@litfamily/lithermes 1.0.8 → 1.0.9

package/assets/lithermes-plugin/skills/lit-comprehend/SKILL.md

@@ -117,8 +117,9 @@ full diff during the gate — that happens in step 1 after approval.
 - The real diff (`git diff`, `git log`), not a summary.
 - The CURRENT contents of every file you will quote — open it and read.
 - Durable litgoal state (`goals.json`, `ledger.jsonl`) if present.
-- Every evidence artifact the ledger cites. A missing cited artifact is a
-  finding for the honesty ledger, not a detail to smooth over.
+- Consult relevant durable records when they help explain the change. Keep the
+  detailed verification trail in the internal journal; surface a missing record
+  only when it materially changes what the reader should understand or do.
 
 ### 2. Delta anchoring
 
@@ -144,13 +145,16 @@ A small interactive widget — faithful miniature, slider, step-through, or
 old/new toggle — so the reader can feel the behavior. Always labelled as a
 simplified model. Omit for purely structural changes.
 
-### 6. Honesty ledger ("확인 안 된 것")
+### 6. Reader-facing qualifications and sources
 
-- What is verified by running a command vs merely read.
-- Code that exists but was never executed in this session.
-- Work the objective asked for that did not happen.
-- Missing or stale cited artifacts.
-- Explicit uncertainty — things the artifact does not know.
+Write the explainer for the person who needs to understand the change. Do not
+add a dedicated honesty ledger, evidence table, confidence badge, or source
+label to each claim. Keep command output, verification status, missing records,
+and the detailed reasoning trail in the internal journal. If uncertainty or a
+limitation changes the conclusion or a useful next action, state it once in the
+chat reply in plain language. In the artifact, use relevant citations naturally
+where they help a reader check a source; do not require a citation for every
+sentence or claim.
 
 ### 7. Quiz
 
@@ -175,9 +179,8 @@ confirm the quoted lines actually exist. Without it, the explainer's strongest
 anti-fabrication check (the phantom-quote guard) silently grades nothing, and
 the artifact becomes a plausible story about code that may not exist.
 
-The verifier will FAIL an artifact that contains `<pre>` code blocks but none
-carry the `data-src` attribute. If a code change walkthrough has no code
-excerpts at all, the verifier issues a warning.
+The verifier will FAIL if any `<pre>` code block lacks the `data-src` attribute.
+If a code change walkthrough has no code excerpts, the verifier issues a warning.
 
 ### 9. Build the artifact
 
@@ -193,24 +196,28 @@ template. Populate the canonical sections.
 
 Run `scripts/verify-explainer.py <artifact-path> --repo <worktree>` from the
 installed skill root. The verifier exits nonzero on failure. The skill must not
-claim completion until the verifier passes. Fix any reported defects and re-run.
+claim completion until the verifier passes. Keep its full transcript in the
+internal run record; report the result briefly unless the user asks for detail.
+Fix any reported defects and re-run.
 
 ## #contract.outputs
 
 ### Canonical sections
 
-These Korean headers are byte-identical across all four LitFamily harnesses.
-They must appear in this exact order:
+Use these Korean headers in this order as the explainer's useful structure:
 
-1. **한눈에** — one-paragraph orientation, confidence note
+1. **한눈에** — one-paragraph orientation
 2. **이미 알고 있던 것** — the reader's starting position (delta anchor)
 3. **직관** — essence of each theme; toy data, diagrams
 4. **바뀐 것** — literate walkthrough in conceptual order
 5. **직접 만져보기** — micro-world widget (omit for structural-only changes)
-6. **확인 안 된 것** — honesty ledger
-7. **증거** — claim-to-artifact evidence table
-8. **퀴즈** — interactive quiz
-9. **다음** — three concrete next entry points
+6. **퀴즈** — interactive quiz
+7. **다음** — concrete next entry points
+
+Do not add a fixed source/evidence or verification-status section. Keep useful
+citation links in ordinary prose or a conventional references list when the
+chosen format calls for one. Reader-facing explanations do not need exhaustive
+claim-by-claim proof; retain that detail in the internal research journal.
 
