| name | lazyweb-explain-flow |
| route | Explain how something works with a hosted flow diagram (never touches the canonical chart) |
| router-terms | explain with a diagram, explain using the flowchart, walk me through how X works, diagram why this failed, failure trace, trace what happened, show how the pieces fit, visualize how it works, explain this flow |
| description | EXPLAIN how something works with a hosted, tappable flow diagram — a walkthrough,
a failure trace, an answer to "how does X work?", a hypothetical — WITHOUT
touching the product's canonical flow chart. Saves via lazyweb_explain_flow to
separate insert-only storage, hosted at /explainer/<id>/. Use THIS (never the
generate/save-flowchart path) whenever the goal is explanation rather than
recording the product's current state.
Trigger on: "explain this with a diagram", "walk me through how X works",
"diagram why this failed", "explain using the flowchart", "show me how the
pieces fit", "trace what happened".
|
| allowed-tools | ["Bash","Read","Grep","Glob","Agent"] |
Lazyweb — Explain a Flow
Host an explanatory diagram: same sectioned flow-view shape and viewer as the
canonical chart, but stored separately and insert-only — it can never overwrite
or outrank the product's canonical flow chart or its Architecture dashboard.
Hard rule: an explanation NEVER goes through lazyweb_save_flowchart. That
tool is the system of record (one chart per product, upserted in place, shown on
the user's dashboard); saving an explanation there overwrites or hijacks it.
Explanations go through lazyweb_explain_flow, full stop.
MCP Setup
Use the hosted Lazyweb MCP tools at https://www.lazyweb.com/mcp.
Required tools:
lazyweb_health — verify connectivity when the MCP surface is uncertain
lazyweb_explain_flow — save the explanatory diagram (returns explainer_id + explainer_url)
lazyweb_get_flowchart — optionally fetch the canonical chart as a starting reference
If Lazyweb MCP is missing — or lazyweb_explain_flow isn't in the tool list — tell
the user to run curl -fsSL https://www.lazyweb.com/install.sh | bash, reload, rerun.
Steps
- Understand what needs explaining. A question ("how does search work?"), a
failure ("why did this report ignore the intent?"), a walkthrough, a comparison.
Ground it in the real code/logs — read the handlers, don't invent.
- Optionally start from the canonical chart.
lazyweb_get_flowchart({ product })
gives you the product's real structure — copy and reshape it freely for the
explanation (highlight the relevant path, drop irrelevant sections, add the
failure step). You are editing a COPY; never re-save it through save_flowchart.
- Build the explanatory diagram in the same schema as the canonical chart
(sections, actors, nodes/edges with
notes and real data payloads,
bidirectional on round trips — the lazyweb-generate-flowchart conventions all
apply; an explanation earns its keep with REAL content, not boxes). Data
snippets meet the same exactness bar as the canonical chart: actual payloads
from the code/logs, … only to redact secrets (never to shorten content), and
a sliced array keeps its first item(s) full-shape with the omission count in
the note.
Shape it for the QUESTION: order sections as the narrative, name the section
titles as claims (e.g. "Where the run died — no model call"), put the "so what"
in each node's note.
- Save it:
lazyweb_explain_flow({ title, summary, product?, diagram }) →
share the returned explainer_url. The canonical chart is untouched.
When NOT to use
- Recording the product's actual current architecture →
/lazyweb-generate-flowchart
(canonical, matter-of-fact, one per product).
- Refreshing the canonical chart after code changes →
/lazyweb-update-flowchart.
- Proposing changes the user should Accept/Decline →
/lazyweb-propose-ui-changes.
Notes
- Explainers are point-in-time and cheap — make as many as the conversation needs;
they never collide with each other or with anything canonical.
- If an explanation reveals the canonical chart is WRONG about the current state,
say so and offer
/lazyweb-update-flowchart as a separate follow-up — don't fold
the correction into the explainer.