This repository was archived by the owner on Jul 15, 2026. It is now read-only.
Conversation
install.sh client mode downloaded only 3 lib files but the wrapper routes push-all/pull-all to sync-push-all.mjs/sync-pull-all.mjs, causing MODULE_NOT_FOUND on those commands.
7 scripts exploring LLM-judge quality measurement: - v1-v2: baseline + simulated refine vs direct - v3: real multi-persona broadcast vs direct - v4: added context_fitness dimension to rubric - v5: stricter specificity rubric anchors - v6: diverse domains (frontend/ML/devops) - v7: persona ablation + seed variance + judge drift Key findings stored in mem-mesh: judge drift severe on trade_off (r=-0.087), seed=3 insufficient (σ=2-3), bcast-1 already captures most of the multi-persona benefit. out-*/ directories gitignored (LLM output noise).
… command (unified diagnostic + --fix)\n- x-kit: extend init with --dry-run, --rollback, auto-backup, status labels\n- x-kit: First-Run Init Check gates subcommands on missing hooks\n- x-kit: init now installs x-sync client via install.sh\n- x-sync: include sync-push-all/pull-all in client install\n- x-build: release CLI maintenance (detect/bump/squash/commit)\n- test: isolate core-unit test from user's ~/.xm/config.json
…scover, single decision gate, plain-git fallback\n - Step 0 bundles detect/diff-report/git probes into one Bash call\n - resolve_xmb() auto-detects CLI path (local → cache → plain-git mode)\n - Single AskUserQuestion gate combines squash+bump+commit+push preview\n - README check inlined into commit step\n - Trace runs in background, ship message immediate\n - Adds Plain-Git Fallback section for standalone projects without x-build CLI\n - Common Rationalizations +5, Red Flags section new
…/status/squash → haiku, decision gate → sonnet)\n- x-humble: add Model Routing (CHECK-IN/RECALL → haiku, ANALYZE+ → sonnet guardrail)\n- x-handon: route session restore to haiku, escalate analysis to sonnet\n- x-handoff: route session capture to haiku, narrative summary → sonnet\n- x-dashboard: route start/stop/status/open to haiku (pure script ops)\n- x-ship: flip default behavior — proceed silently, halt only on blockers (bump ambiguity, breaking change, main branch, push-rewrite squash, test/review block, scope leak)\n- x-release: same blocker-only confirmation policy\n- All routing additions include downgrade guardrails to protect reasoning-critical steps
…atcher\n - subcommands: dashboard, handoff, handon, memory, build, trace, sync, solver\n - auto-resolve lib path: → cwd repo → ~/.claude/plugins/cache/x-kit\n - install.sh: local + curl modes, PATH check, bun preflight\n - rationale: single binary preferred over per-tool binaries (gh/kubectl pattern)\n - usage: x-kit dashboard: started at http://127.0.0.1:19841, [memory] 0 memories found for "...", lib: /Users/jinwoo/work/project/agentic/x-kit x-dashboard /Users/jinwoo/work/project/agentic/x-kit/x-dashboard/lib x-build /Users/jinwoo/work/project/agentic/x-kit/x-build/lib x-memory /Users/jinwoo/work/project/agentic/x-kit/x-memory/lib x-trace /Users/jinwoo/work/project/agentic/x-kit/x-trace/lib x-sync /Users/jinwoo/work/project/agentic/x-kit/x-sync/lib x-solver /Users/jinwoo/work/project/agentic/x-kit/x-solver/lib
… add Karpathy 4-principle "Before Non-Trivial Work" (assumptions, ambiguities, tradeoffs, success criteria)\n- x-build: add Planning Principle #6 — surface assumptions before task decomposition\n- x-solver: add Assumptions to init AskUserQuestion — block progress on low-confidence assumptions\n- rationale: discipline pattern (not new skill) — Karpathy's "Think Before Coding" applied as cross-cutting rule
…top-level "Before Non-Trivial Work" (9 lines)\n- CLAUDE.md: add 2 novel rules (#6 surface ambiguities, #7 name low-confidence assumptions) to Planning Principles Universal section\n- rationale: x-eval adversarial judge caught bloat — rules 3-4 of prior section duplicated existing Planning Principles. Net -9 lines, same novel content preserved in better location.\n- self-demonstration: this change applies Simplicity First + Surgical Changes to the prior release itself.
