feat: Scope local-clone scanners to the core-repos allowlist

[rubambiza]

Wire the link-health and dep-bump discovery loops to the shared core-repos allowlist introduced in #36, so they act only on the curated core repos instead of every locally cloned directory.

What changed

• program-lib.sh: add is_core_repo() (exact whole-line membership test). Document canonical_repo_for_dir() as a transitional shim (see chore: Rename clone dirs to canonical names and retire the remap shim #37).
• Each clone-iteration loop now: maps the dir basename to its canonical repo, skips non-core/archived repos (allowlist), and dedups duplicate clone dirs (e.g. a stale kagenti/ dir alongside rossoctl/) so each canonical repo is processed once. API refs are built as rossoctl/<canonical>.
• dep-bump-scanner: fixes the false "0 open Dependabot PRs" report — gh pr list --author silently returns empty across a rename redirect, so querying the canonical repo restores results (e.g. rossoctl/rossoctl now returns 11 vs 0).
• extract-broken-links.sh: arg 2 is now a full owner/name ref emitted verbatim (was a bare name prefixed with kagenti/); test updated.

Scope

Discovery / read paths only. Fixer write paths (ensure_fork / create_fork_pr / $ORG in PR creation) are intentionally left for Phase 6, where they are resolved alongside the fork-naming decision.

Verification

• Verified against the real clone set: 8 core repos scanned (from 25), duplicate clones deduped, archived/non-core excluded.
• Full test suite passes on bash 3.2 (test-core-repos, test-extract-broken-links).

Part of #29 (epic #32) — completes the Phase 4 (local-clone scanners) portion of the centralization work; #36 landed Phases 2–3.

Fixes #29

Assisted-By: Claude Code

[clawgenti]

Solid refactor — the canonical-remap + allowlist filter + dedup pattern is applied consistently across all five scripts, and the is_core_repo helper uses grep -Fx to avoid substring false positives. Tests updated to match the new owner/name arg contract.

Finding:

• dep-bump-fixer.sh line 96 initialises seen_canon="" (lowercase) while the outer-scope loops in the same file use SEEN_CANON / SEEN_CANON_TTM (uppercase). No functional impact here since the variable stays local to the if block, but the inconsistency is worth normalising.

---

Reviewed by clawgenti using github:pr-review

[clawgenti]

Naming inconsistency: seen_canon is lowercase here, but the equivalent variables in the two outer-scope loops below are SEEN_CANON and SEEN_CANON_TTM (uppercase). No functional issue since this variable is scoped to the if block, but consider renaming to SEEN_CANON for consistency with the rest of the file.

[cwiklik]

Scopes the link-health and dep-bump discovery loops to the core-repos allowlist and fixes the rename-redirect bug — which is the exact class of problem behind operator #498 (bots acting on stale/duplicate clone dirs and pre-rename names).

Verified:

• is_core_repo() uses grep -qxF (exact whole-line, fixed-string) so there are no substring false positives; the test explicitly guards rosso ≠ rossoctl.
• The allowlist + canonical_repo_for_dir remap + SEEN_CANON dedup block is consistent across all five scanners, SEEN_CANON is reset before each loop (no cross-loop leakage), and the " $SEEN_CANON " / *" $canon "* padding is the correct whole-word membership idiom.
• dep-bump-scanner's Step 1 stores the canonical name and Step 2 prefixes rossoctl/ — source parity holds across the handoff.
• The gh pr list --repo rossoctl/<canon> --author fix matches the reported '0 vs 11' symptom (the author-filtered list doesn't follow a rename redirect).
• extract-broken-links drops the hardcoded kagenti/ prefix for a verbatim owner/name (a latent rename bug removed); no lingering $REPO_NAME; tests updated and assertive.
• canonical_repo_for_dir is clearly documented as transitional with a deprecation note and issue ref.

One non-blocking suggestion inline about factoring the duplicated filter block. DCO green, both commits signed, no .claude/.vscode. LGTM — nice structural fix.

Assisted-By: Claude Code

[cwiklik]

suggestion (non-blocking): the allowlist + canonical-remap + dedup block is now copy-pasted ~8 times across the five scanners (3× in dep-bump-fixer alone), which is real drift risk for security-sensitive automation. Consider a small program-lib helper — e.g. canon=$(repo_filter "$dir") || continue that performs the canonical_repo_for_dir remap + is_core_repo gate and echoes the canonical name, leaving only the SEEN_CANON dedup inline — so the allowlist policy lives in exactly one place.

[rubambiza]

Great suggestion. This reinforces what I had already noticed. That is, program-lib.sh was growing too long for easy maintainability, and I have an open issue as part of the epic to refactor it into a smaller set of subprograms like you suggested. I will add this suggestion to the context for the planning.

[rubambiza]

See Design Input section in #35