academic-army 0.1.1 → 0.1.2
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/README.md +24 -24
- package/README.zh-CN.md +24 -24
- package/dist/cli.d.ts +1 -0
- package/dist/cli.js +1 -0
- package/dist/cli.js.map +1 -1
- package/metaskills/README.md +10 -10
- package/metaskills/README.zh-CN.md +11 -11
- package/metaskills/academic-army-coding-plan/ENVOLVETASK.md +1 -1
- package/metaskills/academic-army-experiment-plan/ENVOLVETASK.md +1 -1
- package/metaskills/academic-army-experiment-plan/METASKILL.md +0 -1
- package/package.json +1 -1
- package/skills/academic-army-architect/SKILL.md +12 -12
- package/skills/academic-army-architect/references/blueprint-schema.md +27 -0
- package/skills/academic-army-coding-plan/SKILL.md +8 -5
- package/skills/academic-army-experiment-plan/SKILL.md +2 -2
- package/skills/academic-army-repo-scaffold/SKILL.md +1 -1
package/README.md
CHANGED
|
@@ -18,11 +18,11 @@ Parts that require fine-grained research should mainly be handled by skills that
|
|
|
18
18
|
|
|
19
19
|
First use three planning skills to produce three AI-facing Markdown artifacts:
|
|
20
20
|
|
|
21
|
-
| Step
|
|
22
|
-
|
|
23
|
-
| `academic-army-architect`
|
|
24
|
-
| `academic-army-experiment-plan` | `experiment_plan.md` | The experiment strategy that maps paper claims to evidence, datasets or workloads, metrics, baselines, ablations, robustness checks, and reviewer-facing validation needs.
|
|
25
|
-
| `academic-army-coding-plan`
|
|
21
|
+
| Step | Artifact | Role |
|
|
22
|
+
| ------------------------------- | -------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
|
|
23
|
+
| `academic-army-architect` | `paper_blueprint.md` | The strategic paper blueprint that fixes the paper identity, target venue posture, claims, contribution boundary, candidate method space, evidence needs, and downstream constraints. |
|
|
24
|
+
| `academic-army-experiment-plan` | `experiment_plan.md` | The experiment strategy that maps paper claims to evidence, datasets or workloads, metrics, baselines, ablations, robustness checks, and reviewer-facing validation needs. |
|
|
25
|
+
| `academic-army-coding-plan` | `coding_plan.md` | The implementation contract that turns the blueprint and experiment plan into logical module boundaries, interface and entrypoint semantics, harnesses, testing categories, raw-result artifact schemas, and method-freeze rules. |
|
|
26
26
|
|
|
27
27
|
Each planning skill also writes a Chinese `*.explain.md` companion for human review, but the development runner consumes the three English Markdown files above.
|
|
28
28
|
|
|
@@ -119,16 +119,16 @@ Find the closest papers to this research idea, compare their methods, and return
|
|
|
119
119
|
|
|
120
120
|
## Project Structure
|
|
121
121
|
|
|
122
|
-
| Path
|
|
123
|
-
|
|
124
|
-
| `agent-forge.yaml` | Agent and team wiring.
|
|
125
|
-
| `install_mcp.py`
|
|
126
|
-
| `mcp-server/`
|
|
127
|
-
| `skills/`
|
|
128
|
-
| `metaskills/`
|
|
129
|
-
| `runs/`
|
|
130
|
-
| `src/`
|
|
131
|
-
| `output/`
|
|
122
|
+
| Path | Purpose |
|
|
123
|
+
| ------------------ | ----------------------------------------------------------------- |
|
|
124
|
+
| `agent-forge.yaml` | Agent and team wiring. |
|
|
125
|
+
| `install_mcp.py` | Installs the project MCP server into Codex for direct skill runs. |
|
|
126
|
+
| `mcp-server/` | Local stdio MCP implementation that exposes `deepresearch`. |
|
|
127
|
+
| `skills/` | Prepared AcademicArmy skills. |
|
|
128
|
+
| `metaskills/` | Matching metaskill design/evolution files. |
|
|
129
|
+
| `runs/` | Convenience wrappers around TypeScript pipelines. |
|
|
130
|
+
| `src/` | TypeScript pipeline structure and implementation notes. |
|
|
131
|
+
| `output/` | Generated planning artifacts, codebase output, and archives. |
|
|
132
132
|
|
|
133
133
|
Agent and team wiring lives in [`agent-forge.yaml`](agent-forge.yaml). The current TypeScript agents are implemented under [`src/developing/agents`](src/developing/agents) and [`src/evolve-skill/agents`](src/evolve-skill/agents).
