Distillation

Date: July 21, 2026

• 🟡 (19:30) User asked for an adversarial senior-code-reviewer pass on getsentry/craft branch feat/prefixed-tags-fixes vs origin/master, called "PR A" — salvageable subset of parked PR #844 (monorepo prefixed-tag support) plus two review fixes. Explicit constraint: do NOT modify files, do NOT rubber-stamp.
• 🔴 (19:30) User stated PR A contents: (1) prefix-aware read paths from #844 — getLatestTag(git, tagPrefix='') uses git describe --match '<prefix>*', threaded through prepare.ts/changelog.ts; getGitTagPrefix() in config.ts warns on multiple github targets with differing tagPrefix and returns the first; version.ts extracts version from prefixed tags (e.g. cli@1.2.3→1.2.3). (2) Fix 1 (Bugbot Medium): changelog.ts wraps getGitTagPrefix() in try/catch so invalid .craft.yml doesn't abort standalone craft changelog, falls back to empty prefix. (3) Fix 2 (BYK review + Bugbot Low): git.ts findReleaseBranches treats release-branch prefix as opaque string via lastIndexOf('/')+slice (replacing old segment-count split/join), fuzzy matching (Levenshtein ≤3) unchanged.
• 🟡 (19:30) User specified verification commands: npx tsc --noEmit -p tsconfig.build.json, pnpm test src/utils/__tests__/git.test.ts src/commands/__tests__/changelog-versioning-policy.test.ts src/__tests__/config.test.ts, full pnpm test, pnpm exec prettier --check on changed files.
• 🟡 (19:30) User requested prioritized findings list (CRITICAL/MAJOR/MINOR) with file:line and concrete fixes, explicitly asking whether the opaque-prefix rewrite introduces ANY behavior regression vs old segment logic (enumerated) and whether HEAD-entry/trailing-slash/empty-version edges are handled correctly.
• 🟡 (19:31) [investigation] Assistant ran git log --oneline origin/master..HEAD → single commit 412b796 "feat: support prefixed tags for monorepo multi-product releases".
• 🟡 (19:31) Assistant investigation hypothesis raised: git diff origin/master...HEAD (committed diff) shows OLD prefix.split('/').length segment-count logic in findReleaseBranches, NOT the lastIndexOf opaque-prefix logic described in the task — flagged as a critical discrepancy requiring working-tree inspection.
• 🟡 (19:32) Assistant read src/utils/git.ts (lines 218-275, working tree) confirming the WORKING TREE contains the new opaque-prefix logic: lastSlash = withoutRemote.lastIndexOf('/'); skip if lastSlash <= 0; branchPrefix = withoutRemote.slice(0, lastSlash); exact match if branchPrefix === prefix; fuzzy if levenshtein(branchPrefix, prefix) <= MAX_EDIT_DISTANCE (MAX_EDIT_DISTANCE=3).
• 🔴 (19:32) [investigation] Root cause found: Fix 2 (opaque-prefix rewrite) is UNCOMMITTED — exists only in working tree, not in HEAD commit 412b796. git status showed branch ahead of origin/master by 1 commit, with unstaged modifications to: .lore.md, docs/src/content/docs/targets/github.md, src/commands/tests/changelog-versioning-policy.test.ts, src/commands/changelog.ts, src/utils/tests/git.test.ts, src/utils/git.ts; untracked: .craft-issue-842-plan.md, .opencode/.
• 🟡 (19:32) Unstaged diff for git.ts confirmed: OLD logic used prefixSegmentCount = prefix.split('/').length, branchSegments = withoutRemote.split('/'), branchPrefix = branchSegments.slice(0, prefixSegmentCount).join('/'), skip condition !branchPrefix || branchSegments.length <= prefixSegmentCount. NEW logic (working tree) replaces all of this with the lastIndexOf/slice approach described above.
• 🟡 (19:33) Assistant confirmed getLatestTag (git.ts lines 45-71, working tree) signature: getLatestTag(git, tagPrefix = ''); when tagPrefix set, appends --match '<prefix>*' to git describe --tags --abbrev=0 args; catches errors starting with 'fatal: No names found' or 'Nothing to describe' → returns empty string; otherwise rethrows.
