Cross-repo unification: conventions pointer, remote form, and shared Jev tooling #1
Loading…
Reference in a new issue
No description provided.
Delete branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
websourcesis the sixth Clawdie repo and is currently referenced by none of thefive that share conventions (canonical table:
colibri/AGENTS.md; siblings:clawdie-iso, colibri, bherdr, zot, agent-skills). Tracking convergence items
found while unifying the docs conventions. Read-only analysis; no edits made.
Deterministic findings
colibri scripts/cross-repo-consistency.pyover thisAGENTS.md:[REMOTE-FORM]§4 uses the SCP formgit@code.clawdie.si:clawdie/websources.git;canonical is
ssh://git@code.clawdie.si:2222/clawdie/websources.git.[GATES-PARTIAL]declares theokffrontmatter gate but not the shareddates (
DD.mmm.YYYY) / markdown formatter / wiki-lint gates.OMISSION-TABLEhit was a scanner false positive on the §3 boundarymatrix (repos named in a topics cell, not a coordination table). Fixed in
colibri #95; no table change is needed here.
Convention gaps
authoritative" clause.
hello@clawdie.si,Sam & <Agent> (<host>)), nor the branch/PR flow. §4's"never push without explicit human instruction" is stricter than the
canonical branch workflow and should win locally.
okfschema drift: §6 requiresclassified_by/status; the colibri gaterequires
verified_by/platform_notes. This file itself usesverified_by.Jev tooling drift (highest value)
scripts/jev-classify.pyre-implements the Jev transport separately from theshared
agent-skills/jev-triagetooling:typesafe-ai/jev; no TypeSafe-primary /OpenRouter fallback;
caller_attribution,provider, orprovider_attempts(agent-skills #37/#41);
Its own
content-intake/UMBRELLA-WIKI-GOAL.mdalready requires cache identity =source hash + excerpt/section + model + question schema + policy version, and
"preserve raw confidence and per-choice probability fields separately" — the
script does not implement this. Stamp ownership is tracked in colibri #94.
Docs ownership overlap
sites/docsandsites/wikiclaim canonical ownership of Colibri/bherdr/zot/ZFSplatform documentation, while the component repos keep their own
docs/wiki/(e.g. colibri
docs/wiki/architecture.md). Decide source-of-truth and linkrather than duplicate.
Decision needed
Is
websourcesa platform repo (named in every coordination table — thecanonical four become five) or a satellite (linked to the canonical table,
discovered by the scanner as
EXTRA_REPOS, like agent-skills)?Recommendation: satellite — it consumes the platform docs rather than
owning the code platform. Once decided: add it to the colibri coordination
table + scanner, then open a
websourcesPR for the pointer and remote form.Verified — closing.
#20 is merged (
51ace40). Most of this issue was already resolved before the PR;here is the itemised disposition.
Already resolved on
main(verified, no change needed)[REMOTE-FORM]— §4 already uses the canonicalssh://git@code.clawdie.si:2222/clawdie/websources.git.scripts/jev-classify.pyloads the sharedjev_eval(TypeSafe-first with fallback) via
JEV_TRIAGE_ROOT/PATH, and recordscaller_attribution,provider,provider_attempts,usage,cost. Thehardcoded Vercel transport is gone.
DOCUMENTATION_REPOS = ("websources",), i.e. the recommended satelliterole, and §1 already requires linking to component sources rather than
duplicating their contracts.
python3 colibri/scripts/cross-repo-consistency.py→ 0 findings across 6repos (re-run just now against the merged
main).Fixed in #20
classified_by/status; now namessource_of_truth/last_verified(30-day window) /verified_by, plusplatform_noteswhere the source touches platform-specific code.push without explicit instruction rule explicitly stricter than the
shared
main-is-merge-only workflow.build produces 98 (agents 57, wiki 10). Corrected and reworded as a smoke
signal rather than a fixed contract.
Verified:
npm run check:all✅ ·npm run build✅ (98 pages, 5 workspaces).Noted, not actioned
The
sites/docs↔ component-repodocs/wiki/overlap is left to §1's existing"link, don't duplicate" rule rather than a new §3 decision. Reopen or file
separately if you want that stated explicitly.