|
|
134
134
|
|
|
@@ -136,11 +136,11 @@ Prepared AcademicArmy skills live under [`skills/`](skills/), and their matching
|
|
|
136
136
|
|
|
137
137
|
## Configuration Reference
|
|
138
138
|
|
|
139
|
-
| File or variable
|
|
140
|
-
|
|
141
|
-
| `.env` / `OPENAI_API_KEY` | DeepResearch MCP
|
|
142
|
-
| `agent-forge.yaml`
|
|
143
|
-
| `secret.yaml`
|
|
139
|
+
| File or variable | Required for | Notes |
|
|
140
|
+
| ------------------------- | ---------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
|
|
141
|
+
| `.env` / `OPENAI_API_KEY` | DeepResearch MCP | Read by the MCP server and by `install_mcp.py`. |
|
|
142
|
+
| `agent-forge.yaml` | Project pipelines | Launches `academic_army_mcp_tools` as `python -m mcp-server` with `PYTHONPATH=.` and `cwd=.`. |
|
|
143
|
+
| `secret.yaml` | Prepared shell scripts | Local ignored config overlay used by the prepared wrappers. It may contain passwords, API keys, runtime credentials, or other private values that must not be committed or uploaded to GitHub. |
|
|
144
144
|
|
|
145
145
|
To override or add environment variables directly when installing MCP into Codex, repeat `-e/--env NAME=VALUE`:
|
|
146
146
|
|
|
@@ -152,11 +152,11 @@ Running the installer refreshes the Codex `academic_army_mcp_tools` entry, regis
|
|
|
152
152
|
|
|
153
153
|
## Troubleshooting
|
|
154
154
|
|
|
155
|
-
| Problem
|
|
156
|
-
|
|
157
|
-
| `OPENAI_API_KEY` is missing
|
|
155
|
+
| Problem | Likely cause | Fix |
|
|
156
|
+
| ----------------------------------- | ------------------------------------------------------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------- |
|
|
157
|
+
| `OPENAI_API_KEY` is missing | `.env` is not present or was not forwarded to Codex MCP. | Create `.env`; when running skills directly in Codex, rerun `python install_mcp.py`. |
|
|
158
158
|
| A wrapper cannot find `secret.yaml` | The prepared shell scripts pass a local config overlay for private values such as passwords, API keys, and runtime credentials. | Create local `secret.yaml` or adjust the script to use your config files. Do not commit or upload this file to GitHub. |
|
|
159
|
-
| Development output is drifting
|
|
159
|
+
| Development output is drifting | The planning artifacts are not specific enough. | Revise `paper_blueprint.md`, `experiment_plan.md`, and `coding_plan.md` before continuing development. |
|
|
160
160
|
|
|
161
161
|
## Development
|
|
162
162
|
|
package/README.zh-CN.md
CHANGED
|
@@ -18,11 +18,11 @@ AcademicArmy 的主体核心可以概括为一句话：按图施工。
|
|
|
18
18
|
|
|
19
19
|
先使用三个规划类 skill 交互生成三份面向 AI 执行的 Markdown 产物：
|
|
20
20
|
|
|
21
|
-
| 步骤
|
|
22
|
-
|
|
23
|
-
| `academic-army-architect`
|
|
24
|
-
| `academic-army-experiment-plan` | `experiment_plan.md` | 实验策略，把论文 claims 映射到证据链、数据集或 workload、指标、baselines、消融、鲁棒性检查和审稿人关心的验证点。
|
|
25
|
-
| `academic-army-coding-plan`
|
|
21
|
+
| 步骤 | Artifact | 作用 |
|
|
22
|
+
| ------------------------------- | -------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------ |
|
|
23
|
+