• 🟡 (19:33) Assistant confirmed parseGitBranchOutput (git.ts lines 198-207) filters out lines containing -> (i.e., origin/HEAD -> origin/main entries) BEFORE the findReleaseBranches matching loop runs — HEAD-entry edge case is handled at parse time, not in the matching loop itself.
• 🟡 (19:34) Assistant read working-tree src/utils/tests/git.test.ts findReleaseBranches describe block (lines 129-316), noting it contains NEW/updated tests not in the committed diff: "matches slashed (monorepo) release-branch prefixes exactly" (lines 272-289, prefix release/cli against branches release/cli/1.2.0, release/cli/1.2.1, release/mcp/2.0.0, release/1.0.0 — expects exact=[cli/1.2.1, cli/1.2.0], fuzzy=[mcp/2.0.0] since levenshtein('release/mcp','release/cli')=3); "does not match a slashed prefix branch that lacks a version segment" (lines 291-300, branch release/cli with no version part → no matches); "treats the prefix opaquely: 'release' does not claim 'release/cli/x' branches" (lines 302-315, prefix release against release/1.0.0 and release/cli/1.2.3 → exact=[release/1.0.0] only, fuzzy=[] since branch-prefix 'release/cli' is distance 4 from 'release').
• 🔴 (19:35) [bug-found] Assistant identified a genuine behavior regression between OLD (committed) and NEW (uncommitted working-tree) findReleaseBranches logic: for branch release/cli/1.2.3 with search prefix release, OLD logic computes branchPrefix as first segment only ('release'), branchSegments.length(3) > prefixSegmentCount(1) → considered → EXACT MATCH; NEW logic computes branchPrefix via last-slash as 'release/cli', levenshtein('release/cli','release')=4 → EXCLUDED entirely (no exact or fuzzy match). Flagged as a real divergence needing explicit callout in the review findings.
• 🟡 (19:36) Assistant read working-tree src/commands/changelog.ts (lines 50-89): full try/catch (lines 69-78) wraps BOTH findConfigFile() and getGitTagPrefix(); on error, falls back to tagPrefix='' and logs debug message "Could not read tag prefix from config; using latest tag overall"; then calls since = await getLatestTag(git, tagPrefix).
• 🔴 (19:36) [bug-found] Assistant confirmed the COMMITTED diff version of changelog.ts had no try/catch — just const tagPrefix = findConfigFile() ? getGitTagPrefix() : ''; — meaning Fix 1 (Bugbot Medium fix) is present only in the uncommitted working tree; the committed commit 412b796 still has the original bug where an invalid .craft.yml would abort a standalone craft changelog run.
• 🟡 (19:37) [investigation] Verification run 1: npx tsc --noEmit -p tsconfig.build.json → exit code 0 (no type errors).
• 🟡 (19:37) [investigation] Verification run 2: pnpm test on targeted files (git.test.ts, changelog-versioning-policy.test.ts, config.test.ts) → 3 test files passed, 53 tests passed, duration 2.46s. Debug logs noted during run: "Failed to fetch from remote, using locally cached refs" and "Failed to list remote branches" (expected mock behavior in tests).
• 🟡 (19:37) Docs changes noted (working tree): docs/src/content/docs/configuration.md adds a section on slashed releaseBranchPrefix for monorepos with example releaseBranchPrefix: release/cli + github.tagPrefix: "cli@" producing branches release/cli/1.2.3 and tags cli@1.2.3; docs/src/content/docs/targets/github.md adds "Monorepo: independently-versioned products" section explaining tagPrefix namespacing (e.g. cli@1.2.3, mcp@2.0.0), states read-side scoping covers latest-tag detection/changelog base/CalVer scans, includes a caution note and a "Known limitation: the GitHub 'Latest' badge is repo-wide" note.
• 🟡 (19:36) src/tests/config.test.ts (working tree) contains tests for getGitTagPrefix(): warns via logger.warn when multiple github targets have differing tagPrefix (e.g. cli@ vs mcp@) and returns the first prefix; does not warn when multiple github targets share the same tagPrefix (e.g. two targets both cli@, one with id: 'second').