Agent搜索
- 作者仓库星标 435
- 作者更新于 实时读取
- 作者仓库 nvim
- 领域
- 工程开发
- 兼容 Agent
-
- Claude Code
- Cursor
- Cline
- Codex
- Windsurf
- Gemini CLI
- +20
- 信任分
- 88 / 100 · 社区维护
- 作者 / 版本 / 许可
- @benbrastmckie · 未声明 license
- Token 消耗评级
- 低消耗
- 接入复杂程度
- 需简单配置
- 是否需要外部 API Key
- 不需要
- 兼容的系统
- 未声明(默认跨平台)
- 底层运行要求
- 无特殊要求
- 文件与系统权限
-
- 只读
- 允许写入 / 修改
- Shell 执行
- 网络行为
- 仅限本地
- 安装命令数
- 26 条
档案由构建时根据 SKILL.md 与安装命令自动衍生,可能与作者实际意图存在差异。
需要注意: 未限定 allowed-tools,默认拥有全部工具权限。
---
name: skill-lean-implementation
description: Implement Lean 4 proofs and definitions using lean-lsp tools. Invoke for Lean-language implement…
category: 工程开发
runtime: 无特殊运行时
---
# skill-lean-implementation 输出预览
## PART A: 任务判断
- 适用问题:代码实现、重构、调试或代码审查。
- 输入要求:目标材料、限制条件、期望输出和验收方式。
- 证据边界:围绕“Trigger Conditions / Execution Flow / Stage 1: Input Validation”读取原文规则,不把推断写成作者承诺。
## PART B: 执行结果
- **01** 任务判断:确认你的需求是否属于代码实现、重构、调试或代码审查,并标出输入、限制和预期结果。
- **02** 执行计划:优先按“Trigger Conditions / Execution Flow / Stage 1: Input Validation”拆成步骤,说明每一步会读取什么、修改什么、产出什么。
- **03** 交付结果:给出可复制的命令、文件改动、检查清单或内容草稿,并说明如何继续迭代。
- **04** 风险边界:结合 读取文件、写入/修改文件、执行终端命令、主要在本地完成、通常不需要额外 API Key 给出执行前确认项。
## Running Rules
- 读取文件、写入/修改文件、执行终端命令;主要在本地完成;通常不需要额外 API Key。
- 先小样例验证,再放大到真实任务。
- 交付时同时给结果、检查口径和下一步迭代建议。 原文出现了 `/implement`、`/team-implement` 这类斜杠命令;如果你的 Agent 支持命令触发,优先用命令开场,再补充目标和边界。
告诉 Agent 目标文件或材料、期望结果、不可改范围、是否允许联网或执行命令。本 Skill 的权限画像是:读取文件、写入/修改文件、执行终端命令。
先用一个小任务确认它会围绕“Trigger Conditions / Execution Flow / Stage 1: Input Validation”工作;涉及文件或命令时,先看 diff、日志、预览或测试结果。
检查最终产物是否包含明确结果、必要证据和下一步动作;如果输出泛泛而谈,就补充输入、边界和验收标准后重跑。
---
name: skill-lean-implementation
description: Implement Lean 4 proofs and definitions using lean-lsp tools. Invoke for Lean-language implement…
category: 工程开发
source: benbrastmckie/nvim
---
# skill-lean-implementation
## 什么时候使用
- 把工程方向的常用动作沉淀成 Agent 可调用的技能 适合处理工程开发场景下的代码实现、调试、重构、测试或代码审查,核心价值是把输入、判断、执行、验证和交付边界固定下来,避免 Agent 泛泛回答。 把任务拆成可执行、可检查、可继续迭代…
- 面向代码实现、重构、调试或代码审查,优先处理能明确输入、步骤和验收标准的工作。
## 需要提供什么
- 目标材料、目录范围、期望结果和不可改动内容。
- 是否允许联网、执行命令、读写文件或调用外部服务。
## 执行规则
- 围绕「Trigger Conditions / Execution Flow / Stage 1: Input Validation」组织步骤,不把推断写成作者事实。
- 读取文件、写入/修改文件、执行终端命令;主要在本地完成;通常不需要额外 API Key。
- 先跑小样例,确认结果可检查后再扩大任务范围。
## 输出要求
- 给出最终产物、关键证据、验证方式和下一步动作。
- 信息不足时标记 unknown,不编造命令、平台或依赖。 作者原文负责流程事实;仓库文件负责来源和命令;流狐只补充适用场景、限制和质量判断。
skill "skill-lean-implementation" {
输入层 -> 用户目标 + 目标文件 + 禁止范围 + 验收标准
上下文层 -> Trigger Conditions / Execution Flow / Stage 1: Input Validation
规则层 -> SKILL.md 触发条件 / 执行顺序 / 输出格式
运行层 -> 无特殊运行时 | 读取文件、写入/修改文件、执行终端命令 | 主要在本地完成
安全层 -> 通常不需要额外 API Key + 小任务验证 + diff / 日志复核
输出层 -> 可复制结果 + 检查清单 + 下一步迭代
} Lean Implementation Skill
Thin wrapper that delegates Lean 4 proof implementation to lean-implementation-agent subagent.