| `academic-army-architect` | `paper_blueprint.md` | 论文战略蓝图，用来固定论文身份、目标 venue 姿态、核心 claims、贡献边界、候选方法空间、证据需求和下游约束。 |
|
|
24
|
+
| `academic-army-experiment-plan` | `experiment_plan.md` | 实验策略，把论文 claims 映射到证据链、数据集或 workload、指标、baselines、消融、鲁棒性检查和审稿人关心的验证点。 |
|
|
25
|
+
| `academic-army-coding-plan` | `coding_plan.md` | 代码实现契约，把论文蓝图和实验方案转成逻辑模块边界、接口与 entrypoint 语义、实验 harness、测试类别、raw result artifact schema 和 method freeze 规则。 |
|
|
26
26
|
|
|
27
27
|
每个规划类 skill 还会同时生成一份中文 `*.explain.md` 解释文件，方便用户审阅；但后续开发 runner 读取的是上面三份英文 Markdown。
|
|
28
28
|
|
|
@@ -119,16 +119,16 @@ Find the closest papers to this research idea, compare their methods, and return
|
|
|
119
119
|
|
|
120
120
|
## 项目结构
|
|
121
121
|
|
|
122
|
-
| 路径
|
|
123
|
-
|
|
124
|
-
| `agent-forge.yaml` | Agent 和团队 wiring。
|
|
125
|
-
| `install_mcp.py`
|
|
126
|
-
| `mcp-server/`
|
|
127
|
-
| `skills/`
|
|
128
|
-
| `metaskills/`
|
|
129
|
-
| `runs/`
|
|
130
|
-
| `src/`
|
|
131
|
-
| `output/`
|
|
122
|
+
| 路径 | 用途 |
|
|
123
|
+
| ------------------ | --------------------------------------------------------- |
|
|
124
|
+
| `agent-forge.yaml` | Agent 和团队 wiring。 |
|
|
125
|
+
| `install_mcp.py` | 把项目 MCP server 安装到 Codex，供直接运行 skill 时使用。 |
|
|
126
|
+
| `mcp-server/` | 本地 stdio MCP 实现，暴露 `deepresearch`。 |
|
|
127
|
+
| `skills/` | 已准备的 AcademicArmy skills。 |
|
|
128
|
+
| `metaskills/` | 对应的 metaskill 设计与 evolution 文件。 |
|
|
129
|
+
| `runs/` | TypeScript pipelines 的便捷 wrappers。 |
|
|
130
|
+
| `src/` | TypeScript pipeline 的目录结构和实现说明。 |
|
|
131
|
+
| `output/` | 生成的规划产物、代码库输出和归档。 |
|
|
132
132
|
|
|
133
133
|
Agent 和团队 wiring 位于 [`agent-forge.yaml`](agent-forge.yaml)。当前 TypeScript agents 分别实现于 [`src/developing/agents`](src/developing/agents) 和 [`src/evolve-skill/agents`](src/evolve-skill/agents)。
|
|
134
134
|
|
|
@@ -136,11 +136,11 @@ Agent 和团队 wiring 位于 [`agent-forge.yaml`](agent-forge.yaml)。当前 Ty
|
|
|
136
136
|
|
|
137
137
|
## 配置参考
|
|
138
138
|
|
|
139
|
-
| 文件或变量
|
|
140
|
-
|
|
141
|
-
| `.env` / `OPENAI_API_KEY` | DeepResearch MCP
|
|
142
|
-
| `agent-forge.yaml`
|
|
143
|
-
| `secret.yaml`
|
|
139
|
+
| 文件或变量 | 用于 | 说明 |
|
|
140
|
+
| ------------------------- | ------------------ | ------------------------------------------------------------------------------------------------------------------------ |
|
|
141
|
+
| `.env` / `OPENAI_API_KEY` | DeepResearch MCP | MCP server 和 `install_mcp.py` 会读取。 |
|
|
142
|
+
| `agent-forge.yaml` | 项目 pipelines | 以 `PYTHONPATH=.` 和 `cwd=.` 运行 `python -m mcp-server`。 |
|
|
143
|
+
| `secret.yaml` | 预设 shell scripts | 预设 wrappers 使用的本地忽略 config overlay。它可以包含密码、API key、runtime 凭据等不能提交或上传到 GitHub 的隐私内容。 |
|
|
144
144
|
|
|
145
145
|
如果需要覆盖或补充环境变量，可以重复使用 `-e/--env NAME=VALUE`：
|
|
146
146
|
|
|
@@ -152,11 +152,11 @@ python install_mcp.py -e OPENAI_API_KEY=your_api_key_here
|
|
|
152
152
|
|
|
153
153
|
## 常见问题
|
|
154
154
|
|
|
155
|
-
| 问题
|
|
156
|
-
|
|
157
|
-
| 缺少 `OPENAI_API_KEY`
|
|
155
|
+
| 问题 | 常见原因 | 解决办法 |
|
|
156
|
+
| ---------------------------- | -------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------- |
