Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use BIRD Book Deconstructor 2.2 to apply the formal BIRD 2.1 Knowledge Address protocol to complex manuscript text and TheBrain Thoughts. Use when splitting books into chapter/section/item knowledge nodes, assigning Book Address, structured Knowledge Index (Weight, Type, Keyword, Alias), Routes, verified Deep Links, and Semantic Roles; producing BIRD Excel workbooks, TheBrain scaffolds, Roam JSON, monochrome printable double nine-grid cards, or a routed handoff from iMandalArt to A4 eight-page b
.claude/skills/twhsi-thebrain-bird-address/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | -2% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 49% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 119% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 64% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 86% | 0% |
Convert complex structural text into addressable, searchable, cross-linked book nodes for TheBrain, Excel, Roam Research, Obsidian, and AI Agents.
Version 2.2 is the Skill integration release. Canonical interchange objects remain BIRD-2.1 for backward compatibility.
BIRD is the address and route king in the three-skill publishing chain:
iMandalArt 2.2 eight angles -> BIRD 2.2 Book Address + Index + Route + Deep Link -> A4 Booklet 2.2
W + T + K + A Knowledge Indexes.B = Book Address: stable book/project hierarchy address; answer "where in the book?"I = Knowledge Index: structured index object W + T + K + A; answer "what knowledge object?"R = Route: ordered cross-node or cross-chapter paths; answer "where does it lead?"D = Deep Link: exact permanent application address; answer "how is it opened?"Read references/bird-2.1-spec.md before assigning Index Weight, Index Type, or Semantic Role.
Do not use the old B = Branch or unstructured I = title interpretation. Mark legacy input as BIRD 1.x and map it into 2.1. Preserve valid BIRD 2.0 fields during migration; add Role only when evidence supports it.
StructuralType: book function, such as 部 / 章 / 節 / 項.IndexType: knowledge kind inside I, using C / M / P / B / T / O / E / L / S / A / X.Never put 章 into IndexType, or Concept into StructuralType.
.xlsx with BIRD analysis, manuscript Notes, and code tables.BIRD分析 rows into a Roam-importable page/block array.待編 only when no reliable address exists; do not silently renumber.章 and 節 as structural containers. Treat 項 as the normal manuscript unit: one claim, explanation, evidence/example, and transition.B;I.W, one I.T, one canonical I.K, and zero or more deduplicated I.A;R targets when supported;D byte-for-byte or leave it blank/pending for proposed nodes.待補; never fabricate support.Read references/manuscript-structuring.md for segmentation and output templates.
For every node, apply this order:
K: one canonical retrieval term, 4-10 Chinese characters or 2-4 English words.T: one code from the controlled Index Type table.W: importance to this specific book, not general fame.A: genuine synonyms, translations, abbreviations, or established alternate spellings. Exclude the canonical keyword itself.Role only when graph or hierarchy evidence supports it; otherwise leave blank.Do not turn every noun into an Index. Create an index object only when the term supports retrieval, interpretation, or routing.
B for the table-of-contents location and R for meaningful cross-links.FIRE -> 語意索引 -> AI對話.待建 rather than inventing its address.Copy each supplied D exactly from brain:// through the final character. Never decode, encode, normalize, shorten, repair, rename, or regenerate its slug. For a proposed Thought, use a blank cell in Excel and 待建立 Thought 後貼入 in prose.
Before returning, compare every displayed D with its source string. Do not claim tool verification unless a tool returned that exact value in the current run.
Read references/excel-schema.md. Create these sheets:
BIRD分析: one row per knowledge node with normalized BIRD fields.拆書正文: one row per item with Note text and BIRD JSON.代碼表: Weight, Index Type, Semantic Role, StructuralType, and Tag dictionaries.Use a real spreadsheet library. Freeze headers, enable filters, wrap long text, set practical widths, and validate the workbook after saving. Preserve legacy columns only in a separate 舊表對照 sheet when migration is requested.
Read references/roam-json-import.md, then use the deterministic converter:
bashnode scripts/bird_to_roam_json.mjs bird.json roam-import.json
items, records, or flat Excel rows.uid by default; add UIDs only when preserving existing block references is explicitly required and collision risk has been audited.D_DeepLink byte-for-byte inside the D:: block.Read references/black-white-double-nine-grid.md whenever the user asks for a double nine-grid, print card, printable image, or left-directory/right-keyword visual.
