@wanshi-kg/wanshi 0.1.0

package/dist/core/llm/prompts/templates/partials/examples/research.md

@@ -0,0 +1,208 @@
+### Example 1: Paper introduction (proposed method + hypothesis)
+
+Input:
+
+## File information
+
+Path: `papers/blocksparse-attention.md`
+
+## Content to analyze
+```
+# Efficient Sparse Attention for Long-Context Transformers
+Authors: A. Rivera, L. Chen — Stanford NLP Lab
+
+We propose BlockSparse, a sparse-attention mechanism that lowers attention memory
+from O(n^2) to O(n log n). We hypothesize that most attention mass is local, so a
+windowed pattern loses little accuracy. On the PG-19 benchmark, BlockSparse cuts
+perplexity from 33.2 to 31.8 versus a dense baseline. The design builds on the
+Longformer windowed-attention approach (Beltagy et al., 2020).
+```
+
+Output:
+
+```json
+{
+  "entities": [
+    {
+      "name": "BlockSparse",
+      "entityType": "methodology",
+      "observations": [
+        "Sparse-attention mechanism for long-context transformers",
+        "Lowers attention memory from O(n^2) to O(n log n)",
+        "Cuts PG-19 perplexity from 33.2 to 31.8 versus a dense baseline"
+      ]
+    },
+    {
+      "name": "locality of attention",
+      "entityType": "hypothesis",
+      "observations": ["Most attention mass is local, so a windowed pattern loses little accuracy"]
+    },
+    {
+      "name": "A. Rivera",
+      "entityType": "researcher",
+      "observations": ["Co-author of the BlockSparse paper"]
+    },
+    {
+      "name": "Stanford NLP Lab",
+      "entityType": "institution",
+      "observations": ["Affiliation of the BlockSparse authors"]
+    },
+    {
+      "name": "PG-19",
+      "entityType": "dataset",
+      "observations": ["Long-document language-modeling benchmark used for evaluation"]
+    },
+    {
+      "name": "Longformer",
+      "entityType": "methodology",
+      "observations": ["Windowed-attention approach of Beltagy et al., 2020"]
+    }
+  ],
+  "relations": [
+    { "from": "BlockSparse", "to": "locality of attention", "relationType": ["tests"] },
+    { "from": "BlockSparse", "to": "PG-19",                  "relationType": ["uses"] },
+    { "from": "BlockSparse", "to": "Longformer",             "relationType": ["builds_upon"] },
+    { "from": "A. Rivera",   "to": "BlockSparse",            "relationType": ["produces"] },
+    { "from": "A. Rivera",   "to": "Stanford NLP Lab",       "relationType": ["member_of"] }
+  ]
+}
+```
+
+Note: the perplexity figures (33.2 → 31.8) and the `O(...)` bounds are measured
+values — observations on the method, not their own nodes. `methodology` is reused
+for both the proposed method and the prior one it builds on; the type set stays small.
+
+---
+
+### Example 2: Experiment + statistical result (with prior context — extend, don't duplicate)
+
+Input:
+
+## File information
+
+Path: `papers/blocksparse-attention.md`
+Chunk 4 of 7
+
+## Already extracted (do not re-emit)
+
+- **BlockSparse** (methodology): Sparse-attention mechanism for long-context transformers
+- **PG-19** (dataset): Long-document language-modeling benchmark used for evaluation
+
+## Content to analyze
+```
+We ran an ablation across 4 random seeds comparing BlockSparse against the dense
+baseline. The perplexity reduction was significant (paired t-test, p < 0.01).
+Throughput, measured in tokens/second, rose 2.3x at 16k context length.
+```
+
+Output:
+
+```json
+{
+  "entities": [
+    {
+      "name": "BlockSparse ablation",
+      "entityType": "experiment",
+      "observations": [
+        "Ablation across 4 random seeds against the dense baseline",
+        "Throughput rose 2.3x at 16k context length"
+      ]
+    },
+    {
+      "name": "paired t-test",
+      "entityType": "statistical_test",
+      "observations": ["Perplexity reduction significant at p < 0.01"]
+    },
+    {
+      "name": "throughput",
+      "entityType": "variable",
+      "observations": ["Measured in tokens/second"]
+    }
+  ],
+  "relations": [
+    { "from": "BlockSparse ablation", "to": "BlockSparse",   "relationType": ["validates"] },
+    { "from": "BlockSparse ablation", "to": "throughput",    "relationType": ["measures"] },
+    { "from": "BlockSparse ablation", "to": "paired t-test", "relationType": ["uses"] }
+  ]
+}
+```
+
+Note: `p < 0.01`, `4 seeds`, and `2.3x` are observations, not nodes. `BlockSparse`
+is referenced by its exact prior name in a relation without being re-emitted (it
+gained no new fact here). One predicate per edge — no `["validates", "tests"]`.