|
|
157
|
+
| 缺少 `OPENAI_API_KEY` | 没有 `.env`，或没有把变量转发给 Codex MCP。 | 创建 `.env`；如果直接在 Codex 中跑 skill，再执行 `python install_mcp.py`。 |
|
|
158
158
|
| Wrapper 找不到 `secret.yaml` | 预设脚本传入了本地 config overlay，用来放密码、API key、runtime 凭据等隐私内容。 | 创建本地 `secret.yaml`，或调整脚本使用你的 config 文件。不要把这个文件提交或上传到 GitHub。 |
|
|
159
|
-
| 开发输出偏离规划
|
|
159
|
+
| 开发输出偏离规划 | 三份规划产物还不够具体。 | 先修订 `paper_blueprint.md`、`experiment_plan.md` 和 `coding_plan.md`，再继续开发。 |
|
|
160
160
|
|
|
161
161
|
## 开发
|
|
162
162
|
|
package/dist/cli.d.ts
CHANGED
package/dist/cli.js
CHANGED
package/dist/cli.js.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"cli.js","sourceRoot":"","sources":["../src/cli.ts"],"names":[],"mappings":"
|
|
1
|
+
{"version":3,"file":"cli.js","sourceRoot":"","sources":["../src/cli.ts"],"names":[],"mappings":";AAEA,OAAO,EAA2B,cAAc,EAAE,MAAM,eAAe,CAAC;AACxE,OAAO,EAAE,kBAAkB,EAAE,uBAAuB,EAAE,MAAM,uBAAuB,CAAC;AACpF,OAAO,EAAE,mBAAmB,EAAE,MAAM,yBAAyB,CAAC;AAE9D,SAAS,SAAS,CAAwD,KAIzE;IACC,OAAO;QACL,IAAI,EAAE,KAAK,CAAC,IAAI;QAChB,WAAW,EAAE,KAAK,CAAC,WAAW;QAC9B,GAAG,EAAE,CAAC,IAAuB,EAAE,EAAE,CAAC,cAAc,CAAC,KAAK,CAAC,UAAU,EAAE,IAAI,CAAC;KACzE,CAAC;AACJ,CAAC;AAED,MAAM,cAAc,GAAG;IACrB,SAAS,CAAC;QACR,IAAI,EAAE,YAAY;QAClB,WAAW,EAAE,gCAAgC;QAC7C,UAAU,EAAE,kBAAkB;KAC/B,CAAC;IACF,SAAS,CAAC;QACR,IAAI,EAAE,kBAAkB;QACxB,WAAW,EAAE,qDAAqD;QAClE,UAAU,EAAE,uBAAuB;KACpC,CAAC;IACF,SAAS,CAAC;QACR,IAAI,EAAE,cAAc;QACpB,WAAW,EAAE,+BAA+B;QAC5C,UAAU,EAAE,mBAAmB;KAChC,CAAC;CACM,CAAC;AAEX,SAAS,SAAS;IAChB,MAAM,YAAY,GAAG,cAAc;SAChC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,KAAK,QAAQ,CAAC,IAAI,CAAC,MAAM,CAAC,EAAE,CAAC,IAAI,QAAQ,CAAC,WAAW,EAAE,CAAC;SAC1E,IAAI,CAAC,IAAI,CAAC,CAAC;IAEd,OAAO;;;EAGP,YAAY,EAAE,CAAC;AACjB,CAAC;AAED,MAAM,CAAC,YAAY,EAAE,GAAG,YAAY,CAAC,GAAG,OAAO,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC;AAC9D,MAAM,kBAAkB,GAAG,cAAc,CAAC,IAAI,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,QAAQ,CAAC,IAAI,KAAK,YAAY,CAAC,CAAC;AAE7F,IAAI,kBAAkB,KAAK,SAAS,EAAE,CAAC;IACrC,OAAO,CAAC,GAAG,CAAC,SAAS,EAAE,CAAC,CAAC;IACzB,OAAO,CAAC,QAAQ,GAAG,CAAC,CAAC;AACvB,CAAC;KAAM,CAAC;IACN,MAAM,kBAAkB,CAAC,GAAG,CAAC,YAAY,CAAC,CAAC;AAC7C,CAAC"}
|
package/metaskills/README.md
CHANGED
|
@@ -22,13 +22,13 @@ This is especially useful for the three planning skills used at the start of the
|
|
|
22
22
|
|
|
23
23
|
Prepared AcademicArmy skill metaskills:
|
|
24
24
|
|
|
25
|
-
| Skill
|
|
26
|
-
|
|
27
|
-
| `academic-army-architect`
|
|
25
|
+
| Skill | Edit this file | Run this script |
|
|
26
|
+
| ------------------------------- | ------------------------------------------------------------------------------------------ | -------------------------------------------------------------------------------------------------- |
|
|
27
|
+
| `academic-army-architect` | [`academic-army-architect/METASKILL.md`](academic-army-architect/METASKILL.md) | [`academic-army-architect/envolve.sh`](academic-army-architect/envolve.sh) with `bash` |
|
|
28
28
|
