v0.5.0 — 25-issue batch: run/jd-check/review-diff/doctor, agentic persona, per-locale review

[0.5.0] — 2026-05-28

Issue-driven release closing 25 issues filed against 0.4.0 (one declined).
No breaking changes — all additive or bug fixes.

Fixed (P0)

• review: persona+locale resolution (#2). review --persona X --locale Y
  without --file/--version now globs resume_v*_<locale>[_<persona>].md
  and scores the highest matching version instead of silently scoring the
  lexically-latest file (which previously overwrote N−1 results in a batch).
• review: DOB false positive (#3). The "full DOB in header" red flag
  required only a bare ISO date in the first 14 lines, so summary /
  target-role / experience-start dates triggered a phantom −2 on every
  en_US/en_GB render. Now requires an explicit DOB / Date of birth / Born /
  生年月日 / 出生 label adjacent to the date. (The en_US template never
  rendered profile.dob — the bug was entirely in the reviewer's scanner.)
• render: filename locale suffix (#4). en_US renders dropped the locale
  suffix (resume_v006_tech_lead.md) while other locales kept it, breaking
  resume_v*_en_US_*.md globs. Every render now uses
  resume_v<NNN>_<locale>[_<persona>].md uniformly.

Added

• vibe-resume run orchestrator (#10). One command for multi-persona ×
  multi-locale pipelines. Phase A: extract + aggregate (if cache stale) +
  enrich emit across the matrix, then stops for session processing. Phase B
  (--continue): ingest --all + render matrix + review matrix + trend.
  No auto-dispatch — preserves the session-quota model.
• vibe-resume jd-check (#23). Reports JD-keyword coverage across
  enriched bullets without a full render+review round. --threshold filters
  to under-covered keywords.
• vibe-resume review-diff <vA> <vB> (#26). Per-check scorecard delta
  between two résumé versions.
• vibe-resume doctor (#19). Diagnoses CLI/plugin version drift,
  profile/config presence, optional-dep (pandoc/claude) availability.
• Agentic Engineer persona (#22). --persona agentic — 2026 AI-agent
  narrative (agent loops, MCP, RAG triad, eval harness) with a Head-of-AI
  reviewer lens and keyword-echo-weighted scoring.
• enrich --tailor-keywords / --tailor-keywords-cap / --tailor-keywords-strict (#7).
  Manual keyword overrides; override terms always lead the merged set.
• enrich --ingest --all (#9). Walks every (persona, locale) under
  data/enrich_jobs/ and ingests each; comma-separated persona ingest too.
• enrich --status + --ingest --all-ready (#12). Progress table across
  job dirs; batch-ingest only the complete ones.
• enrich --clean (#20). Clears stale *.yaml on re-emit; a warning
  fires when old yaml survives a re-emit without --clean.
• render --all-locales --persona X,Y,Z matrix (#11). Persona-list
  expansion across all locales.
• status --enriched / --pending / --all (#25). Cache-state views per
  (persona, locale) and pending job progress.
• Profile-derived redaction (#24). derive_profile_redactors auto-builds
  name/email patterns from profile.yaml (incl. locale name variants) and
  scrubs bullets at ingest, regardless of which invocation path ran.
• Manifest JD provenance (#8). Emit records the JD's sha256 + mtime +
  extracted keywords; ingest warns when the JD file changed since emit.
• Persona-specific review weights (#16). review --persona now reweights
  the 8-point scorecard (e.g. tech_lead weights metrics 1.5×, page-count
  0.7×) and normalises back to the same scale, not just appends a lens note.
• Per-locale page-count targets + gradual scoring (#14). ja_JP=1.0,
  de_DE/fr_FR=2.5, others=2.0; 4-band scoring (10/8/5/2) replaces the
  10→5 cliff.
• Trend grouped by (locale, persona) (#15). Default grouping splits
  apples-vs-oranges persona runs; --group-by + --persona/--locale
  filters.
• render warns when profile.summary empty (#13) — surfaces the −4
  top-fold penalty before it lands.
• uv tool install support (#18). Wheel now force-includes cli.py so
  uv tool install git+… puts vibe-resume on PATH. (No src/ restructure —
  tracked separately if desired.)

Docs

• SKILL.md documents the --level axis in the main Procedure + Quick
  Reference (#6); personas-compare quick-ref notes the --locale
  requirement since 0.4.0 (#5).
• New references/tailor-keyword-extraction.md explains the two-pass
  extractor + 12-keyword cap rationale (#21).

Closed without code

• #1 (list in awesome-codex-plugins) — declined; discovery is via
  skills.sh + Claude Code / Codex marketplaces.
• #17 (plugin gitCommitSha drift) — upstream Claude Code behaviour, not
  a vibe-resume bug; workaround documented; the work-tree drift we can
  detect ships as doctor (#19).