text:
+<source>
+主角的父母死后，他去找地主，请求看在父亲给他当了一辈子佃户的分上，找个地方埋葬父亲，但地主干净利落地拒绝了他。
+</source>
+
+Core facts (ALL required):
+- 主角父母死后
+- 主角去找地主求坟地
+- 主角提到父亲当了一辈子佃户
+- 地主拒绝了他
+
+Completeness check:
+- After verifying core facts, compare the compressed text with source text above
+- Check if any substantive information is missing, understated, or significantly altered
+- This is a "detailed" chunk - it should preserve ~100% information with rephrasing allowed
+- If information is incomplete → report MAJOR issue (guaranteed minimum even if core facts are satisfied)
+- The student can achieve 100% satisfaction by preserving the original text closely
+```
+
+**Example (TIER 1a - detailed/verbatim chunk, literary):**
+```
+**命运的预示** [LITERARY]
+
+Source text:
+<source>
+历史为这个王朝准备了一个终结者：某人。当然他不会想到命运会交给他这样一个重要的任务。这一年，他十七岁。很快一场灾难就要降临到他的身上，但同时，一个伟大的使命也在等待着他，只有像传说中的凤凰一样，历经苦难,投入火中，经过千锤百炼，才能浴火重生，成为光芒万丈的神鸟。来吧，命运之子正在等待着你！
+</source>
+
+Core facts (ALL required):
+- 使用metaphorical/prophetic language描述主角被命运选中(如"历史准备的终结者"、"命运之子"等)
+- 这一年主角十七岁
+- Foreshadowing双重命运：一场灾难即将降临 + 一个伟大的使命在等待
+- 使用"凤凰"/"浴火重生"等metaphorical imagery描述转变历程
+
+Completeness check:
+- After verifying core facts, compare with source text for metaphorical/prophetic language preservation
+- Check: Are metaphors like "千锤百炼"、"光芒万丈的神鸟"、"来吧，命运之子" preserved?
+- This is a literary chunk - factual substitutions are NOT sufficient
+- The prophetic tone and dramatic expressions should be largely preserved
+- If tone/metaphors are significantly weakened → report MAJOR issue
+
+Note: This is a literary/prophetic chunk requiring high preservation of style and metaphorical language.
+```
+
+**Example (TIER 1b - critical importance chunk):**
+```
+**主角决定参军**
+
+Source text:
+<source>
+面对师兄的告发和朝廷的追捕，主角算了一卦，得到"逃跑必死，起义有生机"的启示。他意识到别无选择，只能参加起义军。
+</source>
+
+Core facts (ALL required):
+- 师兄告发了主角
+- 主角算卦得到启示
+- 主角决定参加起义军
+
+Minor details (at least ONE required):
+- 卦象内容是"逃跑必死，起义有生机"
+- 主角意识到别无选择
+- 朝廷在追捕主角
+```
+
+---
+
+### TIER 2 (medium priority): focused OR important
+**Verification Condition:** "Any One"
+- Must mention AT LEAST ONE item from the **key points list**
+
+**If none mentioned:** MAJOR issue
+**If at least one mentioned but incomplete:** MINOR issue
+
+**Strategy:**
+- Extract core meaning without requiring full text
+- List key points (any one sufficient)
+- Allow semantic equivalence, paraphrasing, reordering
+
+**Example:**
+```
+**统领审问新兵**
+
+Key points (at least ONE required):
+- 统领审问主角
+- 主角表现镇定
+- 统领想看到慌乱但没看到
+- 得知是熟人介绍后相信了
+```
+
+---
+
+### TIER 3 (low priority): relevant OR helpful
+**Verification Condition:** "Existence"
+- Must have ANY form of mention (any phrasing, any detail counts)
+
+**If not mentioned at all:** MINOR issue
+**If mentioned in any form (even incomplete):** MINOR issue (but acceptable)
+
+**Strategy:**
+- Brief description of what should be present
+- Very flexible - any reference counts
+
+**Example:**
+```
+**主角的家庭背景**
+
+Existence check:
+- 文本中必须有关于主角家庭生活的任何描述（任何形式的提及都算满足）
+```
+
+---
+
+## Output Format (Natural language, will be embedded in reviewer prompt)
+
+Structure your output EXACTLY as follows:
+
+```
+## Clue: "【Clue theme name based on actual content, 3-8 characters】"
+
+---
+
+### TIER 1 (HIGH PRIORITY - CRITICAL if missing, MAJOR if incomplete)