| `academic-army-experiment-plan` | [`academic-army-experiment-plan/METASKILL.md`](academic-army-experiment-plan/METASKILL.md) | [`academic-army-experiment-plan/envolve.sh`](academic-army-experiment-plan/envolve.sh) with `bash` |
|
|
29
|
-
| `academic-army-coding-plan`
|
|
30
|
-
| `academic-army-repo-scaffold`
|
|
31
|
-
| `academic-army-coding-style`
|
|
29
|
+
| `academic-army-coding-plan` | [`academic-army-coding-plan/METASKILL.md`](academic-army-coding-plan/METASKILL.md) | [`academic-army-coding-plan/envolve.sh`](academic-army-coding-plan/envolve.sh) with `bash` |
|
|
30
|
+
| `academic-army-repo-scaffold` | [`academic-army-repo-scaffold/METASKILL.md`](academic-army-repo-scaffold/METASKILL.md) | [`academic-army-repo-scaffold/envolve.sh`](academic-army-repo-scaffold/envolve.sh) with `bash` |
|
|
31
|
+
| `academic-army-coding-style` | [`academic-army-coding-style/METASKILL.md`](academic-army-coding-style/METASKILL.md) | [`../runs/develop-skill.sh`](../runs/develop-skill.sh) with `bash` |
|
|
32
32
|
|
|
33
33
|
Before calling `evolve-skill` for `academic-army-architect`, create or confirm [`academic-army-architect/ENVOLVETASK.md`](academic-army-architect/ENVOLVETASK.md). This fixed task is what the runner uses to test the architect skill across evolution rounds.
|
|
34
34
|
|
|
@@ -124,8 +124,8 @@ Copy an existing [`envolve.sh`](academic-army-architect/envolve.sh) and update t
|
|
|
124
124
|
|
|
125
125
|
## Troubleshooting
|
|
126
126
|
|
|
127
|
-
| Problem
|
|
128
|
-
|
|
127
|
+
| Problem | Likely cause | Fix |
|
|
128
|
+
| ------------------------------------------- | -------------------------------------------------------- | ------------------------------------------------------------------------- |
|
|
129
129
|
| The skill output is still not stable enough | The metaskill guidance is still too vague or incomplete. | Add concrete tips about the failure mode and repeat the evolution script. |
|
|
130
|
-
| Different versions are hard to compare
|
|
131
|
-
| The script fails before the loop starts
|
|
130
|
+
| Different versions are hard to compare | The fixed evolution task is not stable enough. | Rewrite `ENVOLVETASK.md` as a representative fixed task. |
|
|
131
|
+
| The script fails before the loop starts | Dependencies or pipeline config are missing. | Run `npm install` and check the TypeScript entry points and config paths. |
|
|
@@ -22,13 +22,13 @@
|
|
|
22
22
|
|
|
23
23
|
当前已准备的 AcademicArmy skill metaskill 如下：
|
|
24
24
|
|
|
25
|
-
| Skill
|
|
26
|
-
|
|
27
|
-
| `academic-army-architect`
|
|
25
|
+
| Skill | 修改这个文件 | 运行这个脚本 |
|
|
26
|
+
| ------------------------------- | ------------------------------------------------------------------------------------------ | ----------------------------------------------------------------------------------------------------- |
|
|
27
|
+
| `academic-army-architect` | [`academic-army-architect/METASKILL.md`](academic-army-architect/METASKILL.md) | 用 `bash` 运行 [`academic-army-architect/envolve.sh`](academic-army-architect/envolve.sh) |
|
|
28
28
|
