index-note

Index and Network Maintenance (index-note)

Positioning

Knowledge network entry system and structural health expert. Core Concept: An index is an "entry point" rather than a "category"—restrained, carefully selected, reachable. It does not ask "which category does this belong to," only "where can it be found." (03_Index = MOC, using links to point to atomic notes in 05_Daily Records)

Navigation and Contract (when to read what)

Operation Mode	Trigger Scenario / Core Action	Basis Documents That Must Be Loaded Before Execution
Mode 1: Create Index	Establish a hub point for a brand-new topic	Read references/index-design-concept.md to determine boundaries, read references/output-templates.md to get the corresponding structure
Mode 2: Network Repair	Analyze the checkup report and prescribe remedies for network defects	Read references/output-templates.md to get the prescription template
Mode 3: Content Networking	Passively or manually establish network connections for scattered notes	Read references/入网流程.md to determine the operation to perform
Mode 4: Daily Networking	Daily one-click wrap-up 05_每日记录 latest notes collection and distribution	Read references/output-templates.md to get the aggregation checklist plan

---

Mode 1: Create Index

Mandatory Requirement: Indexes are named after the domain (e.g., 索引_营销), and must not contain book titles/article titles/authors/dates.

1. Read the Concept Dictionary First: Load references/index-design-concept.md to determine whether the current need should establish a "Concept Knowledge Index" or a "Project Event Index" that includes a timeline. Clearly distinguish the topic boundaries of the index.
2. Cluster and Filter Entry Points: Search the entire library for cards related to this domain. Sort out their internalkeyword clusters. Select nodes that can truly serve as entry points to answer search intent; there is no need to list everything indiscriminately.
3. Load Template and Create File: Read references/output-templates.md, and use the corresponding index skeleton to output the brand-new content node.
4. Add Backtracking Jump Points: Ensure that all associated underlying cards have a return link to the new index inserted at the bottom (bidirectionally reachable).

Mode 2: Network Repair

Input and Output: Use the network-health report as the input analysis, and produce an executable checkup repair prescription. This skill is not responsible for recalculating the total count and network layer thickness data.

1. Read Data Source: Extract the latest report from memory/network-health/ , and retrieve: island list, high-frequency dead links, hub bridge notes.
2. Identify Hidden Risks in Deep Waters (index-layer-specific judgment)：
   • Check Topic Vacuums: Search for whether there are notes clustered together (more than 5 notes on the same concept in the last 3 months) but with no corresponding 03_索引at the MOC layer, and propose a "missing topic index."
   • Clear Dormant Anchors: There are nodes at the bottom of the spot check that have SEED tags and have not moved for too long—propose activation, bridge connections, or have them participate in the next random roaming verification.
   • Island Rescue Analysis: For thrown-out islands, only take responsibility for finding an attributable index or hooking them to nearby nodes; do not force-fit them.
3. Submit Prescription: Read references/output-templates.md 's [Mode 2 Template] and list detailed repair countermeasures to the user. After user confirmation, mount and connect them one by one. Note: Because dead links are usually paths changed by moving files, do not directly operate on dead links on behalf of the user.

Mode 3: Content Networking (Single/Batch)

Objective Requirements: Use a point-to-point approach to lock new free-floating knowledge into the graph (≥2 path principle). The core of executing this mode is to use the 03_folder as an MOC system distribution pool.

1. Define Scope: Clarify whether this session is assigned one file or a group of files, as the to-do list of entities to be networked.
2. Study and Determine Rules: Load and read through references/入网流程.md 's handling methods for extracting claims/classification IF-ELSE.
3. Implement and Announce: Implement additions to existing index entries, or independently generate new graph bidirectional link intersections. After completion, read references/output-templates.md [Mode 3] section's provided syntax to send the "Batch Networking Report".

Mode 4: Daily Networking (Clean Inbox)

Objective Requirements: The fallback action of daily clearing and daily settling. Connect today's unplaced new cards back to the network's main artery to prevent them from being lost with the wind and "disconnecting from the network".

1. Lock Increment: Filter out independent articles created today under 05_每日记录/YYYY/MM/YYYYMMDD that have not yet been recorded with 归网：✅ .
2. Quick Pre-read and Qualify: Read by priority YAML frontmatter → title → first paragraph. First extract tags(candidate index main signal) and links(existing connections; skip duplicate networking during networking); when tags exist, usually no need to read the body to complete the judgment; when tags are missing or point unclearly, then read the first paragraph for supplementation. Judgment: Does this card belong to the project category? Knowledge category? Both? Or is it undecided and should not be forced into a category?
3. Load Template and Write Plan: Read references/output-templates.md and strictly follow [Mode 4 Today's Networking Plan]. Merge and display files belonging tothe same index, rejecting loose, sand-like流水 list displays. Display the modified connection sketch and timeline action description.
4. Wait for Confirmation to Execute: After accepting user instructions, uniformly inject at the end of each card:← 已归网至 [[索引_XXX]] accompanied by 归网：✅ YYYY-MM-DD confirmation wording and write the processing log in memory/入网/YYYYMMDD.md .
5. Open Loops Wrap-up: After networking is complete, review all Open loops in today's memory/YYYY-MM-DD.md log, and judge each one: closed loop → cross out; not closed but will naturally be encountered next time → leave in the log; not closed and "if not specifically sought out, it will be forgotten" → elevate to memory/open loops.md(attach source note link). After closing the loop, delete from open loops.md , do not archive.

---

Quality and Review CheckList (Required)

• Execute Pre-judgment: Before clearly executing the current process, whether it has already andmust first readthe references plug-in auxiliary document specified in the table above? (to prevent hallucinated template operations by large models)
• Restrained and concise review: Does the output and the constructed index page exceed one screen? Does every entry node added (especially in Mode 4) provide a "one-sentence core value summary" for that node?
• Closed-loop completeness: Regardless of which mode is executed, do all lower-level atomic notes that are attached to the index not only add pointing relationships but also add clear label identification?
• Mode 4 aggregation degree: In the daily network-return plan, has content with the same attribution already been gathered and displayed? Has content that cannot be decided been separately set up as "Pending Judgment," preventing irresponsible and crude merging?

---

Appendix: Batch network-return automation extension (assetized implementation)

For the large-scale queue of unclosed network-return notes accumulated after confirmation, you can directly use the scaffold provided by the chassis to complete idempotent closed-loop operations, without the model needing to copy them word by word:

• Execution method: After ensuring the confirmation list architecture is intact, execute in the terminal: python3 ".agents/skills/index-note/scripts/batch_network_integration.py"
• The script will actively find the mapping targets already in the list, automatically penetrate to complete reverse link filling and end seal annotation, and nodes that have already been identified will be harmlessly skipped.