Domain Knowledge Sweep
Use when the user wants to audit an engineering domain comprehensively (e.g., "strengthen llm-wiki with X" / "research domain Y" / "find gaps in our hydrodynamics coverage") so that:
- Standards, academic sources, industry practice, and codebase coverage are all surveyed
- A defensible coverage map is produced with proper citations
- Implementation gaps are spawned as actionable subissues
- LinkedIn is captured as marketing-surface only (not primary technical source)
When this skill is the right fit
- User asks to research, strengthen, or audit a technical domain end-to-end
- User mentions parallel-account dispatch (multiple AI accounts available)
- Coverage gaps in llm-wiki, digitalmodel, or related repos need systematic discovery
- A LinkedIn post or single source kicked off the request but completeness matters
When NOT to use
- User wants a one-off single-source extraction → use
field-dev-code-recon instead
- User wants to ingest a specific document already in hand → use
gsd:ingest-docs
- User wants implementation of a known gap → use normal feature-dev pipeline
Parent feature
#2667 — Domain Knowledge Sweep meta-feature. Every domain sweep spawned via this skill must reference this parent.
Workflow
Phase 1: Domain Charter
Confirm with user:
- Domain name (e.g., "Offshore Hydrodynamics", "Mooring Design")
- Scope IN (what topics are inside the domain)
- Scope OUT (what's a separate domain to avoid overlap)
- Existing codebase anchor (which modules to audit)
Phase 2: Spawn Parent Issue
Use template at templates/domain-parent-issue.md. Title format:
Domain Sweep: <Domain Name> (<key concepts>)
Labels: enhancement, priority:medium, cat:knowledge-domain, cat:engineering
Phase 3: Spawn 6 Research Subissues (sequential)
To avoid the feedback_parallel_gh_issue_create_reverses_numbers pitfall, create subissues sequentially. Templates in templates/:
| ID |
Title pattern |
Account |
Template |
| R1 |
Standards & Codes inventory |
2 (rigor) |
r1-standards.md |
| R2 |
Academic sources sweep |
2 (rigor) |
r2-academic.md |
| R3 |
Industry practice |
3 (broad) |
r3-industry.md |
| R4 |
LinkedIn expert mapping |
3 (broad) |
r4-marketing.md |
| R5 |
Code coverage audit |
1 (main) |
r5-code-audit.md |
| R6 |
Synthesis + gap spawning |
1 (main) |
r6-synthesis.md |
Phase 4: Update Parents with Cross-refs
After all 6 are created:
- Comment on Domain Parent with subissue tree + account assignments
- Comment on Feature Parent (#2667) with Domain entry added to queue
Phase 5: Dispatch (User-Driven)
The user dispatches R1-R4 in parallel across accounts. Account 1 handles R5 (after R1+R2) and R6 (after all R1-R5).
Account Distribution Playbook
| Account |
Streams |
Rationale |
| 1 (Claude / main) |
R5 (code audit), R6 (synthesis) |
Needs holistic codebase context + synthesis judgment |
| 2 (high-rigor: Codex or Gemini) |
R1 (standards), R2 (academic) |
Requires careful sourcing, citation discipline |
| 3 (broader sweep) |
R3 (industry), R4 (LinkedIn) |
Lower-rigor but wider net for current practice |
Rules to honor
- Citations: All gap subissues must conform to
.claude/rules/calc-citation-contract.md
- LinkedIn = marketing surface only: Per
feedback_llm_wiki_concept_pages_need_public_references, never import LinkedIn content as primary technical source. R4 is for outreach tracking, NOT technical extraction.
- Sequential issue creation: Per
feedback_parallel_gh_issue_create_reverses_numbers, don't bake cross-refs at create-time when parallelizing.
- Inline issue refs: Per
feedback_inline_gh_issue_url, render #NNNN as Markdown hyperlink in chat output.
- No self-approval: Per
feedback_never_offer_to_self_label_plan_approved, never self-approve. Implementation gaps spawned by R6 still need user approval before code work begins.
Domain queue
Priority order (update as domains complete):
- Offshore Hydrodynamics — #2668 launched 2026-05-12
- Mooring Design — queued (40 knowledge seeds anchor)
- Subsea Pipelines — queued
- VIV / Riser Dynamics — queued
- CCS / CO2 Transport — queued
Output artifacts
Each domain sweep produces:
docs/field-development/<domain>-coverage-map.md (R6 deliverable)docs/standards/<domain>-inventory.yaml (R1 deliverable)docs/research/<domain>-academic-references.md (R2 deliverable)docs/research/<domain>-industry-practice.md (R3 deliverable)docs/marketing/<domain>-linkedin-experts.yaml (R4 deliverable)- Gap implementation subissues in
digitalmodel (or relevant repo)
- llm-wiki ingestion checklist for synthesized concepts
Related skills
field-dev-code-recon — single-source predecessor pattern (still useful for quick LinkedIn extraction)gsd:ingest-docs — for ingesting docs already in handgsd:new-milestone — when a domain sweep reveals enough work to constitute a milestone
Related memory
project_domain_knowledge_sweep — durable workflow recordproject_llm_wiki_strategic_role — why coverage gaps are first-class defectsfeedback_llm_wiki_concept_pages_need_public_references — root cause for shift from LinkedIn-only sourcing