| `academic-army-experiment-plan` | [`academic-army-experiment-plan/METASKILL.md`](academic-army-experiment-plan/METASKILL.md) | 用 `bash` 运行 [`academic-army-experiment-plan/envolve.sh`](academic-army-experiment-plan/envolve.sh) |
|
|
29
|
-
| `academic-army-coding-plan`
|
|
30
|
-
| `academic-army-repo-scaffold`
|
|
31
|
-
| `academic-army-coding-style`
|
|
29
|
+
| `academic-army-coding-plan` | [`academic-army-coding-plan/METASKILL.md`](academic-army-coding-plan/METASKILL.md) | 用 `bash` 运行 [`academic-army-coding-plan/envolve.sh`](academic-army-coding-plan/envolve.sh) |
|
|
30
|
+
| `academic-army-repo-scaffold` | [`academic-army-repo-scaffold/METASKILL.md`](academic-army-repo-scaffold/METASKILL.md) | 用 `bash` 运行 [`academic-army-repo-scaffold/envolve.sh`](academic-army-repo-scaffold/envolve.sh) |
|
|
31
|
+
| `academic-army-coding-style` | [`academic-army-coding-style/METASKILL.md`](academic-army-coding-style/METASKILL.md) | 用 `bash` 运行 [`../runs/develop-skill.sh`](../runs/develop-skill.sh) |
|
|
32
32
|
|
|
33
33
|
对 `academic-army-architect` 调用 `evolve-skill` 前，先创建或确认 [`academic-army-architect/ENVOLVETASK.md`](academic-army-architect/ENVOLVETASK.md)（Windows 路径：`metaskills\academic-army-architect\ENVOLVETASK.md`）。这个固定任务是 runner 在 evolution 轮次中测试 architect skill 的输入。
|
|
34
34
|
|
|
@@ -124,8 +124,8 @@ envolve.sh
|
|
|
124
124
|
|
|
125
125
|
## 常见问题
|
|
126
126
|
|
|
127
|
-
| 问题
|
|
128
|
-
|
|
129
|
-
| Skill 输出仍然不稳定
|
|
130
|
-
| 不同版本难以比较
|
|
131
|
-
| 脚本在 loop 开始前失败 | 依赖或 pipeline config 缺失。
|
|
127
|
+
| 问题 | 常见原因 | 解决办法 |
|
|
128
|
+
| ---------------------- | --------------------------------------- | ---------------------------------------------------------------- |
|
|
129
|
+
| Skill 输出仍然不稳定 | Metaskill guidance 仍然太模糊或不完整。 | 继续补充具体 failure mode 和 tips，然后再次运行 evolution 脚本。 |
|
|
130
|
+
| 不同版本难以比较 | 固定 evolution task 不够稳定。 | 把 `ENVOLVETASK.md` 改成更有代表性的固定任务。 |
|
|
131
|
+
| 脚本在 loop 开始前失败 | 依赖或 pipeline config 缺失。 | 运行 `npm install`，并检查 TypeScript 入口和 config 路径。 |
|
|
@@ -1 +1 @@
|
|
|
1
|
-
Create a coding plan based on the paper blueprint in output/paper_blueprint.md and the experiment plan in output/experiment_plan.md.
|
|
1
|
+
Create a coding plan based on the paper blueprint in output/paper_blueprint.md and the experiment plan in output/experiment_plan.md.
|
|
@@ -1 +1 @@
|
|
|
1
|
-
Create an experiment plan based on the paper blueprint in output/paper_blueprint.md.
|
|
1
|
+
Create an experiment plan based on the paper blueprint in output/paper_blueprint.md.
|
|
@@ -44,7 +44,6 @@ experiment plan主要是给AI看的，不是主要给人看的；它只需要客
|
|
|
44
44
|
experiment plan中不应出现`Assumptions to validate`、`Artifact cautions`、`Do not assume reviewers will run code`这类面向用户的部分。
|
|
45
45
|
如果确实存在不确定信息，应将其表达为实验方案中的开放变量、依赖条件或后续规划需要继承的信息，而不是写成用户提醒。
|
|
46
46
|
|
|
47
|
-
|
|
48
47
|
`experiment_plan.md`和`experiment_plan.explain.md`都应为Markdown格式。
|
|
49
48
|
`experiment_plan.md`固定只放experiment plan本身，且使用英文输出。
|
|
50
49
|
`experiment_plan.explain.md`固定只放experiment plan解释和思考流程，且使用中文输出。
|
package/package.json
CHANGED
|
@@ -64,22 +64,22 @@ artifact access, or runtime environment.
