~ / rigs / a-green-hand-jack / ml-project-repo-agent-native-template

a-green-hand-jack/ml-project-repo-agent-native-template

Agent-native ML research repo template for AI PhD workflows with Claude Code: repo-as-control-plane, repo-local subagents/skills/hooks, anti-drift ANATOMY + runnable governance validators.

↗ GitHub ★ 0 no license updated 3mo ago project Claude Code Codex

share on X

ARCHETYPE

Fort Knox

Deny lists, pre-tool hooks, sandboxing. Nothing touches prod without a signature.

CONTEXT TAX · EVERY TURN

~2.7k tokens

Moderate · median rig: 2.2k · breakdown

GUARDRAILS

5/5

Blocks destructive commands · Protects secrets · Pre-tool screening hook · No YOLO mode · Sandbox or ask-first rules · details

Copy this rig

# review before running: this installs third-party code
$ npx degit a-green-hand-jack/ml-project-repo-agent-native-template/.claude ./rig-ml-project-repo-agent-native-template  # inspect, then merge into .claude/

MCP servers are added to Claude Code at local scope; env vars are shown as YOUR_… placeholders — we never store values. Files are fetched with degit into a separate folder so you can review before merging.

$ npx degit a-green-hand-jack/ml-project-repo-agent-native-template/.agents/skills/adopt-existing-repo .claude/skills/adopt-existing-repo
$ npx degit a-green-hand-jack/ml-project-repo-agent-native-template/.agents/skills/anatomy-drift-control .claude/skills/anatomy-drift-control
$ npx degit a-green-hand-jack/ml-project-repo-agent-native-template/.agents/skills/artifact-indexing .claude/skills/artifact-indexing
$ npx degit a-green-hand-jack/ml-project-repo-agent-native-template/.agents/skills/bootstrap-project .claude/skills/bootstrap-project
$ npx degit a-green-hand-jack/ml-project-repo-agent-native-template/.agents/skills/coding-agent-quota .claude/skills/coding-agent-quota
$ npx degit a-green-hand-jack/ml-project-repo-agent-native-template/.agents/skills/command-adopt-existing-repo .claude/skills/command-adopt-existing-repo
$ npx degit a-green-hand-jack/ml-project-repo-agent-native-template/.agents/skills/command-checkpoint .claude/skills/command-checkpoint
$ npx degit a-green-hand-jack/ml-project-repo-agent-native-template/.agents/skills/command-experiment-watch .claude/skills/command-experiment-watch
$ npx degit a-green-hand-jack/ml-project-repo-agent-native-template/.agents/skills/command-feature-split .claude/skills/command-feature-split
$ npx degit a-green-hand-jack/ml-project-repo-agent-native-template/.agents/skills/command-paper-reproduce .claude/skills/command-paper-reproduce
$ npx degit a-green-hand-jack/ml-project-repo-agent-native-template/.agents/skills/command-pr-review .claude/skills/command-pr-review
$ npx degit a-green-hand-jack/ml-project-repo-agent-native-template/.agents/skills/command-result-promote .claude/skills/command-result-promote
$ npx degit a-green-hand-jack/ml-project-repo-agent-native-template/.agents/skills/command-weekly-maintenance .claude/skills/command-weekly-maintenance
$ npx degit a-green-hand-jack/ml-project-repo-agent-native-template/.agents/skills/experiment-workflow .claude/skills/experiment-workflow
$ npx degit a-green-hand-jack/ml-project-repo-agent-native-template/.agents/skills/interactive-plan-doc .claude/skills/interactive-plan-doc
$ npx degit a-green-hand-jack/ml-project-repo-agent-native-template/.agents/skills/session-boundary-control .claude/skills/session-boundary-control
$ npx degit a-green-hand-jack/ml-project-repo-agent-native-template/.agents/skills/spawn .claude/skills/spawn
$ npx degit a-green-hand-jack/ml-project-repo-agent-native-template/.agents/skills/subagent-routing .claude/skills/subagent-routing
$ npx degit a-green-hand-jack/ml-project-repo-agent-native-template/.agents/skills/template-feedback .claude/skills/template-feedback
$ npx degit a-green-hand-jack/ml-project-repo-agent-native-template/.agents/skills/template-stress-test .claude/skills/template-stress-test
$ npx degit a-green-hand-jack/ml-project-repo-agent-native-template/.agents/skills/workflow-recipe-harvesting .claude/skills/workflow-recipe-harvesting
$ npx degit a-green-hand-jack/ml-project-repo-agent-native-template/.agents/skills/worktree-pr-flow .claude/skills/worktree-pr-flow

