# Workflow Research > [Workflow] Research web sources and produce a cited, reviewed deliverable. [Workflow] Research web sources and produce a cited, reviewed deliverable. --output={synthesis|business-eval|marketing|course}. - Page: https://ss33.ai/skills/duc01226/workflow-research-99196570738 - Publisher: duc01226 - Category: [AI & Agents](https://ss33.ai/skills/category/ai-and-agents-y6jz7i) › [Orchestration & multi-agent](https://ss33.ai/skills/category/ai-and-agents-y6jz7i/orchestration-and-multi-agent-tlbx40) - License: MIT - Source: https://github.com/duc01226/easyplatform/blob/ed26f0b60f4ac2da864abc4c4eb5ca8c068aa5a8/.claude/skills/workflow-research/SKILL.md ## Skill SKILL.md: https://ss33.ai/api/v1/ss33/download/duc01226/workflow-research-99196570738 ## Workflow Calls and Todo Bootstrap Read [the registry](../../../.claude/workflows.json) → `workflows.workflow-research` together with this skill. Call [`/start-workflow workflow-research`](../start-workflow/SKILL.md) to resolve the selected mode, pre-actions and fingerprint. **Todo FIRST:** before triage, analysis or step execution, create one todo per phase of the selected mode, in order, listing every occurrence of that phase as a line, including conditional/optional ones (a mode that declares no phases gets one todo per occurrence); preserve occurrence IDs, roles and barrier groups. Use native todo tools or an equivalent persistent ledger. Then mark the first todo `in_progress`; close each step line with its run or skip evidence, and complete a phase todo only when every line is closed and its applicable exit gates hold. Mode selection and registry loading prepare tracking; any 'first action' below means the first substantive action after this bootstrap. **Call each step skill:** read its linked SKILL.md and execute its protocol through the active host with the registry args. Reading or naming a skill alone is not execution. Required gates and core/optional flex follow the linked start-workflow Step Execution Protocol; keep this skill's quality gates, loops and evidence requirements. **Conditions:** load each occurrence's registry `applicability.when` and `skipReason` verbatim, and record its run/skip evidence through the linked start-workflow protocol. Do not silently omit a todo or turn a conditional skill into an unconditional call; close, in one batch at triage, the todos whose condition is already false there. Declared parallel groups retain their all-return barrier. Explicit step-skill calls by mode (registry order; roles and conditions remain owned by the registry): - Mode `synthesis`: [`/web-research --chain=deep-dive`](../web-research/SKILL.md) (core) → [`/knowledge-synthesis`](../knowledge-synthesis/SKILL.md) (core) → [`/knowledge-review --fix-loop`](../knowledge-review/SKILL.md) (gate) → [`/workflow-end`](../workflow-end/SKILL.md) (gate) → [`/watzup`](../watzup/SKILL.md) (optional; conditional) Phase todos: Discover (3; exit review-converged) → Close (2; exit run-closed) - Mode `business-eval`: [`/web-research --chain=deep-dive`](../web-research/SKILL.md) (core) → [`/market-analysis`](../market-analysis/SKILL.md) (core) → [`/business-evaluation`](../business-evaluation/SKILL.md) (core) → [`/knowledge-review --fix-loop`](../knowledge-review/SKILL.md) (gate) → [`/workflow-end`](../workflow-end/SKILL.md) (gate) → [`/watzup`](../watzup/SKILL.md) (optional; conditional) Phase todos: Discover (4; exit review-converged) → Close (2; exit run-closed) - Mode `marketing`: [`/web-research --chain=deep-dive`](../web-research/SKILL.md) (core) → [`/market-analysis`](../market-analysis/SKILL.md) (core) → [`/strategy-builder`](../strategy-builder/SKILL.md) (core) → [`/knowledge-review --fix-loop`](../knowledge-review/SKILL.md) (gate) → [`/workflow-end`](../workflow-end/SKILL.md) (gate) → [`/watzup`](../watzup/SKILL.md) (optional; conditional) Phase todos: Discover (4; exit review-converged) → Close (2; exit run-closed) - Mode `course`: [`/web-research --chain=deep-dive`](../web-research/SKILL.md) (core) → [`/course-builder`](../course-builder/SKILL.md) (core) → [`/knowledge-review --fix-loop`](../knowledge-review/SKILL.md) (gate) → [`/workflow-end`](../workflow-end/SKILL.md) (gate) → [`/watzup`](../watzup/SKILL.md) (optional; conditional) Phase todos: Discover (3; exit review-converged) → Close (2; exit run-closed) Regenerate this block with `node .claude/scripts/lib/workflow-skill-contract.cjs --write` after registry edits; [`/sync-codex`](../sync-codex/SKILL.md) refreshes it before mirroring. ## Quick Summary **Goal:** Research a topic from web sources and deliver ONE cited, reviewed artifact in the form `--output` selects: a knowledge report (`synthesis`, default), a business/market viability evaluation (`business-eval`), a marketing strategy (`marketing`), or structured course material (`course`). The workflow produces research artifacts only, never code. **Use it when** the answer must come from external sources and end in a durable, cited deliverable. **Use a sibling instead** for a quick lookup with no artifact (plain `/web-research`) or for questions about this codebase (`/investigate`). **IMPORTANT MANDATORY Steps:** resolve the `workflow-research` manifest variant for `--output` first (default `synthesis`), then create one task per phase with a row per returned occurrence (default: /web-research --chain=deep-dive -> /knowledge-synthesis -> /knowledge-review --fix-loop -> /workflow-end -> /watzup). **Step contract:** steps follow `/start-workflow` → Step Execution Protocol — `gate` steps always run, a step that runs invokes its `Skill` tool, and every other deviation is logged. NEVER batch-complete validation gates. This skill is the canonical Research & Synthesis entry point. Invoke it with `--output=` (default `synthesis`); `/start-workflow workflow-research` resolves the same variants through `.claude/scripts/lib/workflow-manifest.cjs`. Each variant is a complete sequence, so execute the resolved manifest and never a hand-swapped list. ## Contents - [Workflow Calls and Todo Bootstrap](#workflow-calls-and-todo-bootstrap) - [Quick Summary](#quick-summary) - [Output Modes (--output)](#output-modes---output) - [Question-Scope Triage (first action)](#question-scope-triage-first-action) - [Required Quality Gates](#required-quality-gates) - [Recommended Skills](#recommended-skills) - [Orchestration](#orchestration) - [Memory & Reporting](#memory--reporting) - [Evidence Retention](#evidence-retention) - [Findings & Fix Path](#findings--fix-path) - [Closing Reminders](#closing-reminders) ## Output Modes (--output) Pick the mode from the prompt BEFORE creating tasks. When the prompt is ambiguous, use `synthesis` and state the assumption. Every mode shares one research step, `/web-research --chain=deep-dive` (source map, then evidence base; the `/source-deep-dive` procedure runs inside it per `.claude/skills/web-research/references/research-chain.md`), and the `/knowledge-review --fix-loop → /workflow-end` close, followed by `/watzup` only when its condition holds; only the terminal synthesis skill(s) differ. The `/market-analysis` step of `business-eval` and `marketing` reads that evidence base first and searches only the gaps it names. | `--output` | Deliverable | Terminal synthesis skill(s) | | ----------------------- | ------------------------------------ | ------------------------------------------- | | **synthesis** (default) | Cited knowledge report | `/knowledge-synthesis` | | **business-eval** | Business/market viability evaluation | `/market-analysis` → `/business-evaluation` | | **marketing** | Marketing strategy | `/market-analysis` → `/strategy-builder` | | **course** | Learner-ready course | `/course-builder` | **[BLOCKING] Evidence-artifact identity for `business-eval` and `marketing`:** before invoking the first research child, derive one stable `ARTIFACT_SLUG` from the user's topic and record the exact `MARKET_ANALYSIS_PATH = docs/knowledge/strategy/market-analysis/{ARTIFACT_SLUG}.md` in the parent workflow context/task handoff. Pass those exact values to every child skill. `market-analysis` is the only producer; it must write and return that path, and `business-evaluation`/`strategy-builder` must read that exact path rather than deriving a second slug. If a plan directory is active, its `{plan-dir}/research/market-analysis.md` file is a copy of the same producer artifact, not a second identity. This token is not needed for the `synthesis` variant. — why: sequence ordering without a shared artifact key still allows a producer/consumer miss that degrades the final evidence without failing the workflow. ## Question-Scope