docs: enforce English-only language policy + retroactive cleanup of Korean issue bodies (#58–#80)

[ComBba]

Summary

Repo language is inconsistent: every shipped artifact (README, code, commits, CHANGELOG, all PR bodies authored by Sangguen, all release notes) is in English, but 23 GitHub issues authored by the auditor agent (#58–#80) shipped in Korean. This creates a bilingual archive that breaks searchability, accessibility for non-Korean reviewers, and visual parity in the issue list.

This issue formalizes the language policy and tracks retroactive cleanup.

---

Root cause

A stale instruction in the auditor agent's persistent memory file hackathon_advisory_stance.md (dated 2026-04-22):

Korean communication: respond in Korean unless the user switches languages (CLAUDE.md is in Korean; user communicates in Korean).

This rule was inferred from a single signal — the user typed prompts in Korean — and was over-generalized into "all output, including GitHub issue bodies, in Korean." The agent's memory index MEMORY.md reinforced the rule with the hook line "Korean by default", which is auto-loaded at every session start. As a result, even when the agent observed English-only artifacts (PRs #81–#99, README, commits), the memory rule overrode the visible signal.

There is no repo-level CLAUDE.md or CONTRIBUTING.md clause documenting language policy, so the agent had no project-side guardrail to override the user-scoped memory rule.

---

Affected artifacts (target scope)

Primary — issue bodies in Korean (23 issues, all CLOSED)

Range	Tier	Count	State
#58–#62	🔴 Critical (C-1..C-5)	5	All closed
#63–#72	🟡 Important (I-1..I-10)	10	All closed
#73–#79	🟢 Nice-to-have (N-1..N-7)	7	2 closed (#78, #79), 5 open with v1.12-backlog label
#80	📋 Master index	1	Closed

Issue titles are bilingual (English tier label + Korean description); issue bodies are predominantly Korean prose with English file paths and code snippets. None of these are PRs or commit messages, so the git history itself is unaffected.

Secondary — auditor chat reports

Mixed Korean/English progress reports delivered to the user in-session. Not persisted to the repo.

Unaffected

• All commits (English ✅)
• All PR bodies fix(cache): atomic strong+weak alias via hardlink to eliminate concurrent put race (#70) #81–docs: CONTRIBUTING + ideation-lead wording cleanup (#95 / #82 #89 follow-up) #101 (English ✅)
• All CHANGELOG/PROGRESS entries (English ✅)
• All code, comments, fixtures, schemas (English ✅)

---

Resolution

Already applied (2026-04-25)

The auditor agent's memory file has been hard-locked to English-only:

• hackathon_advisory_stance.md — language rule rewritten as HARD LOCK: every output (chat + every artifact) must be English; Korean prompts are not a switch signal; only an explicit user directive overrides for a single turn.
• MEMORY.md — index hook updated so the constraint is visible at every session-start auto-load, preventing regression.
• Why-trail in the memory body records all three escalations (original Korean default → English baseline → hard lock) so a future session cannot silently re-introduce the original rule.

Proposed for this repo (subject of this issue)

• Add a one-paragraph language policy to CONTRIBUTING.md so the standard is repo-discoverable and not just agent-memory-only. Suggested wording:

  Language: All issues, PRs, commits, code, and documentation in this repo are written in English. This keeps the artifact archive uniform and accessible to non-Korean reviewers.

• Retroactive cleanup of issue bodies 🔴 C-1: no end-to-end run evidence — clean-room verification needed #58–📋 v1.11.0 5×3 Review — 22-issue master index (#58~#79) #80 — rewrite each body in English while preserving:

  • Original tier label and ID prefix (🔴 C-1, 🟡 I-1, 🟢 N-1, 📋, etc.)
  • Original file paths, line numbers, code snippets, evidence
  • Acceptance-criteria checkboxes (already English)
  • Cross-references between issues
  • Original close timestamps (issue body edits do not alter close metadata)

  All 23 issues are CLOSED; no triage or labels need re-evaluation. Edits are body-only and reversible via issue history.

• Title cleanup (optional) — current titles are bilingual (e.g., 🔴 C-1: 실제 end-to-end run 증거 부재 — clean-room 검증 필요). Decide whether to rewrite to fully English or leave as-is for archive fidelity. Recommendation: rewrite for searchability since titles are the most-indexed surface.

Will NOT be done

• No git history rewrite. Commits, PR bodies, and merged content stay as-shipped.
• No relabeling or re-triage of closed issues. Cleanup is text-only.
• No edits to N-1..N-5 acceptance criteria during translation — only Korean prose translated.