1---2name: domain-knowledge-sweep3description: Systematic multi-source research of an engineering domain. Spawns parent issue → 6 research subissues (Standards, Academic, Industry, LinkedIn-marketing, Code-audit, Synthesis) → gap implementation subissues. Replaces LinkedIn-only extraction with defensible comprehensive sourcing.4---56# Domain Knowledge Sweep78Use when the user wants to audit an engineering domain comprehensively (e.g., "strengthen llm-wiki with X" / "research domain Y" / "find gaps in our hydrodynamics coverage") so that:9- Standards, academic sources, industry practice, and codebase coverage are all surveyed10- A defensible coverage map is produced with proper citations11- Implementation gaps are spawned as actionable subissues12- LinkedIn is captured as marketing-surface only (not primary technical source)1314## When this skill is the right fit1516- User asks to research, strengthen, or audit a technical domain end-to-end17- User mentions parallel-account dispatch (multiple AI accounts available)18- Coverage gaps in llm-wiki, digitalmodel, or related repos need systematic discovery19- A LinkedIn post or single source kicked off the request but completeness matters2021## When NOT to use2223- User wants a one-off single-source extraction → use `field-dev-code-recon` instead24- User wants to ingest a specific document already in hand → use `gsd:ingest-docs`25- User wants implementation of a known gap → use normal feature-dev pipeline2627## Parent feature2829[#2667](https://github.com/vamseeachanta/workspace-hub/issues/2667) — Domain Knowledge Sweep meta-feature. Every domain sweep spawned via this skill must reference this parent.3031## Workflow3233### Phase 1: Domain Charter3435Confirm with user:361. **Domain name** (e.g., "Offshore Hydrodynamics", "Mooring Design")372. **Scope IN** (what topics are inside the domain)383. **Scope OUT** (what's a separate domain to avoid overlap)394. **Existing codebase anchor** (which modules to audit)4041### Phase 2: Spawn Parent Issue4243Use template at `templates/domain-parent-issue.md`. Title format:44```45Domain Sweep: <Domain Name> (<key concepts>)46```47Labels: `enhancement, priority:medium, cat:knowledge-domain, cat:engineering`4849### Phase 3: Spawn 6 Research Subissues (sequential)5051To avoid the `feedback_parallel_gh_issue_create_reverses_numbers` pitfall, create subissues sequentially. Templates in `templates/`:5253| ID | Title pattern | Account | Template |54|----|---------------|---------|----------|55| R1 | Standards & Codes inventory | 2 (rigor) | `r1-standards.md` |56| R2 | Academic sources sweep | 2 (rigor) | `r2-academic.md` |57| R3 | Industry practice | 3 (broad) | `r3-industry.md` |58| R4 | LinkedIn expert mapping | 3 (broad) | `r4-marketing.md` |59| R5 | Code coverage audit | 1 (main) | `r5-code-audit.md` |60| R6 | Synthesis + gap spawning | 1 (main) | `r6-synthesis.md` |6162### Phase 4: Update Parents with Cross-refs6364After all 6 are created:651. Comment on Domain Parent with subissue tree + account assignments662. Comment on Feature Parent (#2667) with Domain entry added to queue6768### Phase 5: Dispatch (User-Driven)6970The user dispatches R1-R4 in parallel across accounts. Account 1 handles R5 (after R1+R2) and R6 (after all R1-R5).7172## Account Distribution Playbook7374| Account | Streams | Rationale |75|---------|---------|-----------|76| 1 (Claude / main) | R5 (code audit), R6 (synthesis) | Needs holistic codebase context + synthesis judgment |77| 2 (high-rigor: Codex or Gemini) | R1 (standards), R2 (academic) | Requires careful sourcing, citation discipline |78| 3 (broader sweep) | R3 (industry), R4 (LinkedIn) | Lower-rigor but wider net for current practice |7980## Rules to honor81821. **Citations:** All gap subissues must conform to `.claude/rules/calc-citation-contract.md`832. **LinkedIn = marketing surface only:** Per `feedback_llm_wiki_concept_pages_need_public_references`, never import LinkedIn content as primary technical source. R4 is for outreach tracking, NOT technical extraction.843. **Sequential issue creation:** Per `feedback_parallel_gh_issue_create_reverses_numbers`, don't bake cross-refs at create-time when parallelizing.854. **Inline issue refs:** Per `feedback_inline_gh_issue_url`, render `#NNNN` as Markdown hyperlink in chat output.865. **No self-approval:** Per `feedback_never_offer_to_self_label_plan_approved`, never self-approve. Implementation gaps spawned by R6 still need user approval before code work begins.8788## Domain queue8990Priority order (update as domains complete):91921. **Offshore Hydrodynamics** — [#2668](https://github.com/vamseeachanta/workspace-hub/issues/2668) launched 2026-05-12932. **Mooring Design** — queued (40 knowledge seeds anchor)943. **Subsea Pipelines** — queued954. **VIV / Riser Dynamics** — queued965. **CCS / CO2 Transport** — queued9798## Output artifacts99100Each domain sweep produces:101- `docs/field-development/<domain>-coverage-map.md` (R6 deliverable)102- `docs/standards/<domain>-inventory.yaml` (R1 deliverable)103- `docs/research/<domain>-academic-references.md` (R2 deliverable)104- `docs/research/<domain>-industry-practice.md` (R3 deliverable)105- `docs/marketing/<domain>-linkedin-experts.yaml` (R4 deliverable)106- Gap implementation subissues in `digitalmodel` (or relevant repo)107- llm-wiki ingestion checklist for synthesized concepts108109## Related skills110111- `field-dev-code-recon` — single-source predecessor pattern (still useful for quick LinkedIn extraction)112- `gsd:ingest-docs` — for ingesting docs already in hand113- `gsd:new-milestone` — when a domain sweep reveals enough work to constitute a milestone114115## Related memory116117- `project_domain_knowledge_sweep` — durable workflow record118- `project_llm_wiki_strategic_role` — why coverage gaps are first-class defects119- `feedback_llm_wiki_concept_pages_need_public_references` — root cause for shift from LinkedIn-only sourcing
vamseeachanta