# Knowledge Workflow Console

> Start here for end-to-end source-gated knowledge work from broad learning needs, URLs, queries, transcripts, subtitles, media, web articles, social posts, and repositories. Route optional web discovery through web-intent-scout, then acquisition and evidence, and finally a learning article or source-faithful document; report provenance-aware status and do not acquire or write claims directly.

- Skill: `sitabanubanu/knowledge-workflow-console` (Agent Skill, multi-file: 10 files)
- Install (CLI): `npx skillmds@latest add sitabanubanu/knowledge-workflow-console`
- Raw SKILL.md: https://api.skillmd.com/api/skills/sitabanubanu/knowledge-workflow-console/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Product & Planning
- Author: sitabanubanu (https://skillmd.com/u/sitabanubanu)
- Updated: 2026-09-22
- Page: https://skillmd.com/skills/sitabanubanu/knowledge-workflow-console

---


# Knowledge Workflow Console

Use this skill as the product controller.

Before the first command, resolve the installed entry point with `kw version`.
If `kw` is unavailable, install the repository package; do not assume the
current working directory contains `kw.py`.

1. Classify whether the user needs source discovery, processing of an already
   selected source, or both.
2. For a broad open-web learning need, comparison, recommendation, or
   source-selection request, route discovery to `web-intent-scout`. Ask it for
   an intent map, source ledger, candidate shortlist, and selection rationale.
   Skip discovery when the user already supplied the source and did not ask to
   compare, verify, or find alternatives.
3. Treat Web Scout outputs as planning artifacts only. Select a URL or explicit
   query for acquisition; never promote snippets, scorecards, or dossier claims
   directly into source evidence.
4. State the intended `analysis_target` and resolve the required acquisition
   `operation` for the selected source.
5. Run preflight for live URLs, media, or unclear expectations.
6. Route URL/query acquisition to `acquire-source-material`; local files use the
   same Bundle v2 contract through the local builder.
7. When no structured adapter exists, obtain authorized task-primary material
   through a browser, CLI, API, or user export, then route the saved artifact
   through `kw source import`; never downgrade it to a generic web fallback.
8. Route the promoted manifest to `source-gated-evidence-layer`.
9. For a source-faithful report, continue to `knowledge-document-composer` only
   when the current gate and analysis receipts allow it. For personal learning,
   run the console with `--deliverable learning_article`. Read the generated
   `15_learning/learning_enrichment_request.json`, return to its gate-admitted
   normalized source, write evidence-bound `learning_enrichment.json`, and then
   call `kw learn`. Require the learning quality gate and receipt before
   delivery. Never generate Source rows from the heuristic inventory alone.
10. Treat `knowledge-video-decomposer` as an internal script library, never as
   a competing user-facing route.
11. Use a new project root by default. Reuse requires `--resume` and an exact
   source, target, and operation match.
12. Finish with provenance-aware status and result index. Report discovery
    artifacts, source-faithful reports, and learning articles separately, and
    report stale output files separately from current deliverables.

Browser state may supply authorized visible artifacts, but it does not bypass
Bundle v2 or the source gate. Whenever Chrome, Edge, OpenCLI, cookies, an
extension, or a browser export is involved, require the actual host to be
declared; never infer it from a tool name or fall back to the other browser.
Metadata, snippets, screenshots, comments, and captions must not be silently
promoted to another content scope.

Primary command:

```powershell
kw run --input <url-or-file> --target <target> --operation <operation> --mode audit
```

For a learning article:

```powershell
kw run --input <url-or-file> --target <target> --operation <operation> --mode audit --deliverable learning_article
# The Agent reads learning_enrichment_request.json and the admitted source,
# then writes 15_learning/learning_enrichment.json.
kw learn --project-root <project> --depth standard
```

The pause between these commands is intentional. Semantic enrichment is an
Agent judgment step; the CLI prepares and verifies it but must not replace it
with deterministic heuristic prose.

Use explicit stages when diagnosing:

```powershell
kw acquire ...
kw ingest ...
kw audit ...
kw compose ...
kw learn ...
kw status --project-root <project>
kw result --project-root <project>
```

Read `references/routing.md`, `references/stage-contracts.md`, and
`references/output-layout.md` before changing routing or handoff behavior.