# Local Dry-Run Foundation Implementation Plan

> **For Claude:** REQUIRED SUB-SKILL: Use superpowers:executing-plans to implement this plan task-by-task.

**Goal:** Make the current pipeline testable on a local machine without Hermes credentials, blog credentials, or live LLM calls.

**Architecture:** Keep the existing single script as the compatibility entrypoint. Add small, tested helpers for project `.env` loading, dry-run token behavior, and mock LLM responses. This creates a safe base for later Stage 0-8 modularization.

**Tech Stack:** Python standard library, `unittest`, current `script/ai_daily_blog_pipeline.py`.

---

### Task 1: Add Local `.env` Loading

**Files:**
- Modify: `script/ai_daily_blog_pipeline.py`
- Create: `tests/test_env_loading.py`

**Step 1: Write the failing test**

Test that `load_env()` reads project-root `.env` values when Hermes env is absent, and that real process environment variables override file values.

**Step 2: Run test to verify it fails**

Run: `python -m unittest tests.test_env_loading -v`

Expected: FAIL because the script currently only reads `~/.hermes/.env`.

**Step 3: Implement minimal code**

Add a helper to parse env files and update `load_env()` to read:

1. Project `.env`
2. `~/.hermes/.env`
3. process environment

Later sources override earlier ones.

**Step 4: Run test to verify it passes**

Run: `python -m unittest tests.test_env_loading -v`

Expected: PASS.

### Task 2: Let Dry-Run Skip Blog Token Requirement

**Files:**
- Modify: `script/ai_daily_blog_pipeline.py`
- Create: `tests/test_dry_run_config.py`

**Step 1: Write the failing test**

Extract a small helper such as `is_dry_run(env)` and `require_blog_token(env)`, then test:

- `AI_DAILY_DRY_RUN=1` does not require `BLOG_SERVICE_TOKEN`.
- normal publish mode still requires a token.

**Step 2: Run test to verify it fails**

Run: `python -m unittest tests.test_dry_run_config -v`

Expected: FAIL because no helper exists and `main()` checks token before dry-run.

**Step 3: Implement minimal code**

Move dry-run detection before token validation in `main()`.

**Step 4: Run test to verify it passes**

Run: `python -m unittest tests.test_dry_run_config -v`

Expected: PASS.

### Task 3: Add Mock LLM Mode

**Files:**
- Modify: `script/ai_daily_blog_pipeline.py`
- Create: `tests/test_mock_llm.py`

**Step 1: Write the failing test**

Test that `llm_call(prompt, {"AI_DAILY_LLM_MODE": "mock"})` returns valid JSON for:

- semantic dedup prompts
- summary rewrite prompts
- classify prompts

Also test that guide generation can get a non-empty mock response.

**Step 2: Run test to verify it fails**

Run: `python -m unittest tests.test_mock_llm -v`

Expected: FAIL because mock mode does not exist.

**Step 3: Implement minimal code**

Add `AI_DAILY_LLM_MODE=mock` support in `llm_call()`.

**Step 4: Run test to verify it passes**

Run: `python -m unittest tests.test_mock_llm -v`

Expected: PASS.

### Task 4: Add Markdown Smoke Test

**Files:**
- Create: `tests/test_markdown_rendering.py`
- Modify: `script/ai_daily_blog_pipeline.py` only if necessary.

**Step 1: Write the failing or characterization test**

Test that `blog_markdown()` renders:

- `## 导览`
- at least one section
- source links
- no `> >`
- no `[N]`

**Step 2: Run test**

Run: `python -m unittest tests.test_markdown_rendering -v`

Expected: If it already passes, keep it as characterization coverage. If it fails because of `> >`, implement a focused fix.

**Step 3: Implement minimal fix if needed**

Strip leading `>` from guide text before adding blockquote syntax.

**Step 4: Run test to verify it passes**

Run: `python -m unittest tests.test_markdown_rendering -v`

Expected: PASS.

### Task 5: Run Full Verification

**Files:**
- No new files.

**Step 1: Run unit tests**

Run: `python -m unittest discover -s tests -v`

Expected: PASS.

**Step 2: Run compile check**

Run: `python -m py_compile script/ai_daily_blog_pipeline.py`

Expected: exit code 0.

**Step 3: Check git status**

Run: `git status --short`

Expected: only intended files are modified or added.