|
|
|
64
64
|
Operate at paper-strategy levels 0-2 and route tactical work to downstream
|
|
65
65
|
skills.
|
|
66
66
|
|
|
67
|
-
| Level
|
|
68
|
-
|
|
69
|
-
| 0: paper identity
|
|
70
|
-
| 1: thesis and claims
|
|
67
|
+
| Level | This skill owns |
|
|
68
|
+
| ------------------------ | --------------------------------------------------------------------------------------------------------------------------------------------------- |
|
|
69
|
+
| 0: paper identity | idea, field context, target readers, target venue, paper type |
|
|
70
|
+
| 1: thesis and claims | problem pressure, insight, reader promise, goals, claim hierarchy |
|
|
71
71
|
| 2: strategic constraints | prior-work boundary, contribution roles, high-level design logic, candidate method space, evidence principles, downstream contracts, open variables |
|
|
72
72
|
|
|
73
73
|
Translate tactical inputs into strategic contracts:
|
|
74
74
|
|
|
75
|
-
| User input type
|
|
76
|
-
|
|
77
|
-
| algorithm, proof, or optimization idea
|
|
78
|
-
| module or method combination
|
|
79
|
-
| dataset, trace, benchmark, device, or baseline | evidence dimension, fairness obligation, or user-specified evaluation constraint
|
|
80
|
-
| metric formula
|
|
81
|
-
| figure or section idea
|
|
82
|
-
| implementation detail
|
|
75
|
+
| User input type | Blueprint-level representation |
|
|
76
|
+
| ---------------------------------------------- | ------------------------------------------------------------------------------------------------ |
|
|
77
|
+
| algorithm, proof, or optimization idea | required method property, selected strategic direction, or candidate method route |
|
|
78
|
+
| module or method combination | candidate route with purpose, evidence needed, demotion condition, and fair-comparison principle |
|
|
79
|
+
| dataset, trace, benchmark, device, or baseline | evidence dimension, fairness obligation, or user-specified evaluation constraint |
|
|
80
|
+
| metric formula | outcome family and claim-evidence requirement |
|
|
81
|
+
| figure or section idea | visual or narrative argument that downstream planning must preserve |
|
|
82
|
+
| implementation detail | inherited capability, deployment boundary, or downstream engineering constraint |
|
|
83
83
|
|
|
84
84
|
## Strategy Ledger
|
|
85
85
|
|
|
@@ -14,16 +14,27 @@ Use this default structure:
|
|
|
14
14
|
# Paper Blueprint: <Working Title>
|
|
15
15
|
|
|
16
16
|
## Paper Identity
|
|
17
|
+
|
|
17
18
|
## Core Thesis and Reader Promise
|
|
19
|
+
|
|
18
20
|
## Target Venue Fit
|
|
21
|
+
|
|
19
22
|
## Problem Context and Prior-Work Gap
|
|
23
|
+
|
|
20
24
|
## Paper Goals
|
|
25
|
+
|
|
21
26
|
## Contribution Strategy
|
|
27
|
+
|
|
22
28
|
## Claim-Evidence Architecture
|
|
29
|
+
|
|
23
30
|
## High-Level Design Logic and Scope
|
|
31
|
+
|
|
24
32
|
## Candidate Method Space
|
|
33
|
+
|
|
25
34
|
## Evidence Strategy
|
|
35
|
+
|
|
26
36
|
## Downstream Planning Constraints
|
|
37
|
+
|
|
27
38
|
## Open Strategic Variables
|
|
28
39
|
```
|
|
29
40
|
|
|
@@ -182,9 +193,11 @@ Add contracts for relevant planning areas:
|
|
|
182
193
|
### <Planning Area> Contract
|
|
183
194
|
|
|
184
195
|
Preserve:
|
|
196
|
+
|
|
185
197
|
- <strategic invariant>
|
|
186
198
|
|
|
187
199
|
Delegate:
|
|
200
|
+
|
|
188
201
|
- <tactical design space>
|
|
189
202
|
```
|
|
190
203
|
|
|
@@ -243,19 +256,33 @@ Use this default structure:
|
|
|
243
256
|
# 论文蓝图说明：<Working Title>
|
|
244
257
|
|
|
245
258
|
