aftermath Class-1 post-batch residual verification (v0.14.55) - #1067
Merged
Ljferrer merged 9 commits intoJul 24, 2026
Conversation
… (1 round) Patched End state 6 + Task 1.1's stale-claim sweep to name the landed 2026-07-16 plan (docs/plans/2026-07-16-aftermath-class1-gate-evidence.md) as deliberately-left-uncorrected — a landed plan is a point-in-time record, never retro-edited (same rule the plan already applied to the dated spec Context band). Resolves the one needsDecision; deterministically re-verified that every `grep -rn 'not encoded'` hit is now adjudicated. Executed proof: the plan's own backstop #1 (throwaway file:// repo git-semantics demo) run as a probe — PASS: the buggy while-read exclusion filter silently loses the hold ref, the two-sided ls-remote diff catches it, the snapshot-SHA restore recovers it. All 6 spine + 3 bespoke probes on-target; escape guard clean. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
… pins + lesson amendment (#1064) Coupled Task 1.1 — all three files in ONE commit (spec criterion 5: the commit touching the lesson IS the commit touching the SKILL; the drift guard travels with the prose it pins). - skills/aftermath/SKILL.md: new body-only subsection '### Class-1 remote deletes — post-batch residual verification', placed between the stranded-upstream `-d` refusal subsection and Class-4 join rule, carrying six one-line-paragraph items: (a) retained (never retaken) ls-remote snapshot + exact-`refs/heads/`-name complement hold set; (b) set-difference construction advisory naming the piped `while read` subshell silent-no-op; (c) two-sided post-batch `git ls-remote --heads` diff (hold-set survival = invariant, missing = data-loss row, SHA drift informational; delete-list survivor = failed-delete row, never re-batched); (d) clean-verdict rule under any flag incl. --afk; (e) snapshot-SHA restore `git push origin <snapshot-sha>:refs/heads/<ref>`, interactive-confirm-only / --afk print-only; (f) zero-local-refs hygiene note. Gate cell, C3 paragraph, four-class table, known-stranded.tsv, frontmatter byte-untouched. - skills/war-machine/war-pipeline-structure.test.sh: two mandated has_i() prose pins ('hold set must survive', 'before declaring the run clean') + one optional has() pin (':refs/heads/' restore refspec, worker latitude) appended inside the existing Class-1 gate-evidence block, after its two row-scoped checks, before the summary printf; sub-comment cites the 2026-07-22 spec. The block banner and its four pre-existing assertions stay byte-untouched. - docs/learnings/aftermath-remote-stranded-differs-from-local-tip-reachability.md: the 2026-07-16 dated line's 'not encoded anywhere / standing warning' clause amended in place to record the 2026-07-22 encoding, citing the spec path; one line stays one line; metadata.keywords += 'residual-set hold-set'; description frontmatter and third body paragraph byte-untouched. war-memory lint clean. Anchors: each greps to exactly one region of SKILL.md (the new subsection), zero-hit at the 2026-07-22 base; `ls-remote` banned (6× in the SKILL). Red-proof (temp-break — subsection reverted, three new pins go red): RED(absent): has_i 'hold set must survive' RED(absent): has_i 'before declaring the run clean' RED(absent): has ':refs/heads/' Subsection present -> all three GREEN; the block's four pre-existing assertions (git cherry, --unset-upstream, both row-scoped gate-cell pins) stay green byte-unmodified across the revert. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Add the residual-set verification term to the Class-1 glossary cluster, anchored immediately after the stranded upstream entry (acknowledged-stranded -> patch-equivalence probe -> stranded upstream -> residual-set verification). Documents the post-batch git ls-remote two-sided diff against the pre-batch snapshot's hold set: a missing hold-set ref is a data-loss row (snapshot SHA + restore command), a surviving delete-list ref is a failed-delete row, and the run is not clean until the diff is empty or fully reported. References SKILL prose landing in Task 1.1 (same phase) by named construct only. Task 1.2 of docs/plans/2026-07-22-aftermath-class1-postdelete-verify.md.