Triage (first action) Classify the question and record it in the workflow report. Scope sets research DEPTH inside the research step's caps (≤10 searches, then ≤8 fetches); it never lowers the evidence bar. | Scope | Signals | Depth | | ------------------------------ | --------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | **Narrow** | One fact, definition, or comparison of 2–3 named options | 3–5 angle-varied queries; fetch the 2–4 strongest Tier 1–2 sources; a short artifact that still carries every template section the review enforces. | | **Standard** | One topic with several angles | Skill defaults: 5–10 queries, 5–8 fetches. | | **Broad or decision-critical** | Multi-part topic, contested evidence, or a business/strategy decision rides on it | Split into sub-topics, one bounded research pass per sub-topic (each under the caps), then merge the source maps and evidence bases before synthesis. Run the adversarial checks of `/knowledge-review` at full depth. | ## Required Quality Gates | Gate | Evidence that proves it | | ------------------------------------------------ | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | Review converged (`review-converged`) | `/knowledge-review --fix-loop` verdict APPROVED on the final artifact: the gate validated its REVISE/BLOCKED findings, had them fixed at their owner and re-reviewed the fixed artifact in a fresh pass. | | Citation bar | Every factual claim, number, table row and inference ends with an inline `[N]` citation mapped to a Sources row (Title, URL, Author/Publisher, Date, Tier). | | Cross-validated and calibrated | Factual claims rest on 2+ independent sources; a single-source claim is marked unverified and held below 60%; confidence above 80% needs contradicting evidence addressed; Tier 4 is never cited as fact; findings below 60% and open gaps are flagged. | | Artifact identity (`business-eval`, `marketing`) | One `MARKET_ANALYSIS_PATH`, written by `market-analysis` and read by its consumer. | | Variant closure | The workflow report records the selected mode, resolver fingerprint and ordered occurrence IDs. | | Run closed (`run-closed`) | `/workflow-end` ran last. | ## Recommended Skills | Skill | Role | When it earns its cost | Proves / feeds | | --------------------------- | ---- | ---------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------- | | `/web-research --chain=deep-dive` | core | Always; depth per triage. Runs the `/source-deep-dive` procedure inline; the synthesis skills read its evidence base. | Tiered source map with gaps (`tmp/research/_sources-{slug}.md`), then the cross-validated evidence base (`tmp/research/_evidence-{slug}.md`). | | Terminal synthesis skill(s) | core | Per the Output Modes table. | The deliverable. | | `/knowledge-review --fix-loop` | gate | Always. It owns validate → fix → fresh re-review. | `review-converged`: template, citation, confidence, source-quality and anti-bias checks. | | `/workflow-end` | gate | Always, last. | `run-closed`, and the short handoff with the artifact path when nothing was built. | | `/watzup` | optional | The user asked for a session report or an HTML wrap-up, or the session spanned several workflows or prompts that need one consolidated handoff. Skip reason: workflow-end printed the recap, or the short handoff when nothing was built, and every applicable outcome gate passed. | Session report with the artifact path. | ## Orchestration You choose inline vs sub-agent, batching and ordering to minimize wall-clock and tokens at equal quality. Narrow questions run inline. For broad questions, independent sub-topic research passes can run as one parallel wave, because each writes only its own source and evidence files. Merge them before synthesis. Fixed data dependencies: inside the research step the source map exists before the deep dive; the evidence base exists before synthesis; `market-analysis` writes `MARKET_ANALYSIS_PATH` before its consumer reads it; `/knowledge-review` checks the final artifact; `/workflow-end` runs last. ## Memory & Reporting - Create one task per phase, with a row per resolved occurrence. Child skills expand their own steps under the parent row. - Create the workflow report FIRST at `tmp/reports/workflow-research-{YYMMDD}-{HHmm}-{slug}.md`: mode, triage, resolver