feat: checkpoint hints, branch path contract, agent-solvability reviews

- Inject progressive per-checkpoint hints into all 40 checkpoint pairs
  (80 EN/ZH lesson files); DA lessons gain full frontmatter so they
  enter the content manifest like the other courses.
- paths/foundation/path.json declares the first-agent branch group so
  the site can treat Claude Code / Codex modules as alternatives.
- Record real-agent solvability runs (Devin, SWE-2 Max) in all 8 course
  REVIEW.md files — challenge-mode runs with observed deviations noted;
  starters restored and --expect-failure re-verified.
- Regenerate content-manifest.json; update repo-plan status lines.
- Verified: verify_courses (8/8), verify_examples (5/5), manifest,
  catalog export, and readme rendering checks all green.

Generated with [Devin](https://devin.ai)

Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>

Author: xxg1413

content-manifest.json

@@ -20,4 +20,20 @@
 
 ## Deviation log
 
-(none recorded yet — first agent-taught run-through pending)
+### 2026-09-13 — agent solvability run (challenge mode, not taught mode)
+
+- Agent: Devin (SWE-2 Max), CLI session on macOS arm64, system Python
+  3.14 (stdlib only — this course needs no third-party deps).
+- Method: implemented `starter/rules_check.py` from `TASK.md` and the
+  test expectations only; did not read or copy `solution/`.
+- Result: `python verify.py starter` exits 0 (all eleven tests);
+  `verify.py progress` printed all five claim codes (l03/l04
+  `[passed]`, l01/l02/l05 `[attest]`).
+- Observed deviation: one iteration needed, not one pass — my first
+  `diverged` issue text said "differs" where the suite asserts the
+  substring "differ from AGENTS.md". The contract pins observable
+  strings, so the fix was wording, not logic.
+- Limitations: challenge-mode run (TASK.md → tests), not the taught
+  COURSE.md walkthrough — teaching quality remains separately pending.
+- Starter restored to the deliberately-unfinished state after the run
+  (`--expect-failure` verified again).

courses/agent-rules-single-source/REVIEW.md

@@ -7,6 +7,13 @@ lang: en-US
 content_version: 1
 status: reviewed
 reviewed_on: 2026-09-12
+hints:
+  - title: "What this checkpoint tests"
+    body: "That you can measure drift instead of feeling it: run the checker on the three scenario repositories and read each result as a test failure, not a vibe."
+  - title: "Find the three repo states"
+    body: "The scenario folders show the three shapes the contract classifies — a clean single-source repo, one with a thin pointer, and one where a copy has drifted. Name which is which before moving on."
+  - title: "When you are done"
+    body: "You can explain what `ok=false` means in terms of files, not feelings: which file diverged from `AGENTS.md` and how the checker knew."
 ---
 
 # Measure the drift you already have

courses/agent-rules-single-source/lessons/L01.md

@@ -7,6 +7,13 @@ lang: zh-CN
 content_version: 1
 status: reviewed
 reviewed_on: 2026-09-12
+hints:
+  - title: "这个检查点考什么"
+    body: "考你能不能把漂移量化而不是凭感觉：对三个 scenario 仓库跑检查器，把每个结果当测试失败读，不当气氛。"
+  - title: "认出三种仓库状态"
+    body: "scenario 文件夹展示了契约要分类的三种形态——干净的单一真源仓库、带薄指针的仓库、副本已漂移的仓库。先说出谁是谁再往下走。"
+  - title: "完成标准"
+    body: "你能用文件而不是感觉解释 `ok=false`：哪个文件偏离了 `AGENTS.md`，检查器是怎么发现的。"
 ---
 
 # 量化你已经存在的漂移

courses/agent-rules-single-source/lessons/L01_cn.md

@@ -7,6 +7,13 @@ lang: en-US
 content_version: 1
 status: reviewed
 reviewed_on: 2026-09-12
+hints:
+  - title: "What this checkpoint tests"
+    body: "Deciding the single source of truth: `AGENTS.md` is the file agents read; every other rule file becomes a thin pointer or an exact copy."
+  - title: "Pointer vs copy"
+    body: "A pointer is at most 10 non-empty lines and references `AGENTS.md`; a copy is byte-identical. Anything else is `diverged`. Write down which of your own repos' files would classify as which."
+  - title: "When you are done"
+    body: "Your decision is written down: which file is the source, and what every other file may contain."
 ---
 
