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.
- Classify whether the user needs source discovery, processing of an already
selected source, or both.
- 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.
- 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.
- State the intended
analysis_target and resolve the required acquisition
operation for the selected source.
- Run preflight for live URLs, media, or unclear expectations.
- Route URL/query acquisition to
acquire-source-material; local files use the
same Bundle v2 contract through the local builder.
- 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.
- Route the promoted manifest to
source-gated-evidence-layer.
- 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.
- Treat
knowledge-video-decomposer as an internal script library, never as
a competing user-facing route.
- Use a new project root by default. Reuse requires
--resume and an exact
source, target, and operation match.
- 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:
kw run --input <url-or-file> --target <target> --operation <operation> --mode audit
For a learning article:
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:
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.
1---2name: knowledge-workflow-console3description: 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.4---56# Knowledge Workflow Console78Use this skill as the product controller.910Before the first command, resolve the installed entry point with `kw version`.11If `kw` is unavailable, install the repository package; do not assume the12current working directory contains `kw.py`.13141. Classify whether the user needs source discovery, processing of an already15 selected source, or both.162. For a broad open-web learning need, comparison, recommendation, or17 source-selection request, route discovery to `web-intent-scout`. Ask it for18 an intent map, source ledger, candidate shortlist, and selection rationale.19 Skip discovery when the user already supplied the source and did not ask to20 compare, verify, or find alternatives.213. Treat Web Scout outputs as planning artifacts only. Select a URL or explicit22 query for acquisition; never promote snippets, scorecards, or dossier claims23 directly into source evidence.244. State the intended `analysis_target` and resolve the required acquisition25 `operation` for the selected source.265. Run preflight for live URLs, media, or unclear expectations.276. Route URL/query acquisition to `acquire-source-material`; local files use the28 same Bundle v2 contract through the local builder.297. When no structured adapter exists, obtain authorized task-primary material30 through a browser, CLI, API, or user export, then route the saved artifact31 through `kw source import`; never downgrade it to a generic web fallback.328. Route the promoted manifest to `source-gated-evidence-layer`.339. For a source-faithful report, continue to `knowledge-document-composer` only34 when the current gate and analysis receipts allow it. For personal learning,35 run the console with `--deliverable learning_article`. Read the generated36 `15_learning/learning_enrichment_request.json`, return to its gate-admitted37 normalized source, write evidence-bound `learning_enrichment.json`, and then38 call `kw learn`. Require the learning quality gate and receipt before39 delivery. Never generate Source rows from the heuristic inventory alone.4010. Treat `knowledge-video-decomposer` as an internal script library, never as41 a competing user-facing route.4211. Use a new project root by default. Reuse requires `--resume` and an exact43 source, target, and operation match.4412. Finish with provenance-aware status and result index. Report discovery45 artifacts, source-faithful reports, and learning articles separately, and46 report stale output files separately from current deliverables.4748Browser state may supply authorized visible artifacts, but it does not bypass49Bundle v2 or the source gate. Whenever Chrome, Edge, OpenCLI, cookies, an50extension, or a browser export is involved, require the actual host to be51declared; never infer it from a tool name or fall back to the other browser.52Metadata, snippets, screenshots, comments, and captions must not be silently53promoted to another content scope.5455Primary command:5657```powershell58kw run --input <url-or-file> --target <target> --operation <operation> --mode audit59```6061For a learning article:6263```powershell64kw run --input <url-or-file> --target <target> --operation <operation> --mode audit --deliverable learning_article65# The Agent reads learning_enrichment_request.json and the admitted source,66# then writes 15_learning/learning_enrichment.json.67kw learn --project-root <project> --depth standard68```6970The pause between these commands is intentional. Semantic enrichment is an71Agent judgment step; the CLI prepares and verifies it but must not replace it72with deterministic heuristic prose.7374Use explicit stages when diagnosing:7576```powershell77kw acquire ...78kw ingest ...79kw audit ...80kw compose ...81kw learn ...82kw status --project-root <project>83kw result --project-root <project>84```8586Read `references/routing.md`, `references/stage-contracts.md`, and87`references/output-layout.md` before changing routing or handoff behavior.
sitabanubanu