+
+**For verbatim/detailed chunks:**
+
+**【Chunk label】** [Add [LITERARY] tag if this chunk contains metaphorical/prophetic language]
+
+Source text:
+<source>
+【Original chunk text extracted from markup】
+</source>
+
+Core facts (ALL required):
+- 【Core fact 1 - if literary chunk, include metaphorical expressions with quotes】
+- 【Core fact 2 - preserve tone/style requirements】
+- 【Core fact 3】
+- 【...】
+
+Completeness check:
+- After verifying core facts, compare the compressed text with source text above
+- Check if any substantive information is missing, understated, or significantly altered
+- This is a "detailed"/"verbatim" chunk - it should preserve ~100% information with rephrasing allowed
+- If information is incomplete → report MAJOR issue (guaranteed minimum even if core facts are satisfied)
+- The student can achieve 100% satisfaction by preserving the original text closely
+
+[If literary chunk, add a note:]
+Note: This is a literary/[metaphorical/prophetic/symbolic] chunk requiring high preservation of style and [metaphorical/prophetic/symbolic] language.
+
+---
+
+**For critical importance chunks:**
+
+**【Chunk label】** [Add [LITERARY] tag if this chunk contains metaphorical/prophetic language]
+
+Source text:
+<source>
+【Original chunk text extracted from markup】
+</source>
+
+Core facts (ALL required):
+- 【Core fact 1 - if literary chunk, include metaphorical expressions with quotes】
+- 【Core fact 2 - preserve tone/style requirements】
+
+Minor details (at least ONE required):
+- 【Minor detail 1 - include dramatic/metaphorical language if present】
+- 【Minor detail 2】
+- 【Minor detail 3】
+
+[If literary chunk, add a note:]
+Note: This is a literary/[metaphorical/prophetic/symbolic] chunk. Factual substitutions are NOT sufficient. The [metaphorical/prophetic/symbolic] tone and expressions must be preserved.
+
+---
+
+### TIER 2 (MEDIUM PRIORITY - MAJOR if missing)
+
+**【Chunk label】**
+
+Key points (at least ONE required):
+- 【Key point 1】
+- 【Key point 2】
+- 【Key point 3】
+
+---
+
+### TIER 3 (LOW PRIORITY - MINOR if missing)
+
+**【Chunk label】**
+
+Existence check:
+- 【Description of what should be present in any form】
+
+---
+
+### NARRATIVE COHERENCE (OPTIONAL - only if this clue has importance chunks)
+
+**When to include this section:**
+- ONLY if this clue contains chunks with importance attributes (critical/important/helpful)
+- Skip this section entirely if all chunks only have retention attributes
+
+**What to check:**
+Identify potential narrative gaps or logical jumps between chunks that could confuse readers:
+
+**【Potential coherence issue description】**
+- Expected flow: 【chunk A】 → 【chunk B】 → 【chunk C】
+- Potential problem: 【describe what logical connection might be missing】
+- Severity: MAJOR or MINOR
+- Suggestion: 【what bridging information would help】
+
+(Repeat for each identified coherence issue, typically 0-2 issues per clue)
+
+---
+```
+
+---
+
+## Important Guidelines:
+
+1. **Clue naming:** Name based on what the chunks actually discuss (clue's real content), not user's general intent
+
+2. **Tier assignment logic:**
+   - retention = verbatim OR detailed → TIER 1a (with completeness check)
+   - importance = critical → TIER 1b (with core + minor)
