You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
README.md ("Running it across many repos") and CLAUDE.md ("What belongs here vs. in gardener") both link https://github.com/dmccoystephenson/gardener and describe it as "a separate open-source project" /
"this project's open-source companion".
That repo is currently private, so the link 404s for every reader except the account owner:
PR #78 ("Remove references to repos that aren't public") existed specifically to make this repo safe to
publish, and its test plan states that no reference to a non-public repo remains. The gardener cross-link
was added by that same PR, so this one slipped through the check.
It is the only remaining unfollowable reference I found in a full sweep of the repo's docs.
Expected vs. actual
Expected: every repo link in the public docs resolves for an anonymous reader.
Actual: two docs link a private repo and call it open source.
Fix requires a maintainer decision
Two valid resolutions, and picking between them is not something a docs sweep should decide:
Make dmccoystephenson/gardener public. The docs then become correct as written, and this closes
with no code change. This looks like the intended end state given PR Remove references to repos that aren't public #78's framing.
Soften the docs. Drop the hyperlink and the "open-source" wording, describing the split
(create-dev-loop decides what a skill knows; a dispatcher decides when it runs) without naming an
inaccessible repo — the same treatment PR Remove references to repos that aren't public #78 gave the anonymized RESEARCH.md citations.
Note that CLAUDE.md is one of the two files needing the edit under option 2, which the dev-loop treats as
agent-loaded config requiring explicit human authorization to modify.
Notes
scripts/check_docs.py only validates relative links (check_local_links skips anything starting with http:///https://), so CI cannot catch this class of drift today.
Found during a Stage A documentation-accuracy sweep on 2026-07-29.
What went wrong
README.md("Running it across many repos") andCLAUDE.md("What belongs here vs. in gardener") both linkhttps://github.com/dmccoystephenson/gardenerand describe it as "a separate open-source project" /"this project's open-source companion".
That repo is currently private, so the link 404s for every reader except the account owner:
Why this matters
PR #78 ("Remove references to repos that aren't public") existed specifically to make this repo safe to
publish, and its test plan states that no reference to a non-public repo remains. The gardener cross-link
was added by that same PR, so this one slipped through the check.
It is the only remaining unfollowable reference I found in a full sweep of the repo's docs.
Expected vs. actual
Expected: every repo link in the public docs resolves for an anonymous reader.
Actual: two docs link a private repo and call it open source.
Fix requires a maintainer decision
Two valid resolutions, and picking between them is not something a docs sweep should decide:
dmccoystephenson/gardenerpublic. The docs then become correct as written, and this closeswith no code change. This looks like the intended end state given PR Remove references to repos that aren't public #78's framing.
(create-dev-loop decides what a skill knows; a dispatcher decides when it runs) without naming an
inaccessible repo — the same treatment PR Remove references to repos that aren't public #78 gave the anonymized
RESEARCH.mdcitations.Note that
CLAUDE.mdis one of the two files needing the edit under option 2, which the dev-loop treats asagent-loaded config requiring explicit human authorization to modify.
Notes
scripts/check_docs.pyonly validates relative links (check_local_linksskips anything starting withhttp:///https://), so CI cannot catch this class of drift today.