원클릭으로 Manus에서 모든 스킬 실행

ingest-client

스타1

포크0

업데이트2026년 6월 14일 21:54

Run the ingestion sweep that pulls a new client's raw knowledge into Dossier — public website (Firecrawl), files (Unstructured), and Microsoft 365/SharePoint (Graph delta + Copilot connectors) — normalized to clean, provenance-stamped markdown ready for extraction. Use when onboarding a client or refreshing their sources.

설치

Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.

Manus에서 실행

출처

smithdak

smithdak/dossier

GitHub 저장소 열기 Creator 저장소 보기

다운로드

Manus에서 실행

Ingest a Client

Sweep a client's sources into clean, deduplicated, provenance-stamped markdown — the input the author-okf / extraction-engineer turn into OKF concepts. Ingestion is assembled, not invented; reliability and provenance are the job.

Procedure

Inventory sources with the client: public site(s), SharePoint/M365, file shares, Confluence/Drive/etc., and any tribal-knowledge docs. Record access and scope.
Web → Firecrawl: Map to discover URLs, crawl for exhaustive coverage → clean markdown. (WebFetch for one-off pages.)
Files → Unstructured: partition PDFs/Office/HTML/email/images into semantic markdown elements.
Microsoft 365 / SharePoint → Graph delta query for incremental sync (sites, driveItem, listItem) and/or Copilot connectors (ACL-aware). Remember: deletes/ACL changes need a full crawl; Notion needs a custom connector.
Normalize → consistent clean markdown; deduplicate near-identical content.
Stamp provenance on everything — source URI, fetch timestamp, and access/ACL notes — so every downstream OKF atom can carry source/confidence. Provenance is sovereignty.
Stage the normalized output in the client's silo, ready for author-okf.

Rules

Tenant isolation: a client's raw data stays in their silo — never commingled. (DEC: siloed multi-tenancy.)
Incremental where possible — use delta tokens/change feeds; don't re-ingest the world.
Idempotent, resumable, rate-limit-aware. Capture failures to traces.

Status / dependencies

The connectors run on the Agent SDK runtime (platform-engineer). Until that's built, this skill orchestrates what's available (Firecrawl MCP / WebFetch for web; manual file drops for Unstructured) and records gaps for the platform-engineer to close.

Output

A staged, normalized, provenance-stamped corpus in the client's silo + a coverage report (what was ingested, what was skipped and why).

이 저장소의 다른 Skills

같은 저장소

verify-kb

smithdak/dossier

Audit an OKF knowledge base for integrity before it ships. Use to verify a generated/edited KB — checks OKF compliance, atomicity, link integrity, provenance/confidence, coverage gaps, and stale content. Produces a severity-ranked report; does not silently fix.

2026-06-211

adoption-readout

smithdak/dossier

Generate a client's L2 "adoption readout" — the Ladder (DEC-0052's wedge). From a provisioned client's served OKF, run the deterministic 4-axis scorer over their process/workflow atoms, then narrate a sequenced Now/Next/Later transformation roadmap and emit it as a sovereign OKF `artifact` in the CLIENT's own repo. Use after a client is ingested + served (KNOWLEDGE-BRIEF generated) to produce the post-diagnostic readout that lands the wedge ("we'll take you to where we already are") and opens the Agency conversation. Human-curated authoring; reads the client's OWN OKF, never perturbs Dossier's KB.

2026-06-201

docs-ia

smithdak/dossier

Architect a world-class, ATOMIC information architecture for a Dossier docs site — turn a flat OKF knowledge base into a navigable, scannable, readable reading experience whose structure is DERIVED FROM THE KNOWLEDGE GRAPH, never hand-maintained. Use when a docs surface is hard to navigate (giant flat sidebars, no section overviews, no wayfinding, every page a dead end), when onboarding a new client's KB to the docs, or when asked to improve docs IA / navigation / readability.

2026-06-191

generate-landing

smithdak/dossier

Generate a tailored client landing instance — a typed LandingContent module — from a provisioned client's OKF repo and tenant manifest. Use after a client is ingested and served (KNOWLEDGE-BRIEF generated) to produce a client-specific marketing landing that re-frames Dossier's universal messaging in the client's own vocabulary, with a hero artifact that is a REAL atom from the client's repo. Human-curated authoring, not a hosted runtime.

2026-06-181

author-okf

smithdak/dossier

Author or extract atomic, OKF-compliant concept documents that conform to Dossier's knowledge model. Use whenever turning raw/clean source content (or a topic) into knowledge-base concepts — processes, workflows, roles, decisions, systems, policies, artifacts, terms. Produces one well-formed markdown+YAML file per concept, with full provenance and typed relationships.

2026-06-141

github-markdown

smithdak/dossier

Wield the full power of GitHub-Flavored Markdown (GFM) and GitHub's rendering pipeline to design world-class documentation — alerts/admonitions, collapsible sections, tables, task lists, mermaid diagrams, math, footnotes, themed images, auto-anchors, and badges. Use whenever authoring a README, guide, or any markdown that renders on GitHub (and, with the noted caveats, on the Astro Starlight surface). The reference + procedure the documentation-engineer reaches for.

2026-06-141

name	ingest-client
description	Run the ingestion sweep that pulls a new client's raw knowledge into Dossier — public website (Firecrawl), files (Unstructured), and Microsoft 365/SharePoint (Graph delta + Copilot connectors) — normalized to clean, provenance-stamped markdown ready for extraction. Use when onboarding a client or refreshing their sources.

Ingest a Client

Procedure

Inventory sources with the client: public site(s), SharePoint/M365, file shares, Confluence/Drive/etc., and any tribal-knowledge docs. Record access and scope.
Web → Firecrawl: Map to discover URLs, crawl for exhaustive coverage → clean markdown. (WebFetch for one-off pages.)
Files → Unstructured: partition PDFs/Office/HTML/email/images into semantic markdown elements.
Microsoft 365 / SharePoint → Graph delta query for incremental sync (sites, driveItem, listItem) and/or Copilot connectors (ACL-aware). Remember: deletes/ACL changes need a full crawl; Notion needs a custom connector.
Normalize → consistent clean markdown; deduplicate near-identical content.
Stamp provenance on everything — source URI, fetch timestamp, and access/ACL notes — so every downstream OKF atom can carry source/confidence. Provenance is sovereignty.
Stage the normalized output in the client's silo, ready for author-okf.

Rules

Tenant isolation: a client's raw data stays in their silo — never commingled. (DEC: siloed multi-tenancy.)
Incremental where possible — use delta tokens/change feeds; don't re-ingest the world.
Idempotent, resumable, rate-limit-aware. Capture failures to traces.

Status / dependencies

Output

A staged, normalized, provenance-stamped corpus in the client's silo + a coverage report (what was ingested, what was skipped and why).