IMPORTANT: This skill implements the skill-internal postflight pattern. After the subagent returns, this skill handles all postflight operations (status update, artifact linking, git commit) before returning.
Trigger Conditions
This skill activates when:
- Task type is "lean4" or "lean" (either accepted)
- /implement command targets a Lean task
- Plan exists and task is ready for implementation
Execution Flow
Stage 1: Input Validation
Validate required inputs:
task_number- Must be provided and exist in state.json- Task status must allow implementation (planned, implementing, partial)
- Task type must be "lean" or "lean4"
# Lookup task
task_data=$(jq -r --argjson num "$task_number" \
'.active_projects[] | select(.project_number == $num)' \
specs/state.json)
# Validate exists
if [ -z "$task_data" ]; then
return error "Task $task_number not found"
fi
# Extract fields
task_type=$(echo "$task_data" | jq -r '.task_type // "general"')
status=$(echo "$task_data" | jq -r '.status')
project_name=$(echo "$task_data" | jq -r '.project_name')
# Validate task_type (accept both "lean" and "lean4")
if [ "$task_type" != "lean" ] && [ "$task_type" != "lean4" ]; then
return error "Task $task_number is not a Lean task (task_type: $task_type)"
fi
Task Complexity Warning (GATE IN)
After identifying the plan file, extract the estimated effort and warn if total exceeds 20 hours:
plan_file="specs/${padded_num}_${project_name}/plans/$(ls specs/${padded_num}_${project_name}/plans/ 2>/dev/null | tail -1)"
if [ -f "$plan_file" ]; then
effort_line=$(grep -i "Effort\|estimate\|hours\?" "$plan_file" 2>/dev/null | head -5)
effort_hours=$(echo "$effort_line" | grep -oP '\d+(?=\s*h(our)?s?)' | head -1)
if [ -n "$effort_hours" ] && [ "$effort_hours" -gt 20 ] 2>/dev/null; then
echo "WARNING: Task complexity exceeds 20 hours estimated effort (${effort_hours}h)."
echo " Consider using /team-implement or breaking into smaller phases."
echo " Proceeding with single-agent implementation."
fi
fi
This warning is non-blocking. If effort hours cannot be parsed, no warning is emitted.
Stage 2: Preflight Status Update
Update task status to "implementing" BEFORE invoking subagent.
Update state.json:
jq --arg ts "$(date -u +%Y-%m-%dT%H:%M:%SZ)" \
--arg status "implementing" \
--arg sid "$session_id" \
'(.active_projects[] | select(.project_number == '$task_number')) |= . + {
status: $status,
last_updated: $ts,
session_id: $sid,
started: $ts
}' specs/state.json > specs/tmp/state.json && mv specs/tmp/state.json specs/state.json
Update TODO.md: Use Edit tool to change status marker from [PLANNED] to [IMPLEMENTING].
Stage 3: Prepare Delegation Context
Prepare delegation context for the subagent:
{
"session_id": "sess_{timestamp}_{random}",
"delegation_depth": 1,
"delegation_path": ["orchestrator", "implement", "skill-lean-implementation"],
"timeout": 7200,
"task_context": {
"task_number": N,
"task_name": "{project_name}",
"description": "{description}",
"task_type": "${task_type}"
},
"plan_path": "specs/{N}_{SLUG}/plans/implementation-{NNN}.md",
"metadata_file_path": "specs/{N}_{SLUG}/.return-meta.json"
}
Stage 4: Invoke Subagent
CRITICAL: You MUST use the Task tool to spawn the subagent.