…-1 into dev/2026-07-22-aftermath-class1-postdelete-verify
…ktree-lag recurrence 17 Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
…fication (#1066) Bump all four version slots in lock-step to the next free patch above the live integration base: plugin.json `version`, marketplace.json `metadata.version` and `plugins[0].version`, and the README `## Status` line (replace-in-place, no badge). All four slots read 0.14.54 at the phase base — the global max across every local/remote ref and tag — so 0.14.55 is the next free patch, resolved from the four slots as they stand in the rebased tree (version-slots.test.mjs is the arbiter; plan v-literals are non-authoritative). Status blurb (additive/precise; Class-1-remote-batch-scoped doctrine the Lead follows by reading the skill — not a mechanized floor, gate, or test): after the batched `git push origin --delete`, /aftermath re-lists the remote and diffs survivors two ways against the retained pre-batch snapshot — a missing hold-set ref is a data-loss row with snapshot SHA + restore command (interactive-confirm-only / --afk print-only), a surviving delete-list ref is a failed-delete row; the run withholds its clean verdict until the diff is empty or every discrepancy is already a report row. ADR 0027's C3 deletion bar is byte-unchanged (verification adds evidence, no new deletion path, no autonomous ref-creation under --afk). Prior release prose left as-is. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
…thin one plan) Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
cae15d3 (docs(learnings): phase 2) inadvertently committed stale 0.14.54 version files from a lagging verify worktree over the 0.14.55 release set by the phase-2 merge 1edd92d — reverting plugin.json, marketplace.json (metadata + plugins[0]), and the README ## Status blurb back to plan 6's release. version-slots.test.mjs did not catch it: all four slots were reverted in lock-step, so the lock-step invariant still held. Restores the four slots + aftermath Class-1 blurb to their 1edd92d (0.14.55) state. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Campaign plan 7 of 9 —
2026-07-22-run-resilience-and-hardening. Stacked (ADR 0011) ondev/2026-07-22-war-memory-hardening(plan 6, PR #1061). Merge bottom-up: this PR merges after #1061.Closes the one place in
/aftermathwhere a construction bug converts directly into silent remote data loss — the batched Class-1 remote delete (git push origin --delete).What changed
A new
### Class-1 remote deletes — post-batch residual verificationsubsection inskills/aftermath/SKILL.md: after the batch, re-list the remote once and diff survivors two ways against the retained pre-batch snapshot's hold set (the exact-refs/heads/<ref>-name complement of the delete list). A missing hold-set ref is a data-loss row (prints the snapshot SHA + the ready-to-run restore, run only behind an explicit confirm interactively / printed-only under--afk); a surviving delete-list ref is a failed-delete row. The run withholds its clean verdict until the diff is empty or every discrepancy is a report row. A set-difference construction advisory heads off the recorded near-miss (awhile read-subshell exclusion filter that silently no-ops).Prose-only encoding — matching how Class-1 executes (Lead-run by reading the skill). ADR 0027's C3 deletion bar is byte-unchanged: verification adds evidence, opens no deletion path, grows no autonomous ref-creation authority.
Drift protection: two
has_i()pins in the existing Class-1 gate-evidence block (temp-break-proven); the four pre-existing assertions stay byte-untouched. The origin lesson's stale "not encoded anywhere" clause is amended in the same commit as the SKILL edit.CONTEXT.mdgains aresidual-set verificationglossary entry.Phases
f8a06a81edd92dGate at land: full JS + shell suite green,
war-memory lintclean,version-slots.test.mjsgreen. Zero follow-ups.Red-team
CLEARED-WITH-NOTES (1 round). One
needsDecisionself-adjudicated + patched: the stale-claim sweep would hit the landed 2026-07-16 plan — exempted as a point-in-time record. The plan's own backstop #1 (throwawayfile://git-semantics demo) was exercised as a red-team probe and PASSED — git loses the hold ref under the buggy filter, the two-sidedls-remotediff catches it, and the snapshot-SHA restore recovers it.Unexecuted backstops (AI-declared, ADR 0014)
/aftermathon the released plugin🤖 Generated with Claude Code