$ curl -fsSL --create-dirs -o .claude/agents/artifact-librarian.md https://raw.githubusercontent.com/a-green-hand-jack/ml-project-repo-agent-native-template/main/.claude/agents/artifact-librarian.md
$ curl -fsSL --create-dirs -o .claude/agents/branch-reporter.md https://raw.githubusercontent.com/a-green-hand-jack/ml-project-repo-agent-native-template/main/.claude/agents/branch-reporter.md
$ curl -fsSL --create-dirs -o .claude/agents/checkpoint-writer.md https://raw.githubusercontent.com/a-green-hand-jack/ml-project-repo-agent-native-template/main/.claude/agents/checkpoint-writer.md
$ curl -fsSL --create-dirs -o .claude/agents/experiment-monitor.md https://raw.githubusercontent.com/a-green-hand-jack/ml-project-repo-agent-native-template/main/.claude/agents/experiment-monitor.md
$ curl -fsSL --create-dirs -o .claude/agents/experiment-orchestrator.md https://raw.githubusercontent.com/a-green-hand-jack/ml-project-repo-agent-native-template/main/.claude/agents/experiment-orchestrator.md
$ curl -fsSL --create-dirs -o .claude/agents/feature-worker.md https://raw.githubusercontent.com/a-green-hand-jack/ml-project-repo-agent-native-template/main/.claude/agents/feature-worker.md
$ curl -fsSL --create-dirs -o .claude/agents/hook-maker-agent.md https://raw.githubusercontent.com/a-green-hand-jack/ml-project-repo-agent-native-template/main/.claude/agents/hook-maker-agent.md
$ curl -fsSL --create-dirs -o .claude/agents/interactive-plan-writer.md https://raw.githubusercontent.com/a-green-hand-jack/ml-project-repo-agent-native-template/main/.claude/agents/interactive-plan-writer.md
$ curl -fsSL --create-dirs -o .claude/agents/repo-doc-steward.md https://raw.githubusercontent.com/a-green-hand-jack/ml-project-repo-agent-native-template/main/.claude/agents/repo-doc-steward.md
$ curl -fsSL --create-dirs -o .claude/agents/repo-researcher.md https://raw.githubusercontent.com/a-green-hand-jack/ml-project-repo-agent-native-template/main/.claude/agents/repo-researcher.md
$ curl -fsSL --create-dirs -o .claude/agents/session-boundary-agent.md https://raw.githubusercontent.com/a-green-hand-jack/ml-project-repo-agent-native-template/main/.claude/agents/session-boundary-agent.md
$ curl -fsSL --create-dirs -o .claude/agents/sub-agent-maker-agent.md https://raw.githubusercontent.com/a-green-hand-jack/ml-project-repo-agent-native-template/main/.claude/agents/sub-agent-maker-agent.md
$ curl -fsSL --create-dirs -o .claude/agents/subagent-router-agent.md https://raw.githubusercontent.com/a-green-hand-jack/ml-project-repo-agent-native-template/main/.claude/agents/subagent-router-agent.md
$ curl -fsSL --create-dirs -o .claude/agents/test-runner.md https://raw.githubusercontent.com/a-green-hand-jack/ml-project-repo-agent-native-template/main/.claude/agents/test-runner.md
$ curl -fsSL --create-dirs -o .claude/agents/workflow-recipe-harvester.md https://raw.githubusercontent.com/a-green-hand-jack/ml-project-repo-agent-native-template/main/.claude/agents/workflow-recipe-harvester.md
$ curl -fsSL --create-dirs -o .claude/agents/zh-review-gate.md https://raw.githubusercontent.com/a-green-hand-jack/ml-project-repo-agent-native-template/main/.claude/agents/zh-review-gate.md

Merge into .claude/settings.json (project) or ~/.claude/settings.json (user). Hook commands reference scripts in the source repo — copy those too.