 # Decide the source of truth

courses/agent-rules-single-source/lessons/L02.md

@@ -7,6 +7,13 @@ lang: zh-CN
 content_version: 1
 status: reviewed
 reviewed_on: 2026-09-12
+hints:
+  - title: "这个检查点考什么"
+    body: "定出唯一真源：`AGENTS.md` 是 Agent 真正读的文件，其余规则文件只能是薄指针或逐字节副本。"
+  - title: "指针还是副本"
+    body: "指针最多 10 个非空行且引用 `AGENTS.md`；副本与真源逐字节一致。其余都是 `diverged`。写下你自己仓库里每个文件会被判成哪类。"
+  - title: "完成标准"
+    body: "你的决定落了字：谁是真源，其余文件允许装什么。"
 ---
 
 # 确定唯一真源

courses/agent-rules-single-source/lessons/L02_cn.md

@@ -7,6 +7,13 @@ lang: en-US
 content_version: 1
 status: reviewed
 reviewed_on: 2026-09-12
+hints:
+  - title: "What this checkpoint tests"
+    body: "The gate is `python verify.py starter` — implement `load_rule_files`, `is_pointer`, `check_directory`, and `main` until all eleven tests pass."
+  - title: "Classification order matters"
+    body: "`AGENTS.md` missing or empty yields `missing-source` and `ok=false`. Every other file classifies as `pointer`, `copy`, or `diverged` — check pointer and copy before diverged, and `ok` only when no entry is diverged or missing."
+  - title: "Output shape"
+    body: "`main` prints `ok=... files=... diverged=...` and exits 0 only when ok — argument errors print usage to stderr and return 2."
 ---
 
 # Drive a bounded change, test by test

courses/agent-rules-single-source/lessons/L03.md

@@ -7,6 +7,13 @@ lang: zh-CN
 content_version: 1
 status: reviewed
 reviewed_on: 2026-09-12
+hints:
+  - title: "这个检查点考什么"
+    body: "门控是 `python verify.py starter`——实现 `load_rule_files`、`is_pointer`、`check_directory`、`main`，十一个测试全绿。"
+  - title: "分类顺序很关键"
+    body: "`AGENTS.md` 缺失或为空产出 `missing-source` 且 `ok=false`。其余文件分 `pointer`、`copy`、`diverged`——先判指针和副本再判漂移，没有任何漂移或缺失项时 `ok` 才为真。"
+  - title: "输出形态"
+    body: "`main` 打印 `ok=... files=... diverged=...`，仅当 ok 时退出 0——参数错误打印用法到 stderr 并返回 2。"
 ---
 
 # 按测试驱动一次有边界的变更

courses/agent-rules-single-source/lessons/L03_cn.md

@@ -7,6 +7,13 @@ lang: en-US
 content_version: 1
 status: reviewed
 reviewed_on: 2026-09-12
+hints:
+  - title: "What this checkpoint tests"
+    body: "Both suites plus wiring: prove both verifier directions, run all three skins, then add the checker where your review loop actually runs."
+  - title: "Both directions"
+    body: "The starter must fail where the scenario is broken and pass where it is clean — `--expect-failure` and the plain run tell you which side you are on."
+  - title: "Wire it where it bites"
+    body: "A checker nobody runs is a rule file nobody reads: add it to a pre-commit hook or CI step in one repo you own, and record what it caught first."
 ---
 
 # Verify and wire it in

courses/agent-rules-single-source/lessons/L04.md

@@ -7,6 +7,13 @@ lang: zh-CN
 content_version: 1
 status: reviewed
 reviewed_on: 2026-09-12
+hints:
+  - title: "这个检查点考什么"
+    body: "双套件加接线：两个方向都验证、三个皮肤都跑，然后把检查器接进你真正的评审环。"
+  - title: "两个方向"
+    body: "starter 在坏 scenario 上必须失败、在干净 scenario 上必须通过——`--expect-failure` 和普通运行告诉你自己在哪一侧。"
+  - title: "接到咬得动的地方"
+    body: "没人运行的检查器等于没人读的规则文件：把它加进你某个仓库的 pre-commit 或 CI，记下它第一次抓到了什么。"
 ---
 
 # 验证并接入流程