Required Tool Invocation:
Tool: Task (NOT Skill)
Parameters:
- subagent_type: "lean-implementation-agent"
- prompt: [Include task_context, delegation_context, plan_path, metadata_file_path]
- description: "Execute Lean implementation for task {N}"
DO NOT use Skill(lean-implementation-agent) - this will FAIL.
The subagent will:
- Load implementation context files (MCP tools guide, tactic patterns)
- Parse plan and find resume point
- Execute phases sequentially using lean-lsp MCP tools
- Verify proofs with
lean_goalandlake build - Create implementation summary
- Write metadata to
specs/{N}_{SLUG}/.return-meta.json - Return a brief text summary (NOT JSON)
Stage 5: Parse Subagent Return (Read Metadata File)
After subagent returns, read the metadata file:
metadata_file="specs/${padded_num}_${project_name}/.return-meta.json"
if [ -f "$metadata_file" ] && jq empty "$metadata_file" 2>/dev/null; then
status=$(jq -r '.status' "$metadata_file")
artifact_path=$(jq -r '.artifacts[0].path // ""' "$metadata_file")
phases_completed=$(jq -r '.metadata.phases_completed // 0' "$metadata_file")
phases_total=$(jq -r '.metadata.phases_total // 0' "$metadata_file")
else
echo "Error: Invalid or missing metadata file"
status="failed"
fi
Stage 6: Zero-Debt Verification Gate (MANDATORY)
CRITICAL: Before proceeding to status update, verify the zero-debt completion gate.
If status from metadata is "implemented":
# Check for sorries in modified files
sorry_count=$(grep -r "\bsorry\b" Theories/ 2>/dev/null | grep -v "^[[:space:]]*--" | wc -l)
# Check for vacuous definitions (semantically equivalent to sorry)
vacuous_count=$(grep -rn "^\s*\(noncomputable \)\?\(def\|theorem\|lemma\|instance\).*:= \(True\|Unit\|trivial\|Trivial\)\s*$" Theories/ 2>/dev/null | wc -l)
if [ "$vacuous_count" -gt 0 ]; then
echo " vacuous_count=$vacuous_count (def/theorem/lemma/instance := True|Unit|trivial|Trivial)"
fi
# Verify build passes
if ! lake build 2>/dev/null; then
build_failed=true
fi
if [ "$sorry_count" -gt 0 ] || [ "$vacuous_count" -gt 0 ] || [ "$build_failed" = true ]; then
echo "Zero-debt gate FAILED"
status="partial"
fi
Stage 6b: Plan Compliance Spot-Check (MANDATORY)
After the Zero-Debt gate passes, verify that plan-declared deliverables exist and that no replacement function delegates to the function it purports to replace.
This stage only runs if status from metadata is "implemented". If the agent returned "partial" or "failed", skip Stage 6b and proceed to Stage 7.
Step 1: Extract plan goal names
# plan_file is available from Stage 1 (GATE IN complexity warning)
if [ ! -f "$plan_file" ]; then
echo "WARNING: Plan file not found — skipping compliance check"
compliance_check="skipped"
else
# Extract backtick-wrapped identifiers from **Goals**: section
goal_names=$(sed -n '/^\*\*Goals\*\*:/,/^\*\*[^G]/p' "$plan_file" \
| grep -oP '`[a-zA-Z_][a-zA-Z0-9_'"'"']*`' \
| tr -d '`' | sort -u)
if [ -z "$goal_names" ]; then
echo "INFO: No goal names found in plan **Goals**: section — skipping compliance check"
compliance_check="skipped"
else
compliance_failed=false
# Step 2: Deliverable existence check
for name in $goal_names; do
if grep -rq "^\(noncomputable \)\?\(theorem\|def\|lemma\|instance\) $name\b" Theories/ 2>/dev/null; then
echo " [OK] $name — found in Theories/"
else
echo " [MISSING] $name — not found in Theories/ (plan deliverable absent)"
compliance_failed=true
fi
done
# Step 3: Delivery integrity check
replacement_targets=$(grep -oP '(?:replacement for|replaces|bypasses|supersedes)\s+`[a-zA-Z_][a-zA-Z0-9_'"'"']*`' "$plan_file" 2>/dev/null \
| grep -oP '`[a-zA-Z_][a-zA-Z0-9_'"'"']*`' | tr -d '`')
for replaced in $replacement_targets; do
for new_name in $goal_names; do
new_file=$(grep -rl "^\(noncomputable \)\?\(theorem\|def\|lemma\|instance\) $new_name\b" Theories/ 2>/dev/null | head -1)
if [ -n "$new_file" ] && grep -q "\b${replaced}\b" "$new_file"; then
echo " [INTEGRITY FAIL] $new_name references $replaced in $new_file"
echo " Plan declared $new_name as replacement for $replaced, but implementation delegates to it."
