loom-enhanced-research

/loom-enhanced-research

End-to-end enhanced research skill for iterative evidence gathering, user review, and draft shaping.

Mission

This skill turns a research goal into a bounded, reviewable workflow. It clarifies the request, runs explicit research rounds, preserves evidence and user input, presents gathered materials for review, collects structured and freeform comments through a simple UI path, generates a draft report, and lets the user either finalize, request more research, or re-select existing materials before the final report is published.

Inputs

final research goal
optional seed query or seed URLs
maximum depth for this run
maximum round count for this run
optional output root
evidence-retention policy
optional demo mode flag
optional user language for review surfaces and report output
optional native-language intake comments as first-class input

Workflow

Legend: 🧭 intake and setup, 🔎 research and evidence, 💬 user review, 🔁 continuation path, 📝 drafting, ✅ final output, ❓ decision gate.

flowchart TD
    A[🧭 Start: User provides research goal] --> B[🧭 Clarify inputs]
    B --> B1[🧭 Confirm goal, seed URLs or query, depth, round budget, output root, evidence policy, demo mode, and user language]
    B1 --> B2[🧭 Collect freeform native-language intake comments]
    B2 --> C[🧭 Initialize output root and ledgers]
    C --> C1[🧭 Create data, notes, materials, ui, qa-pairs, error-handling, report artifacts]

    C1 --> D[🔎 Enter research round loop]
    D --> D1[🔎 Round N: record trigger and working hypothesis]
    D1 --> D2[🔎 Choose action: search, dig, compare, fact-check, summarize]
    D2 --> D3[🔎 Capture evidence and round summary]
    D3 --> D4[🔎 Update ledgers]
    D4 --> E{❓ Continue research?}

    E -->|Yes| D
    E -->|No| F[🔎 Build full material inventory]

    F --> G[💬 Present all gathered materials to user]
    G --> G1[💬 Show sources, findings, excerpts, screenshots, round provenance]

    G1 --> H[💬 Open simple review UI site]
    H --> H1[💬 Collect structured selections]
    H1 --> H2[💬 Collect freeform native-language comments]
    H2 --> H3[💬 Emit structured continuation payload]

    H3 --> I{❓ Need another research pass before drafting?}
    I -->|Yes| J[🔁 Append selected context and comments to next research seed]
    J --> D
    I -->|No| K[📝 Generate report draft]

    K --> K1[📝 Include conclusion, scope, round history, evidence chain, cited sources, unresolved questions]
    K1 --> K2[📝 Include material review summary and cherry-pick summary]
    K2 --> L[📝 Draft-review checkpoint]
    L --> L1[📝 Collect user review decision]
    L1 --> L2[📝 Collect freeform native-language comments on draft]

    L2 --> M{❓ User-approved next action}
    M -->|Finalize report| N[✅ Publish final Markdown report]
    M -->|More research| O[🔁 Jump to bounded research loop]
    M -->|Re-select materials| P[🔁 Jump to cherry-pick loop]

    O --> D
    P --> G

    classDef intake fill:#E0F2FE,stroke:#0284C7,color:#0C4A6E;
    classDef research fill:#FEF3C7,stroke:#B45309,color:#78350F;
    classDef review fill:#FFEDD5,stroke:#EA580C,color:#9A3412;
    classDef continuation fill:#FCE7F3,stroke:#DB2777,color:#9D174D;
    classDef draft fill:#ECFCCB,stroke:#65A30D,color:#365314;
    classDef complete fill:#DCFCE7,stroke:#15803D,color:#14532D;
    classDef decision fill:#F1F5F9,stroke:#64748B,color:#334155;

    class A,B,B1,B2,C,C1 intake;
    class D,D1,D2,D3,D4,F research;
    class G,G1,H,H1,H2,H3 review;
    class J,O,P continuation;
    class K,K1,K2,L,L1,L2 draft;
    class N complete;
    class E,M decision;

Setup

The workflow starts by clarifying the research goal and the run contract. At this stage the skill confirms the goal, optional seeds, depth, round budget, output root, evidence policy, demo mode, and optional user language. Intake also includes freeform native-language comments as first-class input rather than treating them as side notes. It then creates the working artifacts needed for the rest of the run, including ledgers, materials, UI artifacts, and report outputs.

Research Loop

The research loop is the only part of the workflow that may create new evidence. Each round records a trigger, a working hypothesis, the chosen action, the evidence gathered, and the next-step decision. The loop continues only while the remaining budget and evidence state justify another pass.

Material Review

When bounded research stops, the workflow builds a full material inventory and presents it to the user. This stage is for inspecting gathered evidence rather than reviewing the written report draft. The material presentation should preserve provenance so the user can see where items came from and why they were retained.

UI Review Loop

After materials are presented, the workflow opens a simple review UI path. The user can make structured selections and also provide freeform comments in their native language. Both the structured selections and the freeform comments are preserved as first-class input and merged into the continuation payload. If the user wants another research pass before drafting, the workflow uses that payload to seed the next bounded research loop.

Drafting

When the workflow is ready to leave evidence gathering, it generates a report draft. That draft includes the conclusion, scope, round history, evidence chain, cited sources, unresolved questions, and the summary of material review and cherry-pick decisions.

Draft Review

The draft-review stage is distinct from material review. Here the user reviews the written report draft, not the raw evidence set. The user can provide a structured decision plus freeform native-language comments, after which the workflow branches to one of three destinations: finalize the report, return to bounded research, or return to material reselection.

Node Map

A entry point for the user research goal
B input clarification checkpoint
B1 confirmation of goal, seeds, budget, output root, evidence policy, demo mode, and user language
B2 freeform native-language intake comments as first-class input
C output-root initialization
C1 ledger and artifact creation
D bounded research loop start
D1 round trigger and hypothesis capture
D2 round action selection
D3 evidence capture and round summary creation
D4 ledger update step
E continue-or-stop research decision
F full material inventory build
G material presentation stage
G1 detailed material display contents
H review UI entry point
H1 structured user selections
H2 freeform native-language comments as first-class input
H3 continuation payload emission
I pre-draft re-research decision
J next-round seed enrichment
K report draft generation
K1 core draft assembly
K2 review-summary assembly
L draft-review checkpoint
L1 structured draft review decision
L2 freeform native-language draft comments as first-class input
M post-review branch decision
N final report publication
O branch back to bounded research
P branch back to material reselection

Rules

Every ask-user checkpoint must support both structured choices and freeform native-language comments.
Every user-facing review step must support both structured choices and freeform native-language comments.
Freeform native-language comments are first-class workflow input and must be preserved in the run ledger.
Only the research loop may create new evidence.
Returning to P reopens material review and reselection without claiming new evidence.
Returning to O re-enters bounded research and must preserve explicit round rationale and budget limits.
The draft-review stage reviews the written report draft, not the raw material inventory.

Outputs

material inventory
round trail and ledger updates
structured continuation payloads from user review
report draft with material review summary and cherry-pick summary
final Markdown report

Planning Source

Current workflow planning draft: demos/loom-enhanced-research/1.Planning/plan-loomEnhancedResearch.prompt.md