 ### Language
 
@@ -227,24 +234,28 @@ stay verbatim in every language mode.
 ## #contract.output_channels
 
 ```yaml
-artifact_genre: internal_analysis
-limitations_channel: designated_section
+artifact_genre: client_deliverable
+limitations_channel: reply
 ```
 
 ## #contract.evidence
 
 - Every code quote carries a `data-src` attribute naming the quoted file.
 - The verifier checks that quoted lines actually exist in the cited file.
-- The evidence section is a claim / artifact-path / command / result table.
+- Keep command transcripts, test results, and status ledgers in internal records.
+- Add source citations in the artifact when they help check a material factual
+  statement; do not turn them into per-claim labels or a verification table.
 
 ## #contract.hard_stops
 
-- Never claim the verifier passed without running it and showing output.
+- Never claim the verifier passed unless it ran and returned success. Keep the
+  full transcript in the internal run record.
 - Never place the artifact inside the repository worktree.
 - Never use external resources (CDN scripts, remote images, fetch calls).
 - Never fabricate code quotes — read the file first, quote what exists.
 - Never present the micro-world as the real code. Label it as a simplified model.
-- A missing cited artifact is a finding, not something to skip.
+- Preserve meaningful caveats and real sources for factual claims; do not invent
+  support or hide a limitation that changes the reader's decision.
 - Never build the artifact without passing the execution gate first.
 
 ## #contract.anti_patterns
@@ -255,5 +266,5 @@ limitations_channel: designated_section
 - Quiz with all correct answers in the same position.
 - Quiz with the correct answer always being the longest option.
 - Collapsed/hidden code blocks (the verifier rejects them).
-- Claiming completion without verifier output.
+- Claiming completion before the verifier passes.
 - Skipping the execution gate when the scope was inferred, not stated.

package/assets/lithermes-plugin/skills/lit-comprehend/references/artifact-template.md

@@ -1,29 +1,27 @@
 # Explainer artifact template
 
-Section-by-section contract for the comprehend artifact. Each section names
-its Korean header, its purpose, and the HTML patterns to use.
+Use this as a layout guide, not a mandatory audit form. Build a clear, self-contained
+explanation that helps a person understand the change and choose a useful next step.
+Keep test transcripts, status ledgers, detailed verification notes, and source
+inventories in the internal journal.
 
 ## 한눈에
 
-One paragraph orienting the reader: what changed, in what scope, with what
-confidence level.
+Orient the reader in one paragraph: what changed and why it matters. State the
+scope in ordinary language when it helps the reader follow the rest.
 
 ```html
 <section id="overview">
   <h2>한눈에</h2>
-  <p>...</p>
-  <div class="note ok">신뢰도: 높음 — 모든 결과가 명령 실행으로 확인됨</div>
+  <p>...what changed, and why it matters...</p>
 </section>
 ```
 
-Confidence levels: `ok` (verified by commands), `warn` (partially verified),
-`bad` (significant unknowns or missing artifacts).
-
 ## 이미 알고 있던 것
 
-The delta anchor. State what the reader already knew — their objective, the
-brief they wrote, the context they started from. Wrap optional background in
-a skippable block so experienced readers can collapse it.
+Anchor the explanation in the reader's objective, request, or starting point.
+Include only background that helps explain the change. A skippable block is
+optional when the reader may already know that context.
 
 ```html
 <section id="anchor">
@@ -38,10 +36,12 @@ a skippable block so experienced readers can collapse it.
 
 ## 직관
 
-Essence of each theme. Introduce toy data here and reuse it through the
-whole document. Place 2-3 diagrams from these families:
+Explain the essence of each theme before its implementation details. Reuse small
+toy data when useful. Choose two or three diagrams from these families; use
+HTML/CSS rather than ASCII art.
 
 ### Pipeline diagram
+
 ```html
 <div class="pipe">
   <div class="box">입력</div>
@@ -53,20 +53,16 @@ whole document. Place 2-3 diagrams from these families:
 ```
 
 ### Before/after diagram
+
 ```html
 <div class="ba">
-  <div class="side">
-    <h4>이전</h4>
-    <pre>...</pre>
-  </div>
-  <div class="side">
-    <h4>이후</h4>
-    <pre>...</pre>
-  </div>
+  <div class="side"><h4>이전</h4><p>...</p></div>
+  <div class="side"><h4>이후</h4><p>...</p></div>
 </div>
 ```
 
 ### State/timeline table
+
 ```html
 <div class="scroll-x">
   <table>
@@ -77,6 +73,7 @@ whole document. Place 2-3 diagrams from these families:
 ```
 
 ### Simplified UI mockup
+
 ```html
 <div class="ui-mock">
   <div class="mock-header">제목</div>
@@ -86,27 +83,30 @@ whole document. Place 2-3 diagrams from these families:
 