## 当前论文方案概括
|
|
259
|
+
|
|
246
260
|
## 用户已经明确的内容
|
|
261
|
+
|
|
247
262
|
### 已确认约束
|
|
263
|
+
|
|
248
264
|
### 偏好但未锁定
|
|
265
|
+
|
|
249
266
|
## 当前工作假设
|
|
267
|
+
|
|
250
268
|
## 使用的研究信号
|
|
269
|
+
|
|
251
270
|
### 跨论文模式
|
|
271
|
+
|
|
252
272
|
### 承重信号
|
|
273
|
+
|
|
253
274
|
### 额外背景信号
|
|
275
|
+
|
|
254
276
|
## 核心出发点
|
|
277
|
+
|
|
255
278
|
## 按蓝图顺序解释论文方案
|
|
279
|
+
|
|
256
280
|
## 候选方法空间说明
|
|
281
|
+
|
|
257
282
|
## 开放验证项
|
|
283
|
+
|
|
258
284
|
## 本轮已收缩的开放项
|
|
285
|
+
|
|
259
286
|
## 证据变化时如何校准 Claim
|
|
260
287
|
```
|
|
261
288
|
|
|
@@ -45,18 +45,20 @@ For `output/evolve-*` outputs, or whenever artifact-access feedback is active or
|
|
|
45
45
|
|
|
46
46
|
Use five-backtick fences for full-file handoffs so embedded command fences remain readable:
|
|
47
47
|
|
|
48
|
-
|
|
48
|
+
``````markdown
|
|
49
49
|
## output/evolve-.../coding_plan.md
|
|
50
50
|
|
|
51
|
-
|
|
51
|
+
```markdown
|
|
52
52
|
<full coding_plan.md content>
|
|
53
|
-
|
|
53
|
+
```
|
|
54
|
+
``````
|
|
54
55
|
|
|
55
56
|
## output/evolve-.../coding_plan.explain.md
|
|
56
57
|
|
|
57
|
-
|
|
58
|
+
```markdown
|
|
58
59
|
<full coding_plan.explain.md content>
|
|
59
|
-
|
|
60
|
+
```
|
|
61
|
+
|
|
60
62
|
````
|
|
61
63
|
|
|
62
64
|
When files are long, read each file with a complete read method or bounded chunks before composing the handoff. Paste the read-back contents, not a regenerated approximation. If read-back fails after writing, try another local read mechanism. If read-back remains impossible, report the read-back failure clearly and mark delivery blocked rather than presenting unverified contents.
|
|
@@ -489,3 +491,4 @@ After writing and validating the files, summarize:
|
|
|
489
491
|
- validation performed, including read-back result
|
|
490
492
|
|
|
491
493
|
For `output/evolve-*` outputs or when artifact-access feedback requests pasted contents, add a `Review Handoff` heading immediately after the concise validation sentence and paste the complete read-back contents of both files using the five-backtick handoff format. A path-only response is incomplete for access-limited review.
|
|
494
|
+
````
|
|
@@ -328,7 +328,7 @@ objectives.
|
|
|
328
328
|
## 2. Claim-to-Evidence Map
|
|
329
329
|
|
|
330
330
|
| Claim | Reviewer Concern | Evidence Objective | Story Placement | Expected Evidence Output |
|
|
331
|
-
|
|
331
|
+
| ----- | ---------------- | ------------------ | --------------- | ------------------------ |
|
|
332
332
|
|
|
333
333
|
## 3. Workload and Dataset Registry
|
|
334
334
|
|
|
@@ -414,7 +414,7 @@ is clearer.
|
|
|
414
414
|
## 实时调研如何影响实验取舍
|
|
415
415
|
|
|
416
416
|
| 来源 | 日期 | venue_status | 影响到的规划决定 |
|
|
417
|
-
|
|
417
|
+
| ---- | ---: | ------------ | ---------------- |
|
|
418
418
|
|
|
419
419
|
## 实验故事线
|
|
420
420
|
|
|
@@ -699,7 +699,7 @@ Treat any of these as validation failures:
|
|
|
699
699
|
- README, REFERENCES, harness explanations, test notes, or retained docs use
|
|
700
700
|
internal artifact-management wording such as `External Task Inputs`,
|
|
701
701
|
`external inputs`, whole-word `bucket`, vague `consulted`, `operators
|
|
702
|
-
|
|
702
|
+
execute`, or similar process language,
|
|
703
703
|
- REFERENCES includes sources that were only searched or rejected and have no
|
|
704
704
|
dependency, attribution, license, benchmark, harness, or implementation value,
|
|
705
705
|
- retained template files still contain default template project names, template
|