---

Acceptance criteria

• CONTRIBUTING.md contains a "Language" section stating English-only.
• Issues 🔴 C-1: no end-to-end run evidence — clean-room verification needed #58–📋 v1.11.0 5×3 Review — 22-issue master index (#58~#79) #80 bodies are in English (file paths, code, and evidence preserved verbatim).
• Optional: titles for 🔴 C-1: no end-to-end run evidence — clean-room verification needed #58–📋 v1.11.0 5×3 Review — 22-issue master index (#58~#79) #80 fully in English (decision documented in this issue thread before applying).
• No regression in agent memory — hackathon_advisory_stance.md hard lock remains in place.
• This issue closed only after cleanup is verified.

---

References

• Memory file: ~/.claude/projects/-Users-kimsejun-Documents-GitHub-PreviewForgeForClaudeCode/memory/hackathon_advisory_stance.md
• Memory index: ~/.claude/projects/-Users-kimsejun-Documents-GitHub-PreviewForgeForClaudeCode/memory/MEMORY.md
• Master index of affected issues: 📋 v1.11.0 5×3 Review — 22-issue master index (#58~#79) #80
• User directive timestamp: 2026-04-25 (hard lock issued during session review)

[sgwannabe]

Phase 2 partial complete — translator agent shipped 17 issues (#63–#79).

Issue	Title length	Body length	Korean residue (non-code)
#63	72	1707	0
#64	79	3114	0
#65	65	2338	0
#66	74	4323	0
#67	72	3299	0
#68	73	2050	0
#69	76	3887	0
#70	86	3857	0
#71	86	3812	0
#72	97	3483	0
#73	88	2648	0
#74	79	2211	0
#75	80	2192	0
#76	84	2919	0
#77	68	3147	0
#78	70	1980	0
#79	86	3054	0

Critical (#58–#62) + Master (#80) handled by main session in parallel.

[sgwannabe]

All acceptance criteria met

✅ CONTRIBUTING.md language policy

PR #106 merged on main (dcaeb8a) — adds a "Language" section near the top of CONTRIBUTING.md documenting English-only as the repo standard.

✅ Issue body translations (#58–#80)

All 23 issues retroactively translated from Korean to English using gh issue edit --body-file:

• Critical (5): 🔴 C-1: no end-to-end run evidence — clean-room verification needed #58, 🔴 C-2: A-4/A-5/A-6 are prompt-text only — no code enforcement #59, 🔴 C-3: CHANGELOG/PROGRESS doc sync failure — v1.5~v1.11 undocumented #60, 🔴 C-4: 26 advocate boilerplate is not byte-identical — silent template drift #61, 🔴 C-5: '26 advocate convergence' metric absent — headline architectural claim unmeasured #62 — translated by main session.
• Master (1): 📋 v1.11.0 5×3 Review — 22-issue master index (#58~#79) #80 — translated by main session.
• Important (10): 🟡 I-1: verify-plugin.sh assertion broken — schema count 6 vs hardcoded 5 #63–🟡 I-10: LESSON 0.7 fixture only guards symptoms — root cause (panel composite scoring) unverified #72 — translated by background translator agent.
• Nice-to-have (7): 🟢 N-1: generate-gallery.sh inline CSS spaghetti — extract assets/gallery.css recommended #73–🟢 N-7: T-7 (mock-bootstrap harness) deferral graveyard risk — re-open trigger too weak #79 — translated by background translator agent.

Final audit (Korean fragments outside code blocks / inline code, both title and body):

#58–#80 prose_kor=0 title_kor=0 across all 23 issues.

The single legitimate Korean string preserved verbatim is the seed text inside #58's "Recommended scenarios" block (the literal user-input idea 공방 운영자가 수업·재고·정산을 한 곳에서) — that lives inside an inline code span, so it's archive-fidelity, not translation residue.

✅ Title rewrite

All 23 titles fully English while preserving tier emoji (🔴/🟡/🟢/📋) and ID prefix (C-N / I-N / N-N).

✅ No memory regression

The auditor agent's hackathon_advisory_stance.md hard lock from 2026-04-25 remains in place; CONTRIBUTING.md is now the project-side guardrail backing it up.

Out of scope (not changed)

• Git history (commits + PR bodies fix(cache): atomic strong+weak alias via hardlink to eliminate concurrent put race (#70) #81–docs: CONTRIBUTING + ideation-lead wording cleanup (#95 / #82 #89 follow-up) #101 stay as-shipped — already English)
• Issue close metadata (closed_at, labels, assignees)
• Cross-references between issues (preserved verbatim)

Closing as complete.