fingerprint, per-step evidence and deviations, and the paths of the source map, evidence base and final artifact. Append after each step. - Sub-agent briefs carry the topic, the slug, the exact paths and the evidence bar, and make report writing the first deliverable. - After compaction, re-read `TaskList` and the workflow report before continuing. ## Evidence Retention Retain this run’s source map and evidence base during synthesis, all review rounds, and owner repairs. Cleanup is an orchestrator lifecycle action after `/knowledge-review` returns APPROVED on the final artifact and `/workflow-end` successfully accepts closure; record the acceptance evidence and exact removed paths in the workflow report. REVISE, BLOCKED, interruption, or closure failure retains the inputs. Remove only this run’s `_sources-{slug}.md` and `_evidence-{slug}.md` if its retention policy permits; never erase another run’s evidence. Cleanup after closure is report housekeeping, not another workflow occurrence. ## Findings & Fix Path - `/knowledge-review --fix-loop` owns the loop; each of its review passes is read-only. It validates each REVISE/BLOCKED finding against the evidence, has it fixed at its owner, then re-reviews the fixed artifact in a fresh pass. An evidence gap takes a targeted `/source-deep-dive` pass over the existing source map, or a `/web-research --chain=deep-dive` pass on the gap when the map lacks the needed sources. A synthesis defect (missing section, uncited claim, miscalibrated confidence) is fixed in the artifact through its synthesis skill. - Loop bounds: round 1 exits on zero open findings (LOW deferral); from round 2 the bar is zero CRITICAL/HIGH/MEDIUM, with LOWs deferred and listed; cap 3 review rounds; escalate via `ask user question tool` when a round makes no progress. > **Protocol guides** — Apply each protocol only when its read-when condition holds. If a protocol's text is not in your context, read its file below before you act on it. - `incremental-persistence` — Persist results per file or section while the work proceeds; a sub-agent or heavy step processes more than three files → .claude/skills/shared/protocols/incremental-persistence.md - `session-goal-ledger` — Keep the original goal and every user prompt of the session; running a long or multi-prompt session → .claude/skills/shared/protocols/session-goal-ledger.md - `subagent-return-contract` — Sub-agents return a structured envelope and a report path, never an inline report; spawning a sub-agent → .claude/skills/shared/protocols/subagent-return-contract.md - `workflow-registry-binding` — Read the workflow registry entry and the workflow skill together, since they must agree; executing or editing a workflow → .claude/skills/shared/protocols/workflow-registry-binding.md - **MANDATORY** Session goal ledger per the `Task Planning Rules`: pin `Original goal:`, keep `User prompts this session: P1…Pn`, and map the result to every prompt before claiming done; full text: `.claude/skills/shared/protocols/session-goal-ledger.md`. ## Closing Reminders **IMPORTANT MUST ATTENTION Goal:** one cited, reviewed artifact in the selected `--output` form, with `/knowledge-review` converged and every factual claim traceable to a Sources row. - **MUST ATTENTION** select the `--output` mode and triage the question scope FIRST. Scope sets research depth inside the skill caps; it never lowers the citation or cross-validation bar. - **MUST ATTENTION** for `business-eval` and `marketing`, derive one `ARTIFACT_SLUG` and `MARKET_ANALYSIS_PATH` before the first research step and pass the exact values to every child. - **MUST ATTENTION** `/knowledge-review --fix-loop` fixes validated findings at their owner (evidence gap → targeted `/source-deep-dive` or `/web-research --chain=deep-dive` pass, synthesis defect → the artifact), then re-reviews the fixed artifact in a fresh pass. - **MUST ATTENTION Variant closure:** record the selected `--output` mode, resolver fingerprint, ordered occurrence IDs, each invoked skill's evidence, and every deviation before `/workflow-end`; a variant mismatch is a workflow failure. **Protocols in force (digest; the guide entries above point to the full text):** Nested Task Creation · Incremental Persistence · Sub-Agent Return Contract · Session Goal Ledger · Workflow Registry Binding.