{
  "permissions": {
    "deny": [
      "Agent(general-purpose)",
      "Edit(lab/data/**)",
      "Write(lab/data/**)",
      "NotebookEdit(lab/data/**)",
      "Edit(lab/runs/**)",
      "Write(lab/runs/**)",
      "NotebookEdit(lab/runs/**)",
      "Edit(lab/models/**)",
      "Write(lab/models/**)",
      "NotebookEdit(lab/models/**)",
      "Edit(lab/infra/private/**)",
      "Write(lab/infra/private/**)",
      "NotebookEdit(lab/infra/private/**)",
      "Edit(checkpoints/**)",
      "Write(checkpoints/**)",
      "NotebookEdit(checkpoints/**)",
      "Edit(wandb/**)",
      "Write(wandb/**)",
      "NotebookEdit(wandb/**)",
      "Edit(mlruns/**)",
      "Write(mlruns/**)",
      "NotebookEdit(mlruns/**)",
      "Edit(.env)",
      "Write(.env)",
      "Bash(sudo *)",
      "Bash(curl *| sh)",
      "Bash(pip install *)"
    ],
    "ask": [
      "Bash(git checkout *)",
      "Bash(git switch *)",
      "Bash(git reset *)",
      "Bash(git clean *)",
      "Bash(git rebase *)",
      "Bash(git merge *)",
      "Bash(git branch -d *)",
      "Bash(git branch -D *)",
      "Bash(git worktree remove *)",
      "Bash(uv add *)",
      "Bash(uv sync *)",
      "Bash(uv remove *)",
      "Bash(kill *)",
      "Bash(sbatch *)",
      "Bash(scancel *)",
      "Bash(srun *)",
      "Bash(runai *)",
      "Bash(python lab/infra/launch/fake_job.py launch *)",
      "Bash(python lab/infra/launch/fake_job.py kill *)",
      "Bash(python lab/infra/launch/fake_job.py restart *)",
      "Bash(python lab/infra/launch/fake_job.py _worker *)",
      "Bash(python3 lab/infra/launch/fake_job.py launch *)",
      "Bash(python3 lab/infra/launch/fake_job.py kill *)",
      "Bash(python3 lab/infra/launch/fake_job.py restart *)",
      "Bash(python3 lab/infra/launch/fake_job.py _worker *)",
      "Bash(python -m lab.infra.launch.fake_job launch *)",
      "Bash(python -m lab.infra.launch.fake_job kill *)",
      "Bash(python -m lab.infra.launch.fake_job restart *)",
      "Bash(python -m lab.infra.launch.fake_job _worker *)",
      "Bash(python3 -m lab.infra.launch.fake_job launch *)",
      "Bash(python3 -m lab.infra.launch.fake_job kill *)",
      "Bash(python3 -m lab.infra.launch.fake_job restart *)",
      "Bash(python3 -m lab.infra.launch.fake_job _worker *)",
      "Bash(python lab/infra/launch/expctl.py apply-recovery *)",
      "Bash(python3 lab/infra/launch/expctl.py apply-recovery *)",
      "Bash(env sbatch *)",
      "Bash(env scancel *)",
      "Bash(env srun *)",
      "Bash(env runai *)",
      "Bash(nohup sbatch *)",
      "Bash(nohup scancel *)",
      "Bash(nohup srun *)",
      "Bash(nohup runai *)",
      "Bash(nice *)",
      "Bash(timeout *)",
      "Bash(python -u lab/infra/launch/fake_job.py *)",
      "Bash(python3 -u lab/infra/launch/fake_job.py *)",
      "Bash(env python lab/infra/launch/fake_job.py *)",
      "Bash(env python3 lab/infra/launch/fake_job.py *)",
      "Bash(python -u lab/infra/launch/expctl.py apply-recovery *)",
      "Bash(python3 -u lab/infra/launch/expctl.py apply-recovery *)",
      "Bash(env python lab/infra/launch/expctl.py apply-recovery *)",
      "Bash(env python3 lab/infra/launch/expctl.py apply-recovery *)",
      "Bash(env *)",
      "Bash(/usr/bin/env *)",
      "Bash(CLAUDE_ALLOW_LAUNCH=1 *)",
      "Bash(CODEX_ALLOW_LAUNCH=1 *)",
      "Bash(env -S *)",
      "Bash(env -S*)",
      "Bash(env --split-string *)",
      "Bash(env --split-string=*)",
      "Bash(/usr/bin/env -S *)",
      "Bash(/usr/bin/env -S*)",
      "Bash(/usr/bin/env --split-string *)",
      "Bash(/usr/bin/env --split-string=*)",
      "Bash(bash -c *)",
      "Bash(bash -lc *)",
      "Bash(/bin/bash -c *)",
      "Bash(/bin/bash -lc *)",
      "Bash(sh -c *)",
      "Bash(sh -lc *)",
      "Bash(/bin/sh -c *)",
      "Bash(/bin/sh -lc *)",
      "Bash(zsh -c *)",
      "Bash(zsh -lc *)",
      "Bash(/bin/zsh -c *)",
      "Bash(/bin/zsh -lc *)",
      "Bash(dash -c *)",
      "Bash(dash -lc *)",
      "Bash(/bin/dash -c *)"
    ]
  },
  "hooks": {
    "PreToolUse": [
      {
        "matcher": "Bash|Edit|Write|NotebookEdit",
        "hooks": [
          {
            "type": "command",
            "command": "python3 \"$CLAUDE_PROJECT_DIR/.claude/hooks/pre_tool_guard.py\""
          }
        ]
      }
    ]
  }
}