+   - retention = focused → TIER 2
+   - retention = relevant → TIER 3
+   - When a chunk has BOTH retention and importance: use the higher priority tier (1a > 1b > 2 > 3)
+
+3. **Source text extraction:**
+   - For TIER 1 (all subtypes): Always extract original text from `<chunk>` tags and place in `<source>` tags
+   - For TIER 2-3: No need to quote full text (saves space)
+
+4. **TIER 1a verification (verbatim/detailed chunks):**
+   - Core facts: Absolutely essential information (usually 2-5 items)
+   - NO minor details list - use completeness check instead
+   - Completeness check gives reviewer flexibility to judge "all information" against source text
+   - Student can achieve 100% satisfaction by closely preserving original text
+
+5. **TIER 1b verification (critical importance chunks):**
+   - Core facts: Absolutely essential information (usually 2-4 items)
+   - Minor details: Supporting details that add richness (usually 2-5 items, only need 1)
+   - If a chunk is very simple, you can have 1-2 core facts and 1-2 minor details
+
+6. **Key points quantity (TIER 2):**
+   - List 2-5 key points
+   - Reviewer only needs to find ONE to pass
+   - More options = more flexibility for compression
+
+7. **Only use information explicitly stated in the source text** - do not add facts from general knowledge
+
+8. **Generate abstract requirements, not specific phrasings** (for TIER 2):
+   - ✅ GOOD: "Multiple family members died in rapid succession"
+   - ❌ BAD: "On June 6th father died, June 9th brother died..." (too specific)
+
+9. **Context awareness:**
+   - Unmarked text between chunks provides narrative context
+   - Use surrounding text to better understand chunk meaning
+   - This complete context helps you generate more accurate verification strategies
+
+10. **Every chunk MUST appear in exactly ONE tier** - no chunk should be skipped
+
+11. **Narrative Coherence (for importance chunks only)**:
+   - Check if this clue contains chunks with importance attributes (critical/important/helpful)
+   - If yes, analyze the logical flow between these chunks in the original text
+   - Identify potential narrative gaps:
+     - Are there large jumps between chunks that skip important context?
+     - Does the story jump from topic A to topic B without explanation?
+     - Is there missing motivation for key decisions (especially for importance=critical chunks)?
+   - Common coherence issues:
+     - **Abrupt topic change**: "After discussing X, suddenly talks about Y without transition"
+     - **Missing cause**: "Character decides to do Z, but the triggering event is not mentioned"
+     - **Background info timing**: "Background policy P appears out of context without explaining why it matters now"
+   - Only flag issues that would genuinely confuse readers
+   - Limit to 0-2 coherence issues per clue (don't over-flag)
+   - Example:
+     ```
+     **缺少"元朝抓人政策"与"朱重八决定"的因果连接**
+     - Expected flow: 师兄告发 → 算卦 → **元朝抓人政策** → 朱重八意识到危险 → 决定造反
+     - Potential problem: 压缩文本在"算卦"之后突然插入"元朝官吏抓人"，读者会困惑为什么突然讲政策背景
+     - Severity: MINOR
+     - Suggestion: 在提到元朝抓人政策时，加上"此时"或"而且"等过渡词，或说明"这意味着朱重八即使不造反也可能被抓"
+     ```
+
+---
+
+**Your output will be directly used as the reviewer's instruction manual. Be clear, structured, and actionable.**