Report the smallest correction for:
StructuralType and address-depth conflict;Do not move, rename, merge, or delete live Thoughts without explicit authorization.
text【BIRD 2.1】 B: 全系統/第三部/3.4/3.4.D/3.4.D.B I: W: I3 T: S | Skill K: 八領域週檢視 A: Weekly Review to 8 Rocks | 八岩週檢視 R: 週檢視 -> 八領域週檢視 -> 週計劃 D: brain://... or 待建立 Thought 後貼入 StructuralType: 項 Tag: 草稿 Role:
For Agent interchange, also output:
json{ "version": "BIRD-2.1", "bookAddress": "全系統/第三部/3.4/3.4.D/3.4.D.B", "index": { "weight": "I3", "typeCode": "S", "type": "Skill", "keyword": "八領域週檢視", "aliases": ["Weekly Review to 8 Rocks", "八岩週檢視"] }, "route": ["週檢視", "週計劃"], "deepLink": "brain://...", "structuralType": "項", "tag": "草稿", "semanticRole": null }
references/thebrain-instruction.md.references/thebrain-manuscript-instruction.md.references/thebrain-scaffold-import.md before creating any Thought, Type, Tag, or Link.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 28,894 | 20,366 | -30% | 1 | 1 | 0% | 5,728 | 5,626 | -2% | 0 | 0 | — |
case-02 | fail→pass | 32,793 | 33,996 | +4% | 1 | 1 | 0% | 5,825 | 8,654 | +49% | 0 | 0 | — |
case-03 | fail→fail | 32,140 | 29,572 | -8% | 1 | 1 | 0% | 6,226 | 8,674 | +39% | 0 | 0 | — |
case-04 | pass→pass | 12,068 | 9,500 | -21% | 1 | 1 | 0% | 2,178 | 4,183 | +92% | 0 | 0 | — |
case-05 | pass→pass | 14,461 | 19,021 | +32% | 1 | 1 | 0% | 2,113 | 6,113 | +189% | 0 | 0 | — |
case-06 | pass→pass | 13,378 | 12,817 | -4% | 1 | 1 | 0% | 2,343 | 4,655 | +99% | 0 | 0 | — |
case-07 | pass→pass | 9,201 | 5,014 | -46% | 1 | 1 | 0% | 1,407 | 3,270 | +132% | 0 | 0 | — |
case-08 | fail→pass | 10,422 | 4,680 | -55% | 1 | 1 | 0% | 1,467 | 3,211 | +119% | 0 | 0 | — |
case-09 | fail→pass | 12,779 | 4,943 | -61% | 1 | 1 | 0% | 1,992 | 3,274 | +64% | 0 | 0 | — |
case-10 | fail→pass | 17,192 | 14,209 | -17% | 1 | 1 | 0% | 2,705 | 5,024 | +86% | 0 | 0 | — |
case-11 | pass→pass | 8,704 | 4,031 | -54% | 1 | 1 | 0% | 1,223 | 3,054 | +150% | 0 | 0 | — |
case-12 | fail→pass | 24,383 | 2,856 | -88% | 1 | 1 | 0% | 1,761 | 2,860 | +62% | 0 | 0 | — |
case-13 | fail→pass | 11,170 | 4,991 | -55% | 1 | 1 | 0% | 1,598 | 3,255 | +104% | 0 | 0 | — |
case-14 | fail→pass | 14,290 | 2,148 | -85% | 1 | 1 | 0% | 2,487 | 2,824 | +14% | 0 | 0 | — |
case-19 | fail→pass | 11,995 | 7,950 | -34% | 1 | 1 | 0% | 1,783 | 3,860 | +116% | 0 | 0 | — |
case-15 | fail→pass | 12,089 | 3,930 | -67% | 1 | 1 | 0% | 1,826 | 3,030 | +66% | 0 | 0 | — |
case-16 | pass→pass | 14,125 | 4,003 | -72% | 1 | 1 | 0% | 2,009 | 2,970 | +48% | 0 | 0 | — |
case-17 | fail→pass | 28,136 | 1,998 | -93% | 1 | 1 | 0% | 2,248 | 2,821 | +25% | 0 | 0 | — |
case-18 | fail→pass | 10,510 | 3,897 | -63% | 1 | 1 | 0% | 1,716 | 3,066 | +79% | 0 | 0 | — |
case-20 | fail→pass | 11,586 | 2,595 | -78% | 1 | 1 | 0% | 1,642 | 2,855 | +74% | 0 | 0 | — |
case-21 | pass→pass | 11,642 | 8,232 | -29% | 1 | 1 | 0% | 1,810 | 3,660 | +102% | 0 | 0 | — |
case-22 | pass→pass | 12,089 | 7,448 | -38% | 1 | 1 | 0% | 1,940 | 3,635 | +87% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted. The headline lift of +59 percentage points is the difference between those two pass rates over the 22 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.