Repository navigation
v0.3.30
·
5 commits
to main
since this release
Immutable
release. Only release title and notes can be modified.
- fix a real, previously-unnoticed ADR-immutability violation: round 5's own commit edited
docs/architecture/adr/ADR-0001-adopt-gitflow-for-release-governance.md's accepted Decision/Consequences text in place, directly contradictingdocs/information-item-map.md's own stated Decision Record convention ("a newADR-NNNNfile is added, never edited after acceptance"). Reverted the original accepted text to be byte-identical to its first acceptance commit, and moved round 5's genuinely useful content (thev0.3.28GitFlow-exception disclosure, the stamp-condition clarification) into a new, clearly-marked, append-only "Addendum" section instead - fix 2 remaining dangling
docs/projectbuddy-consumption.mdreferences round 5 missed when it fixed 2 sibling instances of the identical gap:Run-Pipeline.ps1,Run-Pipeline.sh,Run-WindowsSbom.ps1, andRun-LinuxSbom.ps1(4 files, all repointed to README.md's real "Consumption model" section) - fix
.github/skills/projectbuddy-identity/(SKILL.md,agents/openai.yaml): both still asserted the unqualified "ships with zero pre-committed... state" claim afterREADME.mdwas already corrected in round 5 to honestly disclose the real, tracked violations (#68) -- the same class of sibling-location drift this session keeps finding in this specific file pair - fix
.github/workflows/standards-compliance-recheck.yml's header comment: still said public upstream "deliberately carries zero private infrastructure state" afterdocs/architecture/standards-compliance.mdwas corrected for the identical overclaim in round 5 (a real, onlinevm-provisioner-labeled runner is in fact registered against it) - fix
docs/sbom-pipeline/procedure.md: still claimed both containerized pipelines validate againstpipeline-config.schema.jsonuniformly, contradicting the wiki Architecture Packet's own already-corrected disclosure that the Linux container does its own separate, narrower inline validation instead - fix
docs/vm-provisioning/procedure.md: (1) the receipt-verification-chain paragraph claimed apackage-buildreceipt verifier exists; it doesn't (package-vagrant-box.pyvalidates its input vagrant-box receipt but produces no independently-reverifiable receipt of its own output); (2) the Shared Schemas section implied both VM families'provisioning-manifest.jsonpin the same binary-hash categories, when Ubuntu's manifest has no Actions-runner/PowerShell/NSSM equivalents (Windows-specific concepts) -- clarified both are family-specific schemas, not one shared shape - fix
docs/release/release-procedure.md's Hotfix Flow: added an explicit disclosure that this flow has never actually been exercised in this repo's real history (zerohotfix/*branches or PRs, ever), matching the same honesty standard already applied elsewhere in this document - fix the wiki SRS's stale Document Control (
Version: Initial draft... as of v0.3.10,Status: Draft) and 2 references to av0.3.11tag that was never actually tagged or released (confirmed: tags jump fromv0.3.10tov0.3.13) -- both now cite the real, verifiable commit (c3a36ae) that resolved the spec-to-CI wiring question - fix the wiki Architecture Packet's Deployment View: round 5's own fix cited one specific, ephemeral historical workflow run ID as its sole evidence for
validate-vm-provisioning-dispatch.py's restoration -- directly contradicting this same SRS's own stated evidence-citation principle ("never at one specific historical workflow run ID -- a run ID is ephemeral and does not, by itself, prove anything about the current state of the code"). Reframed to cite the restored script's own reproducible code/logic as the primary evidence, with the historical dispatch noted only as a same-day live confirmation, not the proof itself - fix the wiki Architecture Packet: disclose that
vm-fleet-status.yml's scheduled runs fail differently on the fork (svelderrainruiz/projectbuddy, which has novm-provisioner-labeled runner at all, so every scheduled run queues and is cancelled) versus on public upstream (which now cleanly no-ops per round 5's fix) -- same root cause as the already-documentedstandards-review-localpattern, just not previously disclosed for this specific workflow - fix REQ-010 (wiki SRS) and REQ-013 (wiki RTM): round 5 updated one of each requirement's two sibling locations (SRS or RTM) but not both, leaving the un-updated sibling stale -- REQ-010's SRS row now reflects the same
v0.3.28exception already in its RTM row; REQ-013's RTM row now cites the actually-invokedbuild-virtualbox.shinstead of the orphaned, never-called.ps1sibling, matching its own SRS row - added 2 new, previously-uncited
#68(REQ-001 hardcoded-identifier) violation sites as evidence:vm/windows-labview/unattended-template.xml(a Windows-family instance, 4 hardcoded<Description>fields -- all prior evidence in that issue was Ubuntu-family),vm/ubuntu-labview/provision-guest.sh, and a broader, distinct"sandbox-sbom-reporting/..."schema-namespace-identifier pattern spanning 11 files (flagged with an important nuance: these are versioned, already-publicly-documented schema identifiers, not secrets -- likely an unfortunate namespace-naming choice rather than the same class of bug as the other sites, but still worth a deliberate decision) - deleted 1 more already-merged stale branch on the fork
- filed
#80(remaining lower-severity findings: more Architecture Packet Component/Container View gaps, Context View staleness aroundfeature/*branch naming and consumer SHA-pinning precision, CM/release-procedure practice-vs-document gaps including the wiki's own lack of tag/commit-SHA binding to main-repo releases, and further ISO 29119-3/15289 formal-completeness instances compounding the already-open#77) - all findings from this round's
--mode semantic-auditpass (same 3-model delegation, explicitly asked to verify whetherv0.3.29's fixes held up and hunt for anything neither prior round caught): significantly fewer new defects than prior rounds (expected, 6 rounds deep into hardening the same spec/doc surface), but still real, previously-uncaught sibling-location drift and one genuine process violation (the ADR in-place edit) -- GPT found no new severe issues beyond already-tracked baselines; Gemini and Claude each independently caught several real, previously-missed sibling-drift defects, with meaningful overlap between the two (both independently found the ADR-0001 edit and the standards-compliance-recheck.yml staleness)