Skills (22)

Subagents (16)

artifact-librarian model: inherit	维护 dataset/checkpoint/table/figure/result/document 的索引与归档状态、生成 stale asset 报告时使用；绝不删除或移动数据。
branch-reporter model: inherit	汇总所有 active branch/worktree 的状态与依赖关系、生成 worktree 报告时使用；绝不操作 git 写动作。
checkpoint-writer model: inherit	在 compact/clear/handoff 或结束 session 前，把当前工作状态固化到 memory/current-status.md 时使用。
experiment-monitor model: inherit	只读监控长时间运行的实验、做一次性快照健康检查（watcher）并产出结构化 alert 时使用；绝不干预实验。
experiment-orchestrator model: inherit	维护实验从 claim 到 artifact 的证据链、组织实验卡片与状态机 ledger、把 alert 并入台账、校验并交接 human 恢复提案时使用；agent 内恢复执行 fail-closed。
feature-worker model: inherit	在隔离 worktree 中实现单个清晰的功能或 bugfix 时使用；一个 worker 只做一个任务。
hook-maker-agent model: inherit	从反复出现的安全/状态/格式化/compact/artifact 维护动作中提炼 repo-local hook 草稿时使用；需 human review。
interactive-plan-writer model: inherit	把当前阶段写成可让 human 在文件中批注的中文 plan doc、再据批注收敛计划时使用；只写 plans/。
repo-doc-steward model: inherit	维护重要目录的 README/AGENTS/CLAUDE/ANATOMY 导航四件套、根级 DESIGN.md 能力清单、在结构改动时同步 anatomy 时使用。
repo-researcher model: inherit	需要只读地探索代码库、定位实现、收集证据并提出假设时使用；绝不修改文件。
session-boundary-agent model: inherit	判断是否到达 session 边界（该 checkpoint/compact/clear/branch/换 reviewer）并维护 session-tree 时使用。
sub-agent-maker-agent model: inherit	从重复出现的 human-cc 轨迹中提炼出窄边界 subagent 草稿时使用；只产出 draft，不自行启用。
subagent-router-agent model: inherit	为一个 child task 生成 launch packet（agent 类型、预算、模型、边界、停止条件）时使用；只读、只产出 packet。
test-runner model: inherit	运行指定的定向测试命令并总结失败时使用；不擅自全量测试，不粘贴完整日志。
workflow-recipe-harvester model: inherit	从 human-cc trace 中提炼有行为证据的候选 workflow recipe 时使用；只提炼结构片段，不总结主观好用。
zh-review-gate model: haiku	检查即将面向 human 的说明性文字（PR 正文、review、决策文档等）是否遵守"文档默认中文"doctrine、发现大段非中文正文时原地翻译改写时使用；只做语言翻译，不判断内容对错。