compliance_failed=true
fi
done
done
if [ "$compliance_failed" = true ]; then
echo "Plan compliance spot-check FAILED"
compliance_check="failed"
status="partial"
else
echo "Plan compliance spot-check PASSED"
compliance_check="passed"
fi
fi
fi
Step 4: Record compliance result in metadata
The compliance_check value ("passed", "failed", or "skipped") must be included in the .return-meta.json update passed to the GATE OUT stage. Update the metadata read logic to propagate this field.
Stage 7: Update Task Status (Postflight)
If status is "implemented" (verified by Stage 6):
Update state.json to "completed":
jq --arg ts "$(date -u +%Y-%m-%dT%H:%M:%SZ)" \
--arg status "completed" \
'(.active_projects[] | select(.project_number == '$task_number')) |= . + {
status: $status,
last_updated: $ts,
completed: $ts
}' specs/state.json > specs/tmp/state.json && mv specs/tmp/state.json specs/state.json
Update TODO.md: Change status marker from [IMPLEMENTING] to [COMPLETED].
If status is "partial":
Keep status as "implementing" but update resume point.
TODO.md stays as [IMPLEMENTING].
Stage 8: Link Artifacts
Add summary artifact to state.json.
if [ -n "$summary_artifact_path" ]; then
jq --arg path "$summary_artifact_path" \
--arg summary "$summary_artifact_summary" \
'(.active_projects[] | select(.project_number == '$task_number')).artifacts += [{"path": $path, "type": "summary", "summary": $summary}]' \
specs/state.json > specs/tmp/state.json && mv specs/tmp/state.json specs/state.json
fi
Stage 9: Git Commit
Commit changes with session ID. If the subagent already created per-phase commits (Phase Checkpoint Protocol), skip the batch commit to avoid redundant history:
# Check if per-phase commits already exist from subagent Phase Checkpoint Protocol
if git log --oneline -10 | grep -q "phase [0-9]\+:"; then
echo "Per-phase commits detected — skipping batch commit."
echo "Phase commits already capture implementation history."
else
# No per-phase commits found — create a batch commit
git add \
"Theories/" \
"specs/${padded_num}_${project_name}/summaries/" \
"specs/${padded_num}_${project_name}/plans/" \
"specs/TODO.md" \
"specs/state.json"
git commit -m "task ${task_number}: complete implementation
Session: ${session_id}
"
fi
# Always commit state updates if not already staged
git add "specs/TODO.md" "specs/state.json" 2>/dev/null || true
git diff --cached --quiet || git commit -m "task ${task_number}: update task state
Session: ${session_id}
"
Stage 10: Return Brief Summary
Return a brief text summary (NOT JSON). Example:
Lean implementation completed for task {N}:
- All {phases_total} phases executed, all proofs verified
- Lake build: Success
- Key theorems: {theorem names}
- Created summary at specs/{N}_{SLUG}/summaries/implementation-summary-{DATE}.md
- Status updated to [COMPLETED]
- Changes committed
Error Handling
Input Validation Errors
Return immediately with error message if task not found, wrong language, or status invalid.
Metadata File Missing
If subagent didn't write metadata file:
- Keep status as "implementing"
- Report error to user
Git Commit Failure
Non-blocking: Log failure but continue with success response.
Subagent Timeout
Return partial status if subagent times out (default 7200s). Keep status as "implementing" for resume.
Return Format
This skill returns a brief text summary (NOT JSON). The JSON metadata is written to the file and processed internally.
先判断是否适合
作者设计意图
作者的方法与取舍
边界和复核