…principles from honor system to enforced structural contract\n - Add mandatory "Section 0: Assumptions & Open Questions" to PRD template\n - Assumptions confidence-tagged (high/medium/low); low = blocks progression\n - Open Questions tracked with status (blocking|answered)\n - Gate rule: any [*, low] or blocking question HALTS task decomposition until user validates\n - Anti-pattern: "No assumptions made" = rejected (minimum 2 + 1 open question)\n - Core rules list updated to enforce\n- rationale: behavioral test showed rule #6/#7 work but honor-system compliance ~40%.\n Structural contract (required PRD section) converts soft principle → gated output.
- x-build/lib/x-build/release.mjs: fix commit message newlines (use git -F <file> instead of -m with JSON.stringify) - x-kit/skills/x-ship/SKILL.md: add Commit Message Rules table (what changed only, no rationale) - .claude/skills/x-release.md: add same commit message rules
# Conflicts: # x-build/lib/x-build/release.mjs # x-kit/lib/x-build/release.mjs
- rubric v4: trade_off_awareness 재정의 → "cost/alternatives/failure_modes 중 명시 개수"로 요소 분해. 주관성 제거 목적. - batch-eval-v8: sonnet/opus/haiku 3자 judge로 v7 출력 30개 재판정. - sonnet↔opus trade_off r = -0.087 (v7) → 0.627 (v8). 대폭 개선. - haiku judge는 correlation ~0.2로 부적합 확인. - analyze-traces.mjs: 기존 .xm/traces/에서 followup_same_skill, followed_by_eval 등 usage proxy 신호 추출. infra 검증됨, 샘플 수 부족. - labels-session1.md: controlled labeling (n=5) 결과. Pearson r=0.746, 방향성 확인 but p>0.05. rubric gap: architectural depth 미검출.
- Evidence Standards table: 5 Valid/Invalid pairs (file:line + quoted
snippet, command output, cited URL with quote) vs ("likely",
"probably", bare URL, general tech explanation). Rejection rule:
leader drops or returns findings lacking Valid evidence.
- Phase Checkpoint template: every phase transition must emit
**PHASE_N_CHECKPOINT:** checklist before AskUserQuestion. Unchecked
items block the transition. Applies inside compose sub-strategies.
New section after CLI: lockfile/manifest → runner mapping for 8 toolchains (bun / pnpm / yarn / npm / uv / pip / cargo / go), base-branch detection via git symbolic-ref with fallback chain, and a pointer to package.json scripts for validation command discovery. Agents probe once per session and reuse, so done_criteria and Verify-phase commands are derived from the project rather than hardcoded.
…ments-stale) Opt-in only via --lenses; not part of the default preset. - silent-failures: empty catch, swallowed errors, `|| null` fallbacks, ignored promise rejections, discarded error returns. - type-design: `any` at API boundaries, missing discriminated unions, nullable leaks across modules. Typed-languages-only gate returns [Info] not applicable for untyped diffs. - comments-stale: comments contradicting the code, TODO without ticket, commented-out code, "what" comments restating identifiers. Each lens follows the existing structure: Principles / Judgment criteria / Severity calibration / Ignore-when / Good finding example / Output format. Severity uses the same Critical/High/Medium/Low scale as the default 7+1 lenses.
- x-op: Evidence Standards table (5 Valid/Invalid pairs) + Phase Checkpoint rule for every phase transition. Agents must cite file:line, command output, or quoted sources; "likely"/"probably" rejected. PHASE_N_CHECKPOINT checklist precedes AskUserQuestion at every boundary, including compose sub-strategies. - x-build: Phase 0 Project Environment Detection added after CLI section. 8 lockfile → runner mappings (bun, pnpm, yarn, npm, uv, pip, cargo, go), git symbolic-ref base-branch fallback chain, package.json scripts probe. Agents derive done_criteria commands from the project instead of hardcoding. - x-review: 3 optional lenses (opt-in via --lenses), not in default preset. silent-failures (empty catch, swallowed errors, `|| null` fallbacks, ignored promise rejections, discarded error returns); type-design (`any` at API boundaries, missing discriminated unions, nullable leaks — typed-languages-only gate); comments-stale (stale comments contradicting code, TODO without ticket, commented-out code). Each follows existing Principles / Judgment / Severity / Ignore-when / Example / Output structure.
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 subscribe to this conversation on GitHub.
Already have an account?
Sign in.
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.
… command (unified diagnostic + --fix)\n- x-kit: extend init with --dry-run, --rollback, auto-backup, status labels\n- x-kit: First-Run Init Check gates subcommands on missing hooks\n- x-kit: init now installs x-sync client via install.sh\n- x-sync: include sync-push-all/pull-all in client install\n- x-build: release CLI maintenance (detect/bump/squash/commit)\n- test: isolate core-unit test from user's ~/.xm/config.json