package/data/editor/revision_feedback.jinja

@@ -0,0 +1,5 @@
+I reviewed your compressed version and found the following issues:
+
+{{ issues_description }}
+
+Please regenerate a complete compressed version, ensuring all the above issues are addressed.

package/data/editor/text_compressor.jinja

@@ -0,0 +1,127 @@
+# Your Role
+
+You are a student working on a text compression assignment. Your teacher has given you a text segment to compress while preserving important information.
+
+Your teacher will review your work and give you feedback. If your compression misses important information or doesn't meet requirements, the teacher will tell you specifically what to fix. **Listen to your teacher's feedback and make the requested changes.**
+
+---
+
+## Your Assignment
+
+**Original text:** {{ original_length }} characters
+**Target length:** {{ target_length }} characters ({{ compression_ratio }}%)
+**Acceptable range:** {{ acceptable_min }} - {{ acceptable_max }} characters
+
+Your compressed text must fall within the acceptable range. Going over the maximum is not acceptable for your first submission.
+
+---
+
+{% if user_language %}
+## Language Requirements
+
+**Output language:** {{ user_language }}
+
+**CRITICAL:** You must translate your compressed output to **{{ user_language }}** while maintaining:
+- Semantic accuracy (preserve all key information)
+- Natural expression (sound native in {{ user_language }}, not machine-translated)
+- Terminology consistency (technical terms and proper nouns should be accurately rendered)
+
+**Process:**
+1. Read and understand the marked content in whatever language(s) it uses
+2. Compress according to retention levels (`verbatim`, `detailed`, etc.)
+3. Translate the compressed result to {{ user_language }}
+4. Ensure the output reads naturally in {{ user_language }}
+
+**Important Notes:**
+- The input text may be in any language or mix of languages (e.g., an English book with French quotes)
+- Your final output must be entirely in {{ user_language }}
+- Your teacher will review the {{ user_language }} output for both information completeness and translation quality
+
+{% else %}
+**Language Note:** Keep your compressed output in the **SAME LANGUAGE(S) as the input text**. Do not translate.
+
+- If the input contains multiple languages, preserve them as they appear
+- Your output should read naturally in the source language(s)
+{% endif %}
+
+---
+
+## Format Requirements (MUST FOLLOW)
+
+### 1. Remove ALL `<chunk>` tags from your output
+
+The input text contains `<chunk>` markup tags like this:
+```
+<chunk retention="detailed">Some important text here</chunk>
+```
+
+**These tags are ONLY markers to help you identify important content.** Think of them as highlighting on a page.
+
+**Your output must be plain text with NO tags whatsoever:**
+- ❌ WRONG: `<chunk retention="detailed">比如刮风啊...</chunk>`
+- ✅ CORRECT: `比如刮风啊...`
+
+### 2. Output plain text only
+
+Do NOT add:
+- Section headers (like "## Compressed Text")
+- XML/HTML markup of any kind
+- Explanatory comments mixed with the content
+
+Your output should read like a natural excerpt from a longer article.
+
+---
+
+## Compression Guidance
+
+### Identify content priority from the markup:
+
+**`<chunk retention="verbatim">...</chunk>`**: Preserve almost word-for-word
+- Keep the exact wording, only fix obvious typos or punctuation errors
+- Do NOT paraphrase or restructure
+
+**`<chunk retention="detailed">...</chunk>`**: Preserve all facts
+- Keep all substantive information (names, events, numbers, relationships)
+- You may rephrase for better flow, but don't omit facts
+
+**Text without markup tags**: Can be heavily compressed or deleted
+- This is background filler between important content
+- Delete entire paragraphs if they don't have markup
+- Only keep ultra-short transitions if absolutely necessary (e.g., "三年后", "1344年", "此时")
+
+### Your compression strategy:
+
+1. **First, preserve all marked content** (`<chunk>` tags) - this is your priority
+2. **Then, delete most unmarked text** - this is where you achieve compression
+3. **Add minimal transitions** between marked sections if needed for coherence
+
+### How to meet the length requirement:
+
+The marked content will likely already take up significant space. To hit your target length:
+- Delete most unmarked paragraphs entirely
+- Only keep 1-2 word time/place markers between marked chunks if necessary
+- Don't try to "summarize" unmarked text - just delete it
+
+If you're approaching your length limit and still have unmarked text, delete more aggressively.
+
+---
+
+## Context: Segmented Compression
+
+You are compressing a **segment** of a larger work (like part of a book chapter). Your output will be concatenated with other compressed segments.
+
+**Therefore:**
+- Write as continuous prose, not a standalone summary
+- Match the original's tone and structure
+- Don't add concluding remarks or thematic wrap-ups
+- End naturally where the segment ends
+
+---
+
+## Output Format
+
+[Optional: If you face difficult trade-offs, write 1-2 sentences about your approach. Keep this brief.]
+
+---
+
+[Your compressed text - plain text only, no headers, no tags, written as continuous prose]

package/data/guaranteed/response_intent_classifier.jinja

@@ -0,0 +1,8 @@
+Classify the raw assistant reply into exactly one label.
+
+Labels:
+- malformed_json: the reply is trying to provide JSON, but it is broken or truncated
+- natural_language: the reply is plain conversational language, apology, refusal, or explanation instead of JSON
+- ambiguous: unclear or mixed
+
+Reply with the label only.

