v0.6.14

Three propagation-pipeline polish fixes surfaced during the v0.6.13 cycle (2026-06-02). All three address pain points that required manual intervention during v0.6.13 propagation and that the workflow should handle structurally going forward.

Fixed — Doctor content-diff for installed hooks (issue #112 addendum)

logmind doctor now content-diffs installed .git/hooks/post-merge (and post-rewrite) bytes against the bundled canonical body, in addition to the v0.6.10 version-marker equality check. Closes the failure mode the tokenomics reporter flagged in the #112 addendum: "doctor's check is too cheap a signal" — if the hook drifts after install while preserving the marker line (manual edit, clobber-then-restore by another tool, marker-propagation bug), the v0.6.10 check would still report current while the body is genuinely stale.

v0.6.14: marker equality stays as the cheap fast-path; the content-diff is the expensive correctness check that fires when it matters. Drift surfaces as STALE (content drift) with the existing "reinstall via logmind self-update" remediation pointer.

Files: src/logmind/core/doctor.py (_probe_post_merge_hook + _probe_post_rewrite_hook); tests/test_merge_driver.py (new content-drift and byte-identical tests).

Fixed — logmind-self-update.yml v7: [skip-logmind] PR title prefix

Bot-generated propagation PRs from logmind-self-update.yml are by-definition NOT decision commits — they're pin-bumps + template body refreshes owned by the workflow itself. v0.6.13 propagation surfaced this as a consistent failure: consumer-repo check-decisions.yml failed every PR because the diff exceeded the 20-line threshold AND no decision file was present. The fix required manually editing each PR title to include [skip-logmind] (the existing maintainer-override path) + closing and reopening the PR to re-fire the title-cached check-decisions workflow.

v0.6.14: the self-update template now sets [skip-logmind] logmind self-update: X.Y.Z → A.B.C as both the commit message AND the gh pr create --title at workflow creation time. Eliminates the manual editing + close+reopen ritual entirely.

Files: src/logmind/templates/github/logmind-self-update.yml.template (bumped to v7); tests/test_v0_2_1_audit_fixes.py (new test + marker bump).

Fixed — logmind-self-update.yml v7: PyPI CDN-aware retry

The version-check step polls the PyPI simple/ index OR JSON API to find the latest published logmind. Both propagate via a CDN that lags 30-60s behind a publish event. Cron-triggered self-update workflows can fire during the gap, see "Latest: ", and silently no-op. Hit on v0.6.12 (1/5 propagations) and v0.6.13 (1/5 propagations).

v0.6.14: wraps the version-check in a 3-attempt exponential backoff (0s → 30s → 60s). If all 3 see the same version, accept and proceed as today (legitimate no-op). If any sees a higher version, use it. Adds 0-90s to worst-case workflow runtime; eliminates the silent-skip class of bug.

Files: same template; same test file.

Validation gate end-to-end

After v0.6.14 propagates through all 5 thrillmade consumer repos, the next release cycle (v0.6.15+) should:

• 5/5 consumers' self-update workflows pin-bump-only releases succeed first attempt with NO manual title editing, NO close+reopen rituals, NO CDN-race silent-skips
• doctor on any repo with a hand-edited or drifted hook body reports STALE (content drift) with the logmind self-update remediation pointer