+
+---
+
+### Example 3: Related work (citations + a contradicted finding)
+
+Input:
+
+## File information
+
+Path: `papers/blocksparse-attention.md`
+Chunk 6 of 7
+
+## Already extracted (do not re-emit)
+
+- **BlockSparse** (methodology): Sparse-attention mechanism for long-context transformers
+
+## Content to analyze
+```
+## Related Work
+Child et al. (2019) introduced fixed sparse-attention patterns. Smith et al. (2021)
+reported that sparsity yields no perplexity gain at scale — a claim our PG-19
+results contradict.
+```
+
+Output:
+
+```json
+{
+  "entities": [
+    {
+      "name": "Sparse Transformer (Child et al., 2019)",
+      "entityType": "publication",
+      "observations": ["Introduced fixed sparse-attention patterns"]
+    },
+    {
+      "name": "Smith et al. (2021)",
+      "entityType": "publication",
+      "observations": ["Reported that sparsity yields no perplexity gain at scale"]
+    }
+  ],
+  "relations": [
+    { "from": "BlockSparse", "to": "Sparse Transformer (Child et al., 2019)", "relationType": ["cites"] },
+    { "from": "BlockSparse", "to": "Smith et al. (2021)",                      "relationType": ["contradicts"] }
+  ]
+}
+```
+
+Note: a cited prior work is a `publication`; `cites` records the reference, while
+`contradicts` records the disagreement in findings — two distinct predicates, never
+stacked. Direction is consistent (citing work → cited work).
+
+---
+
+### Example 4: Pure notation → empty graph
+
+Input:
+
+## File information
+
+Path: `papers/blocksparse-attention.md`
+Chunk 7 of 7
+
+## Content to analyze
+```
+$$ \mathcal{L}(\theta) = -\sum_{i=1}^{N} \log p_\theta(x_i) + \lambda \lVert \theta \rVert_2^2 $$
+where $\theta \in \mathbb{R}^d$ and $\lambda > 0$.
+```
+
+Output:
+
+```json
+{ "entities": [], "relations": [] }
+```
+
+Note: an isolated equation with no named concept to attach it to yields nothing —
+bare symbols and formulas are observations at most, never nodes.

package/dist/core/llm/prompts/templates/partials/examples/tabular.md

@@ -0,0 +1,178 @@
+### Example 1
+
+Input:
+
+## File Information
+
+Path: `data/inventory/products_2024q1.csv`
+
+## Content to Analyze
+
+```
+SKU,Name,Category,Price,Stock,Supplier
+PRD-001,USB-C Hub 7-Port,Accessories,49.99,312,TechSource Inc
+PRD-002,Mechanical Keyboard TKL,Peripherals,129.99,87,KeyCraft
+PRD-003,27in 4K Monitor,Displays,399.99,44,ViewMax
+PRD-004,Laptop Stand Aluminum,Accessories,39.99,201,TechSource Inc
+PRD-005,Wireless Mouse Ergonomic,Peripherals,69.99,156,KeyCraft
+```
+
+Output:
+
+```json
+{
+  "entities": [
+    {
+      "name": "products_2024q1",
+      "entityType": "table",
+      "observations": [
+        "Inventory snapshot for Q1 2024",
+        "6 columns: SKU, Name, Category, Price, Stock, Supplier",
+        "5 product rows across 3 categories"
+      ]
+    },
+    {
+      "name": "Accessories",
+      "entityType": "category",
+      "observations": [
+        "Contains PRD-001 (USB-C Hub) and PRD-004 (Laptop Stand)",
+        "Both products supplied by TechSource Inc",
+        "Price range: $39.99–$49.99"
+      ]
+    },
+    {
+      "name": "Peripherals",
+      "entityType": "category",
+      "observations": [
+        "Contains PRD-002 (Mechanical Keyboard) and PRD-005 (Wireless Mouse)",
+        "Both products supplied by KeyCraft",
+        "Price range: $69.99–$129.99"
+      ]
+    },
+    {
+      "name": "Displays",
+      "entityType": "category",
+      "observations": [
+        "Contains PRD-003 (27in 4K Monitor)",
+        "Highest unit price at $399.99",
+        "Lowest stock level at 44 units"
+      ]
+    },
+    {
+      "name": "TechSource Inc",