package/data/help/commands/archive/next.jinja

@@ -0,0 +1,29 @@
+Help Type: command
+Command: {{ commandName }} next
+
+Purpose:
+  Continue any paged Wiki Graph result from its `nextCursor`.
+
+Use when:
+  - A JSON or text result prints `nextCursor` or `Next page: {{ commandName }} next <cursor>`.
+  - You do not want to know whether the cursor came from `--query`, scope enumeration, or evidence.
+
+Usage:
+  {{ commandName }} next <cursor> [--limit <n>] [--json|--jsonl] [--help|-h]
+  {{ commandName }} next <archive-uri> <cursor> [--limit <n>] [--json|--jsonl] [--help|-h]
+
+Returns:
+  - The next result page for the cursor.
+
+Notes:
+  - Cursors are opaque short tokens stored in local Wiki Graph state.
+  - Passing `<archive-uri>` is optional; when provided, it must match the archive stored in the cursor.
+  - Use `--limit` to override the next page size.
+
+Examples:
+  {{ commandName }} next c_7Kp9x2aQ
+  {{ commandName }} next wikg:///books/book.wikg c_7Kp9x2aQ
+
+Related help:
+  - URI scopes, pagination, JSON, and JSONL result shapes:
+    {{ commandName }} help uri

package/data/help/commands/gc.jinja

@@ -0,0 +1,38 @@
+Help Type: command
+Command: {{ commandName }} gc
+
+Purpose:
+  Try to run Wiki Graph local garbage collection.
+
+Usage:
+  {{ commandName }} gc [--force] [--dry-run] [--json]
+
+Behavior:
+  GC scans local Wiki Graph state and cache resources, then releases expired or low-value resources that are not in use.
+  If another GC run is active, this command reports that it was skipped instead of waiting or starting a second run.
+
+Options:
+  --force
+    Ignore normal cache lifetimes and remove disposable resources as soon as they are not in use.
+    This can remove warm SQLite caches and completed or failed build jobs that a normal GC run would keep temporarily.
+    Wikipedia page cache entries still follow their own TTL and are not removed early by `--force`.
+  --dry-run
+    Report what would be removed without deleting anything.
+
+Managed areas:
+  - Wiki Graph coordinator workspace files and cached SQLite databases.
+  - Search query cache.
+  - Expired Wikipedia page cache.
+  - Completed build job records and their durable event, cache, log, and workspace files.
+  - Wiki Graph controlled temporary directories under the state directory.
+
+Output:
+  Text output reports scanned resources, removed resources, and freed disk space.
+  With `--dry-run`, text output reports candidate removals and candidate freed disk space.
+  Use `--json` for a machine-readable run report.
+
+Related help:
+  - Runtime cache, logs, workspaces, and local state:
+    {{ commandName }} help runtime
+  - Search index and generated-data readiness after cleanup:
+    {{ commandName }} help readiness

package/data/help/commands/legacy/migrate.jinja

@@ -0,0 +1,20 @@
+Help Type: command
+Command: {{ commandName }} legacy migrate
+
+Purpose:
+  Convert an `.sdpub` archive into the current `.wikg` format.
+
+Usage:
+  {{ commandName }} legacy migrate <path.sdpub> [--output <path.wikg>]
+
+Output:
+  If `--output` is omitted, the output path is the input path with `.sdpub` replaced by `.wikg`.
+  The input archive is not modified.
+
+Examples:
+  {{ commandName }} legacy migrate ./book.sdpub
+  {{ commandName }} legacy migrate ./book.sdpub --output ./book.wikg
+
+Related help:
+  - Current archive format, create/import, and export boundaries:
+    {{ commandName }} help format

package/data/help/commands/legacy.jinja

@@ -0,0 +1,19 @@
+Help Type: command
+Command: {{ commandName }} legacy
+
+Purpose:
+  Expose the single `.sdpub` upgrade command.
+
+Usage:
+  {{ commandName }} legacy migrate <path.sdpub> [--output <path.wikg>]
+
+Commands:
+  migrate    Convert an `.sdpub` archive into the current `.wikg` format.
+
+Examples:
+  {{ commandName }} legacy migrate ./book.sdpub
+  {{ commandName }} legacy migrate ./book.sdpub --output ./book.wikg
+
+See Also:
+  {{ commandName }} legacy migrate --help
+  {{ commandName }} help format