 ## 바뀐 것
 
-Literate walkthrough in conceptual order. Each theme gets a subsection.
-Code quotes carry `data-src` for verifier tracing.
+Group the walkthrough into a few themes in conceptual order. Place code where it
+answers a question the reader now has. Every source excerpt must include a
+repo-relative `data-src` so the verifier can check its provenance.
 
 ```html
 <section id="changes">
   <h2>바뀐 것</h2>
   <h3>테마 1: 라우팅 통합</h3>
   <p>설명...</p>
-  <pre data-src="src/router.py">
-def route(message):
-    ...
-  </pre>
-  <p>추가된 줄은 <span class="add">초록</span>,
-     삭제된 줄은 <span class="del">빨강</span>으로 표시.</p>
+  <pre data-src="src/router.py:20-24">def route(message):
+    ...</pre>
 </section>
 ```
 
+Use natural source links or references when they help readers verify a specific,
+time-sensitive, disputed, or consequential factual statement. Do not add source
+labels after every claim or a claim-to-command evidence table. A real citation
+belongs beside the statement it supports; the internal journal keeps the full
+research and verification trail.
+
 ## 직접 만져보기
 
-Micro-world interactive widget. See `micro-worlds.md` for patterns.
-Always label it as a simplified model. Omit for structural-only changes.
+Add an interactive widget when it helps the reader feel the behavior. Label a
+toy widget as a simplified model; omit it for a purely structural change.
 
 ```html
 <section id="hands-on">
@@ -118,53 +118,11 @@ Always label it as a simplified model. Omit for structural-only changes.
 </section>
 ```
 
-## 확인 안 된 것
-
-Honesty ledger. Must contain substantive content (verifier warns if under
-120 characters). Categories:
-
-- **명령으로 확인됨**: list what was verified by running a command.
-- **읽기만 한 코드**: code that exists but was never executed.
-- **미완료 작업**: work the objective asked for that did not happen.
-- **누락/오래된 증거**: cited artifacts that are missing or stale.
-- **불확실한 점**: explicit uncertainty.
-
-```html
-<section id="unverified">
-  <h2>확인 안 된 것</h2>
-  <ul>
-    <li><strong>읽기만 함</strong>: core_routing.py의 detect_lit_mode — 이 세션에서 실행하지 않음</li>
-    <li><strong>미완료</strong>: 성능 벤치마크는 요청되었으나 수행되지 않음</li>
-  </ul>
-</section>
-```
-
-## 증거
-
-Claim-to-artifact evidence table in a scrollable container.
-
-```html
-<section id="evidence">
-  <h2>증거</h2>
-  <div class="scroll-x">
-    <table>
-      <thead><tr><th>주장</th><th>증거 경로</th><th>명령</th><th>결과</th></tr></thead>
-      <tbody>
-        <tr>
-          <td>테스트 통과</td>
-          <td>test/output.log</td>
-          <td><code>npm test</code></td>
-          <td>347 pass, 0 fail</td>
-        </tr>
-      </tbody>
-    </table>
-  </div>
-</section>
-```
-
 ## 퀴즈
 