+      "entityType": "supplier",
+      "observations": ["Supplies all Accessories category products"]
+    },
+    {
+      "name": "KeyCraft",
+      "entityType": "supplier",
+      "observations": ["Supplies all Peripherals category products"]
+    },
+    {
+      "name": "ViewMax",
+      "entityType": "supplier",
+      "observations": ["Sole supplier of Displays category"]
+    }
+  ],
+  "relations": [
+    { "from": "products_2024q1", "to": "Accessories",    "relationType": ["contains"] },
+    { "from": "products_2024q1", "to": "Peripherals",    "relationType": ["contains"] },
+    { "from": "products_2024q1", "to": "Displays",       "relationType": ["contains"] },
+    { "from": "TechSource Inc",  "to": "Accessories",    "relationType": ["supplies"] },
+    { "from": "KeyCraft",        "to": "Peripherals",    "relationType": ["supplies"] },
+    { "from": "ViewMax",         "to": "Displays",       "relationType": ["supplies"] }
+  ]
+}
+```
+
+---
+
+### Example 2
+
+Input:
+
+## File Information
+
+Path: `db/schema/orders.sql`
+
+## Content to Analyze
+
+```sql
+CREATE TABLE customers (
+    id          UUID PRIMARY KEY DEFAULT gen_random_uuid(),
+    email       TEXT NOT NULL UNIQUE,
+    created_at  TIMESTAMPTZ NOT NULL DEFAULT now()
+);
+
+CREATE TABLE products (
+    id          UUID PRIMARY KEY DEFAULT gen_random_uuid(),
+    sku         TEXT NOT NULL UNIQUE,
+    name        TEXT NOT NULL,
+    price_cents INTEGER NOT NULL CHECK (price_cents > 0)
+);
+
+CREATE TABLE orders (
+    id          UUID PRIMARY KEY DEFAULT gen_random_uuid(),
+    customer_id UUID NOT NULL REFERENCES customers(id) ON DELETE CASCADE,
+    created_at  TIMESTAMPTZ NOT NULL DEFAULT now(),
+    status      TEXT NOT NULL DEFAULT 'pending'
+);
+
+CREATE TABLE order_items (
+    id          UUID PRIMARY KEY DEFAULT gen_random_uuid(),
+    order_id    UUID NOT NULL REFERENCES orders(id) ON DELETE CASCADE,
+    product_id  UUID NOT NULL REFERENCES products(id),
+    quantity    INTEGER NOT NULL CHECK (quantity > 0),
+    unit_price_cents INTEGER NOT NULL
+);
+```
+
+Output:
+
+```json
+{
+  "entities": [
+    {
+      "name": "customers",
+      "entityType": "table",
+      "observations": [
+        "Primary key: UUID with gen_random_uuid() default",
+        "email must be unique and not null",
+        "Soft-delete cascade propagates to orders"
+      ]
+    },
+    {
+      "name": "products",
+      "entityType": "table",
+      "observations": [
+        "SKU is unique identifier alongside UUID primary key",
+        "price_cents enforces positive integer constraint",
+        "No cascade delete — orphan order_items are prevented by FK"
+      ]
+    },
+    {
+      "name": "orders",
+      "entityType": "table",
+      "observations": [
+        "Links to customers via customer_id foreign key",
+        "Default status is 'pending'",
+        "Cascade deletes propagate to order_items"
+      ]
+    },
+    {
+      "name": "order_items",
+      "entityType": "table",
+      "observations": [
+        "Junction table between orders and products",
+        "Stores unit_price_cents at time of purchase (denormalized for history)",
+        "quantity must be positive"
+      ]
+    }
+  ],
+  "relations": [
+    { "from": "orders",       "to": "customers",  "relationType": ["foreign_key_to"] },
+    { "from": "order_items",  "to": "orders",     "relationType": ["foreign_key_to"] },
+    { "from": "order_items",  "to": "products",   "relationType": ["foreign_key_to"] }
+  ]
+}
+```

package/dist/core/llm/prompts/templates/partials/examples/transcript.md

@@ -0,0 +1,204 @@
+### Example 1
+
+Input:
+
+## File Information
+