Hooks (9)

event	matcher	runs
PreToolUse	Bash\|Edit\|Write\|NotebookEdit	python3 "$CLAUDE_PROJECT_DIR/.claude/hooks/pre_tool_guard.py"
PostToolUse	Edit\|Write	python3 "$CLAUDE_PROJECT_DIR/.claude/hooks/format_changed_python.py"
PostToolUse	Edit\|Write	python3 "$CLAUDE_PROJECT_DIR/.claude/hooks/zh_review_advisory.py"
PreCompact	*	python3 "$CLAUDE_PROJECT_DIR/.claude/hooks/pre_compact_memory_check.py"
SubagentStop	*	python3 "$CLAUDE_PROJECT_DIR/.claude/hooks/subagent_report_index.py"
UserPromptSubmit	*	python3 "$CLAUDE_PROJECT_DIR/.claude/hooks/context_threshold_notice.py"
UserPromptSubmit	*	python3 "$CLAUDE_PROJECT_DIR/.claude/hooks/agent_identity_hook.py"
SessionStart	compact\|clear	python3 "$CLAUDE_PROJECT_DIR/.claude/hooks/context_continuity.py"
SessionStart	*	python3 "$CLAUDE_PROJECT_DIR/.claude/hooks/agent_identity_hook.py"

Slash commands (8)

/adopt-existing-repo/checkpoint/experiment-watch/feature-split/paper-reproduce/pr-review/result-promote/weekly-maintenance

Permissions

deny (27)

Agent(general-purpose)

Edit(lab/data/**)

Write(lab/data/**)

NotebookEdit(lab/data/**)

Edit(lab/runs/**)

Write(lab/runs/**)

NotebookEdit(lab/runs/**)

Edit(lab/models/**)

Write(lab/models/**)

NotebookEdit(lab/models/**)

Edit(lab/infra/private/**)

Write(lab/infra/private/**)

NotebookEdit(lab/infra/private/**)

Edit(checkpoints/**)

Write(checkpoints/**)

NotebookEdit(checkpoints/**)

Edit(wandb/**)

Write(wandb/**)

NotebookEdit(wandb/**)

Edit(mlruns/**)

Write(mlruns/**)

NotebookEdit(mlruns/**)

Edit(.env)

Write(.env)

Bash(sudo *)

Bash(curl *| sh)

Bash(pip install *)

ask (91)

Bash(git checkout *)

Bash(git switch *)

Bash(git reset *)

Bash(git clean *)

Bash(git rebase *)

Bash(git merge *)

Bash(git branch -d *)

Bash(git branch -D *)

Bash(git worktree remove *)

Bash(uv add *)

Bash(uv sync *)

Bash(uv remove *)

Bash(kill *)

Bash(sbatch *)

Bash(scancel *)

Bash(srun *)

Bash(runai *)

Bash(python lab/infra/launch/fake_job.py launch *)

Bash(python lab/infra/launch/fake_job.py kill *)

Bash(python lab/infra/launch/fake_job.py restart *)

Bash(python lab/infra/launch/fake_job.py _worker *)

Bash(python3 lab/infra/launch/fake_job.py launch *)

Bash(python3 lab/infra/launch/fake_job.py kill *)

Bash(python3 lab/infra/launch/fake_job.py restart *)

Bash(python3 lab/infra/launch/fake_job.py _worker *)

Bash(python -m lab.infra.launch.fake_job launch *)

Bash(python -m lab.infra.launch.fake_job kill *)

Bash(python -m lab.infra.launch.fake_job restart *)

Bash(python -m lab.infra.launch.fake_job _worker *)

Bash(python3 -m lab.infra.launch.fake_job launch *)

Bash(python3 -m lab.infra.launch.fake_job kill *)

Bash(python3 -m lab.infra.launch.fake_job restart *)

Bash(python3 -m lab.infra.launch.fake_job _worker *)

Bash(python lab/infra/launch/expctl.py apply-recovery *)

Bash(python3 lab/infra/launch/expctl.py apply-recovery *)

Bash(env sbatch *)

Bash(env scancel *)