-Interactive quiz with feedback for every option.
+Add a short interactive quiz when it reinforces the explanation: five questions
+for a broad change, three for a small one. Give each option feedback and vary
+correct-answer positions. Keep choices similar in length to avoid answer cues.
 
 ```html
 <section id="quiz">
@@ -174,24 +132,17 @@ Interactive quiz with feedback for every option.
     <button class="opt" data-i="0">정규식 하나로 전체 분류</button>
     <button class="opt" data-i="1">파일 확장자 기반</button>
     <button class="opt" data-i="2">토큰 경계 매칭 후 모드별 분기</button>
-    <button class="opt" data-i="3">머신러닝 분류기 사용</button>
     <div class="fb" data-i="0">정규식 하나로는 모드 간 충돌을 해결할 수 없습니다.</div>
     <div class="fb" data-i="1">파일 확장자는 라우팅이 아닌 이벤트 감지에 사용됩니다.</div>
     <div class="fb" data-i="2">맞습니다. 토큰 경계를 확인한 후 각 모드 매처로 분기합니다.</div>
-    <div class="fb" data-i="3">외부 의존성 없이 규칙 기반으로 동작합니다.</div>
   </div>
 </section>
 ```
 
-Rules:
-- 5 questions (3 for small changes).
-- Every option has a feedback `<div class="fb">`.
-- Vary the correct-answer position (`data-answer`): no positional tell.
-- Keep option text lengths roughly even: no length tell.
-
 ## 다음
 
-Three concrete next entry points for the reader.
+Offer concrete next entry points when a follow-up would help. The list may be
+shorter or omitted when there is no useful next action.
 
 ```html
 <section id="next">
@@ -199,7 +150,14 @@ Three concrete next entry points for the reader.
   <ol>
     <li>...</li>
     <li>...</li>
-    <li>...</li>
   </ol>
 </section>
 ```
+
+## Qualifications and references
+
+There is no required "확인 안 된 것" or "증거" section. If a caveat changes the
+reader's interpretation or next action, say it once in the chat reply in plain
+language. Do not fill the artifact with execution status, confidence badges,
+internal source labels, or empty limitation lists. Cite real sources in a normal
+format when they help; do not invent citations or hide a material uncertainty.

package/assets/lithermes-plugin/skills/lit-comprehend/scripts/verify-explainer.py

@@ -12,14 +12,13 @@ Checks:
   filename-dated         Basename must start with YYYY-MM-DD-
   outside-repo           Artifact must NOT be inside the repo worktree
   self-contained         No external resource references
-  sections               All 9 canonical Korean section headers present
+  sections               The 7 useful reader-facing section headers are present
   no-collapsed-code      No <details>/<summary> wrapping <pre> blocks
   code-attribution       Code blocks must carry data-src file attribution
   quotes-real            data-src quoted lines must exist in the cited file
   paths-exist            Repo-relative paths in <code> should exist (WARN)
   no-ascii-art           No box-drawing characters outside <pre> blocks
   quiz                   Quiz structure and anti-tell checks
-  honesty-substantive    Honesty ledger must have substantive content (WARN)
 """
 
 from __future__ import annotations
@@ -36,8 +35,6 @@ CANONICAL_SECTIONS = (
     "직관",
     "바뀐 것",
     "직접 만져보기",
-    "확인 안 된 것",
-    "증거",
     "퀴즈",
     "다음",
 )
@@ -154,10 +151,10 @@ def verify(artifact_path: str, repo_dir: str | None = None) -> list[Finding]:
     stripped_for_pre = STYLE_BLOCK_RE.sub(" ", stripped_for_pre)
     all_pres = CONTENT_PRE_RE.findall(stripped_for_pre)
     attributed_pres = CONTENT_PRE_WITH_SRC_RE.findall(stripped_for_pre)
-    if all_pres and not attributed_pres:
+    if all_pres and len(all_pres) != len(attributed_pres):
         findings.append(Finding(
             "code-attribution",
-            f"Artifact has {len(all_pres)} code block(s) but none carry a data-src attribution attribute",
+            f"Artifact has {len(all_pres)} code block(s), but only {len(attributed_pres)} carry a data-src attribution attribute",
         ))
     elif not all_pres:
         findings.append(Finding(
@@ -265,27 +262,6 @@ def verify(artifact_path: str, repo_dir: str | None = None) -> list[Finding]:
                     warn=True,
                 ))
 