+Path: `meetings/2024-03-15-sprint-retro.txt`
+
+## Content to Analyze
+
+```
+Sprint 12 Retrospective — 2024-03-15
+Attendees: Sarah Chen (EM), David Park (Backend), Aisha Osei (Frontend), Tom Rivera (QA)
+Facilitator: Sarah Chen
+
+WHAT WENT WELL
+- CI pipeline improvements cut build time by 40%
+- New feature flags system shipped on time
+
+WHAT NEEDS IMPROVEMENT
+- API contract changes broke the mobile app twice (no shared schema validation)
+- QA was pulled in too late; several bugs caught post-deploy
+
+DECISIONS
+1. All API changes must include an OpenAPI diff reviewed by QA before merging.
+2. QA joins sprint planning from Sprint 13 onward.
+
+ACTION ITEMS
+- David Park: Add openapi-diff step to GitHub Actions by 2024-03-22
+- Aisha Osei: Integrate shared schema package into frontend build by 2024-03-29
+- Sarah Chen: Update Definition of Done to include schema review gate
+- Tom Rivera: Create QA onboarding template for sprint planning by 2024-03-20
+```
+
+Output:
+
+```json
+{
+  "entities": [
+    {
+      "name": "Sprint 12 Retrospective",
+      "entityType": "meeting",
+      "observations": [
+        "Held on 2024-03-15",
+        "Facilitated by Sarah Chen",
+        "Focus: process improvements after two API breakages"
+      ]
+    },
+    {
+      "name": "Sarah Chen",
+      "entityType": "person",
+      "observations": ["Engineering Manager", "Facilitator for the retro", "Owns Definition of Done update"]
+    },
+    {
+      "name": "David Park",
+      "entityType": "person",
+      "observations": ["Backend engineer", "Responsible for OpenAPI diff CI step"]
+    },
+    {
+      "name": "Aisha Osei",
+      "entityType": "person",
+      "observations": ["Frontend engineer", "Owns shared schema integration into frontend build"]
+    },
+    {
+      "name": "Tom Rivera",
+      "entityType": "person",
+      "observations": ["QA engineer", "Tasked with creating QA sprint planning onboarding template"]
+    },
+    {
+      "name": "Add openapi-diff to GitHub Actions",
+      "entityType": "action_item",
+      "observations": ["Due 2024-03-22", "Prevents unreviewed API contract changes from merging"]
+    },
+    {
+      "name": "Integrate shared schema into frontend",
+      "entityType": "action_item",
+      "observations": ["Due 2024-03-29", "Assigned to Aisha Osei"]
+    },
+    {
+      "name": "QA joins sprint planning",
+      "entityType": "decision",
+      "observations": ["Effective from Sprint 13", "Addresses late QA involvement causing post-deploy bugs"]
+    }
+  ],
+  "relations": [
+    { "from": "Sarah Chen",  "to": "Sprint 12 Retrospective",           "relationType": ["facilitates"] },
+    { "from": "David Park",  "to": "Sprint 12 Retrospective",           "relationType": ["attends"] },
+    { "from": "Aisha Osei",  "to": "Sprint 12 Retrospective",           "relationType": ["attends"] },
+    { "from": "Tom Rivera",  "to": "Sprint 12 Retrospective",           "relationType": ["attends"] },
+    { "from": "David Park",  "to": "Add openapi-diff to GitHub Actions","relationType": ["responsible_for"] },
+    { "from": "Aisha Osei",  "to": "Integrate shared schema into frontend", "relationType": ["responsible_for"] },
+    { "from": "Sprint 12 Retrospective", "to": "QA joins sprint planning", "relationType": ["decides"] }
+  ]
+}
+```
+
+---
+
+### Example 2
+
+Input:
+
+## File Information
+
+Path: `interviews/user-research-2024-04-02.txt`
+
+## Content to Analyze
+
+```
+Customer Discovery Interview — 2024-04-02
+Interviewer: Maya Patel (Product)
+Interviewee: Jordan Lee, CTO at Stackwise (B2B SaaS, 80 employees)
+
+MAYA: How does your team currently handle on-call escalation?