Bash(env srun *)

Bash(env runai *)

Bash(nohup sbatch *)

Bash(nohup scancel *)

Bash(nohup srun *)

Bash(nohup runai *)

Bash(nice *)

Bash(timeout *)

Bash(python -u lab/infra/launch/fake_job.py *)

Bash(python3 -u lab/infra/launch/fake_job.py *)

Bash(env python lab/infra/launch/fake_job.py *)

Bash(env python3 lab/infra/launch/fake_job.py *)

Bash(python -u lab/infra/launch/expctl.py apply-recovery *)

Bash(python3 -u lab/infra/launch/expctl.py apply-recovery *)

Bash(env python lab/infra/launch/expctl.py apply-recovery *)

Bash(env python3 lab/infra/launch/expctl.py apply-recovery *)

Bash(env *)

Bash(/usr/bin/env *)

Bash(CLAUDE_ALLOW_LAUNCH=1 *)

Bash(CODEX_ALLOW_LAUNCH=1 *)

Bash(env -S *)

Bash(env -S*)

Bash(env --split-string *)

allow (114)

Edit(**)

Write(**)

Bash(git status*)

Bash(git diff*)

Bash(git log*)

Bash(git show *)

Bash(git branch)

Bash(git branch -v*)

Bash(git branch -a*)

Bash(git branch --list*)

Bash(git branch --show-current)

Bash(git remote -v)

Bash(git remote show *)

Bash(git rev-parse *)

Bash(git ls-files*)

Bash(git ls-tree *)

Bash(git cat-file *)

Bash(git blame *)

Bash(git describe*)

Bash(git shortlog*)

Bash(git config --get *)

Bash(git config --list)

Bash(git config -l)

Bash(git stash list)

Bash(git stash show *)

Bash(git tag)

Bash(git tag --list*)

Bash(git reflog*)

Bash(git for-each-ref*)

Bash(git merge-base *)

Bash(git rev-list *)

Bash(git show-ref*)

Bash(git worktree list)

Bash(git worktree add *)

Bash(git fetch*)

Bash(git add *)

Bash(git commit *)

Bash(git stash)

Bash(git stash push *)

Bash(git stash pop*)

Bash(git restore --staged *)

Bash(git push *)

Bash(ls*)

Bash(cat *)

Bash(head *)

Bash(tail *)

Bash(wc *)

Bash(find *)

Bash(tree*)

Bash(file *)

Bash(stat *)

Bash(du *)

Bash(df*)

Bash(pwd)

Bash(which *)

Bash(command -v *)

Bash(type *)

Bash(echo *)

Bash(printf *)

Bash(sort *)

Context tax

Instructions: 1.3kMCP tool schemas (est.): 0Skill metadata: 932Subagent metadata: 470

ALWAYS-LOADED FILES (codex)

≡

CLAUDE.md · 44 lines · ~356 tok

≡

.agents/CLAUDE.md · 8 lines · ~71 tok

≡

.claude/CLAUDE.md · 17 lines · ~105 tok

≡

AGENTS.md · 71 lines · ~642 tok

≡

.agent/AGENTS.md · 44 lines · ~493 tok

≡

.agents/AGENTS.md · 19 lines · ~155 tok

Estimate: characters ÷ 4 for instruction files and skill/subagent descriptions; MCP schema sizes from typical tool counts. Harnesses with deferred tool loading pay less. Method

Guardrails 5/5

✓

Blocks destructive commands

3 deny/ask rule(s) e.g. Bash(sudo *)

✓

Protects secrets

Denies reads like Edit(.env)

✓

Pre-tool screening hook

1 PreToolUse hook(s)

✓

No YOLO mode

Permission prompts stay on

✓

Sandbox or ask-first rules

91 ask rule(s), e.g. Bash(git checkout *)

Badge

Add to your README:

[![AgentRigs](https://agentrigs.dev/badge/rig/a-green-hand-jack/ml-project-repo-agent-native-template.svg)](https://agentrigs.dev/r/a-green-hand-jack/ml-project-repo-agent-native-template)

Indexed from commit 7c09e90e79c0 on 2026-10-09. Data is structure only — env values and anything secret-looking are stripped.