ara-compile
The ara-compile skill transforms a feeding plan into compiler arguments and executes the external ARA compiler in a single inline call to generate a consistent ARA output directory. Use this when you need to compile research materials into a globally coherent ARA structure with cross-layer bindings, ensuring that claims, proofs, and evidence maintain consistency across semantic deconstruction, cognitive mapping, source layer generation, and exploration graph extraction phases.
git clone --depth 1 https://github.com/yogsoth-ai/de-anthropocentric-research-engine /tmp/ara-compile && cp -r /tmp/ara-compile/skills/ara-compile ~/.claude/skills/ara-compileSKILL.md
# SOP: ARA Compile
**Key question**: 怎么把投喂计划喂给外部 compiler,一次抽出一份内部一致的 ARA?
## Preflight
先确认外部 `compiler` skill 可 load(ARA skills 已装:`npx @ara-commons/ara-skills`)。
若不可用,提示用户安装并**停下**,不要静默继续。
## Why one inline call, not multi-subagent
ARA 的 cross-layer binding(claim→proof→evidence、tree→claim)必须**全局一致**。
分批 compile 会各自从 C01 起撞 ID、断 tree,汇总等于重缝半成品 —— 正是 ARA 要消灭
的事。compiler 自带覆盖度循环(max 3 轮)+ 内建 `Task` 工具;真需要并行由它**内部**
自理,本 SOP 不越俎拆分。
## Procedure
1. **把投喂计划整理成 compiler 的 `$ARGUMENTS`**:
- 主干文件清单 + trace 素材清单 + 图片清单的**路径**(compiler 按路径读);
- 标注哪些是主干(报告线 → claims/problem);
- 大方向作为约束文本(约束 PAPER.md 的 title/abstract);
- `--output ../ara/`(与 `context/` 平级,天然不会被下次 review 当 context 吃回去)。
例:
```
compiler context/2026-06-06-01-30-stage7-...md context/2026-06-05-...stage6...md \
context/figures/*.png \
--output ../ara/ \
主干=stage7(报告线);其余为过程线/图片;大方向:<从 north-star-align 来的一段>
```
2. **一次 inline 运行**:`Skill` load **compiler**,传上面的 `$ARGUMENTS`。
compiler 跑 4 阶段(语义解构 → 认知映射 → src 层 → 探索图抽取)+ 覆盖度循环
+ Seal Level 1。
3. **Seal Level 1 不过**:compiler 自带 fix-iterate(2–3 轮),本 SOP 不接管;
若仍不过,把失败报告**透传**给用户,停。
## Output
`<workspace>/ara/`(`logic/ src/ trace/ evidence/ PAPER.md`),Level 1 已过。
交给 `ara-rigor-review`。Experiment-specific - summarize the DARE executor's research design into a clean research_result report, forced to write back into the spec file produced by formated-specs.
Experiment-specific - replaces writing-specs, emits DARE's 4-layer call plan as a clean research_graph schema. Last step forces load formated-result.
loss-1 judge - read a sample's full dialogue and decide whether the user simulator semantically enacted its Policy Card. check-blind.
loss-2 judge - pairwise quality comparison across the n rungs within one topic; decide monotonicity and endpoint separation. check-blind, D1-D5 only.
Strategy: Inference to the best explanation in the face of anomalies
Remove components one by one, observe system changes to reveal hidden
Map system architecture to ablatable units for ablation studies
Design ablation studies to isolate component contributions in ML systems