llms-full.txt contains 7 links to https://conduit.io/..., a domain that no longer resolves:
$ curl -sS --max-time 12 https://conduit.io/
curl: (7) Failed to connect to conduit.io port 443 after 161 ms: Couldn't connect to server
Three distinct dead URLs, at lines 118, 259, 323, 353, 373, 505 and 612 of llms-full.txt on
v0.19.0:
https://conduit.io/docs/using/connectors/configuration-parameters/output-format (×5)https://conduit.io/docs/connectors/output-formatshttps://conduit.io/docs/core-concepts#standalone-connector
The live docs are at conduitdata.io, so these should be
https://conduitdata.io/docs/using/connectors/configuration-parameters/output-format and so on.
llms.txt is clean; only llms-full.txt is affected.
Why it matters more now
ConduitIO/conduit-site#362 serves these files at https://conduitdata.io/llms.txt and
/llms-full.txt so agents can retrieve them without a checkout — which is the point of the
llms.txt convention. That makes llms-full.txt the canonical agent-facing description of Conduit,
and an agent that follows one of these links for the output-format reference gets a connection
failure.
Where they come from
Not from llmsgen itself — it inlines connector specifications verbatim. The sdk.record.format
description carrying the ×5 link comes from conduit-connector-sdk; the other two come from
connector READMEs/specs. So the fix is upstream of the generator, in whichever source declares
those descriptions, and then a regeneration.
Worth considering alongside the fix: a check that fails when a generated file contains a link to a
domain we no longer serve. The generator is already drift-guarded in CI; a dead-domain assertion on
its output would be cheap and would have caught this.
llms-full.txtcontains 7 links tohttps://conduit.io/..., a domain that no longer resolves:Three distinct dead URLs, at lines 118, 259, 323, 353, 373, 505 and 612 of
llms-full.txtonv0.19.0:https://conduit.io/docs/using/connectors/configuration-parameters/output-format(×5)https://conduit.io/docs/connectors/output-formatshttps://conduit.io/docs/core-concepts#standalone-connectorThe live docs are at
conduitdata.io, so these should behttps://conduitdata.io/docs/using/connectors/configuration-parameters/output-formatand so on.llms.txtis clean; onlyllms-full.txtis affected.Why it matters more now
ConduitIO/conduit-site#362 serves these files at
https://conduitdata.io/llms.txtand/llms-full.txtso agents can retrieve them without a checkout — which is the point of thellms.txt convention. That makes
llms-full.txtthe canonical agent-facing description of Conduit,and an agent that follows one of these links for the output-format reference gets a connection
failure.
Where they come from
Not from
llmsgenitself — it inlines connector specifications verbatim. Thesdk.record.formatdescription carrying the ×5 link comes from
conduit-connector-sdk; the other two come fromconnector READMEs/specs. So the fix is upstream of the generator, in whichever source declares
those descriptions, and then a regeneration.
Worth considering alongside the fix: a check that fails when a generated file contains a link to a
domain we no longer serve. The generator is already drift-guarded in CI; a dead-domain assertion on
its output would be cheap and would have caught this.