-    # honesty-substantive (WARN)
-    honesty_marker = "확인 안 된 것"
-    if honesty_marker in content:
-        idx = content.index(honesty_marker)
-        # Find next section header or end
-        next_section = None
-        for s in CANONICAL_SECTIONS:
-            if s == honesty_marker:
-                continue
-            pos = content.find(s, idx + len(honesty_marker))
-            if pos != -1 and (next_section is None or pos < next_section):
-                next_section = pos
-        honesty_content = content[idx + len(honesty_marker):next_section] if next_section else content[idx + len(honesty_marker):]
-        honesty_text = re.sub(r"<[^>]+>", "", honesty_content).strip()
-        if len(honesty_text) < 120:
-            findings.append(Finding(
-                "honesty-substantive",
-                f"Honesty ledger content is only {len(honesty_text)} chars (expected >= 120)",
-                warn=True,
-            ))
-
     return findings
 
 

package/assets/lithermes-plugin/skills/lit-diagram-drawer/NOTICE

@@ -0,0 +1,33 @@
+# Third-party notices
+
+This is a LitFamily clean-room port of the diagram-design skill. The icon sheet is a retargeted derivative of the icon sheet in diagram-design commit dc1ace47b99a419e42d01a03cb6ace5346efa8ae. Its layout and selection have been adapted; inline icon paths remain subject to their upstream notices. Complete license texts are in assets/licenses/ and assets/fonts/OFL.txt.
+
+## diagram-design
+
+The upstream diagram-design source is copyright Cathryn Lavery, 2025, and is distributed under the MIT License. The original license text is reproduced at assets/licenses/diagram-design-MIT.txt. The canonical workflow and type guides were rewritten for LitFamily. Names, technical identifiers, asset labels, attribution, and license text may retain upstream wording.
+
+## Tabler Icons
+
+The stroked compute, people, network, data, Kubernetes, action, and DevOps icons are from Tabler Icons, licensed under MIT. Copyright notice and permission text: assets/licenses/tabler-MIT.txt. Upstream: https://github.com/tabler/tabler-icons.
+
+## Simple Icons
+
+The filled brand silhouettes in the icon sheet are from Simple Icons and are dedicated under CC0 1.0; the full text is assets/licenses/simple-icons-CC0.md. Upstream: https://github.com/simple-icons/simple-icons. CC0 does not grant trademark rights; review the individual icon source and brand requirements before commercial use.
+
+## log-z/logos
+
+MySQL, Redis, and StarRocks silhouettes are from log-z/logos and licensed under MIT. Copyright notice and permission text: assets/licenses/log-z-MIT.txt. Upstream: https://github.com/log-z/logos.
+
+## Devicon and one-off icons
+
+RStudio, SPSS, and the Stata glyph are attributed to Devicon, licensed under MIT. Copyright notice and permission text: assets/licenses/devicon-MIT.txt. The Stata glyph was selected from the IcePanel Technology Icons collection published via techicons.dev; its source record is https://techicons.dev/icons/stata.
+
+The SAS glyph is listed as public domain by the pinned diagram-design source. Its source record is https://commons.wikimedia.org/wiki/File:SAS_logo_horiz.svg. Verify that source record and any current brand guidance before reuse.
+
+## Pretendard
+
+Pretendard v1.3.9 is distributed under the SIL Open Font License 1.1. Copyright and reserved-name notices and the complete license text are in assets/fonts/OFL.txt. The bundled full variable WOFF2 is not modified. Provenance, archive digest, and per-file hashes are in assets/fonts/provenance.json.
+
+## Trademarks
+
+Company and product names and logos remain the property of their respective owners. Brand marks are included only as illustrative diagram references. Their presence does not imply endorsement, sponsorship, or affiliation. No trademark rights are granted by the software or icon licenses.