+
+JORDAN: We use PagerDuty, but the runbooks live in Confluence. When an alert fires at 2 AM,
+engineers have to context-switch between three tools to figure out what to do. It's painful.
+
+MAYA: What would make that better?
+
+JORDAN: Honestly, if the runbook context surfaced automatically inside the alert — even
+just the last five incidents for that service — that would cut mean time to resolve by half.
+
+MAYA: Have you tried any integrations?
+
+JORDAN: We built a custom Slack bot that pulls Confluence pages, but it broke twice in
+six months when Confluence changed their API. Maintenance is a nightmare.
+
+PAIN POINTS SUMMARY
+- Context fragmentation across PagerDuty, Confluence, Slack
+- Runbook discoverability during incidents
+- Brittle custom integrations
+
+FEATURE REQUESTS
+- Automatic runbook surfacing in alerts
+- Recent incident history per service shown inline
+- Stable API / webhook support to avoid custom bot maintenance
+```
+
+Output:
+
+```json
+{
+  "entities": [
+    {
+      "name": "Jordan Lee",
+      "entityType": "person",
+      "observations": [
+        "CTO at Stackwise",
+        "Manages on-call engineering team",
+        "Built a custom Slack bot for runbook retrieval"
+      ]
+    },
+    {
+      "name": "Maya Patel",
+      "entityType": "person",
+      "observations": ["Product manager", "Conducting customer discovery interview"]
+    },
+    {
+      "name": "Stackwise",
+      "entityType": "organization",
+      "observations": ["B2B SaaS company", "~80 employees"]
+    },
+    {
+      "name": "Context fragmentation across tools",
+      "entityType": "pain_point",
+      "observations": [
+        "Engineers must switch between PagerDuty, Confluence, and Slack during incidents",
+        "Particularly painful during 2 AM on-call alerts"
+      ]
+    },
+    {
+      "name": "Brittle custom Slack bot",
+      "entityType": "pain_point",
+      "observations": [
+        "Breaks when Confluence changes their API",
+        "Failed twice in six months",
+        "High maintenance burden"
+      ]
+    },
+    {
+      "name": "Automatic runbook surfacing in alerts",
+      "entityType": "feature_request",
+      "observations": [
+        "Show runbook content inline within alert notifications",
+        "Expected to cut MTTR by ~50% per Jordan Lee"
+      ]
+    },
+    {
+      "name": "Recent incident history per service",
+      "entityType": "feature_request",
+      "observations": ["Last 5 incidents for the affected service shown inline in alert"]
+    }
+  ],
+  "relations": [
+    { "from": "Jordan Lee",  "to": "Stackwise",                              "relationType": ["works_at"] },
+    { "from": "Maya Patel",  "to": "Jordan Lee",                             "relationType": ["interviews"] },
+    { "from": "Jordan Lee",  "to": "Context fragmentation across tools",     "relationType": ["reports"] },
+    { "from": "Jordan Lee",  "to": "Brittle custom Slack bot",               "relationType": ["reports"] },
+    { "from": "Jordan Lee",  "to": "Automatic runbook surfacing in alerts",  "relationType": ["requests"] },
+    { "from": "Jordan Lee",  "to": "Recent incident history per service",    "relationType": ["requests"] }
+  ]
+}
+```

package/dist/core/llm/prompts/templates/partials/retrieved-context.hbs

@@ -0,0 +1,18 @@
+{{#if retrievedEntities}}
+## Related Context
+
+The following entities were found in previously processed files that may be relevant:
+
+{{#each retrievedEntities}}
+### {{name}}
+- **Type**: {{entityType}}
+- **File**: {{file}}
+{{#if observations}}
+- **Key Facts**:
+{{#each observations}}
+  - {{this}}
+{{/each}}
+{{/if}}
+
+{{/each}}
+{{/if}}