Agent转换
- 作者仓库星标 435
- 作者更新于 实时读取
- 作者仓库 nvim
- 领域
- AI 智能
- 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-reviser
description: Thin wrapper that delegates plan revision to reviser-agent subagent. Invoke for /revise command.…
category: AI 智能
runtime: 无特殊运行时
---
# skill-reviser 输出预览
## PART A: 任务判断
- 适用问题:提示词、Agent 工作流、模型评估或自动化推理。
- 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“Context References / Trigger Conditions / Execution Flow”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提示词、Agent 工作流、模型评估或自动化推理,并标出输入、限制和预期结果。
- **02** 执行计划:优先按“Context References / Trigger Conditions / Execution Flow”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 `/revise`、`/plan` 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“Context References / Trigger Conditions / Execution Flow”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-reviser
description: Thin wrapper that delegates plan revision to reviser-agent subagent. Invoke for /revise command.…
category: AI 智能
source: benbrastmckie/nvim
---
# skill-reviser
## 什么时候使用
- 把 AI / Agent方向的常用动作沉淀成 Agent 可调用的技能 适合处理AI Agent、提示词、模型评估与自动化推理,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…
- 面向提示词、Agent 工作流、模型评估或自动化推理,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「Context References / Trigger Conditions / Execution Flow」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-reviser" {
输入层 -> 用户目标 + 目标文件 + 禁止范围 + 验收标准
上下文层 -> Context References / Trigger Conditions / Execution Flow
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
} Reviser Skill
Thin wrapper that delegates plan revision to reviser-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. This eliminates the "continue" prompt issue between skill return and orchestrator.
Context References
Reference (do not load eagerly):
- Path:
.claude/context/formats/return-metadata-file.md- Metadata file schema - Path:
.claude/context/patterns/postflight-control.md- Marker file protocol - Path:
.claude/context/patterns/file-metadata-exchange.md- File I/O helpers - Path:
.claude/context/patterns/jq-escaping-workarounds.md- jq escaping patterns (Issue #1132) - Path:
.claude/context/formats/plan-format.md- Plan file format specification
Note: This skill is a thin wrapper with internal postflight. Context is loaded by the delegated agent.
Trigger Conditions
This skill activates when:
/revisecommand is invoked- Task exists in state.json
Execution Flow
Stage 1: Input Validation
Validate required inputs:
task_number- Must be provided and exist in state.json
# 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 "$task_data" | jq -r '.project_name')
description=$(echo "$task_data" | jq -r '.description // ""')
No status-based ABORT rules. The skill works regardless of task status. Routing is determined by plan file existence, not status.
Stage 2: Preflight
No intermediate "revising" status is needed for revision. The task transitions directly to "planned" on success (via postflight). Skip preflight status update.
Rationale: Unlike /plan which sets "planning" as an intermediate status, /revise is lightweight enough that an intermediate status adds no value. The postflight script handles the final status update.
Stage 3: Create Postflight Marker
Create the marker file to prevent premature termination:
# Ensure task directory exists
padded_num=$(printf "%03d" "$task_number")
mkdir -p "specs/${padded_num}_${project_name}"
cat > "specs/${padded_num}_${project_name}/.postflight-pending" << EOF
{
"session_id": "${session_id}",
"skill": "skill-reviser",
"task_number": ${task_number},
"operation": "revise",
"reason": "Postflight pending: status update, artifact linking, git commit",
"created": "$(date -u +%Y-%m-%dT%H:%M:%SZ)",
"stop_hook_active": false
}
EOF
Stage 3a: Calculate Artifact Number
Read next_artifact_number from state.json and use (current-1) since revised plan stays in the same round:
# Read next_artifact_number from state.json
next_num=$(jq -r --argjson num "$task_number" \
'.active_projects[] | select(.project_number == $num) | .next_artifact_number // 1' \
specs/state.json)
# Plan uses (current - 1) to stay in the same round as research
# If next_artifact_number is 1 (no research yet), use 1
if [ "$next_num" -le 1 ]; then
artifact_number=1
else
artifact_number=$((next_num - 1))
fi
# Fallback for legacy tasks: count existing plan artifacts
if [ "$next_num" = "null" ] || [ -z "$next_num" ]; then
padded_num=$(printf "%03d" "$task_number")
count=$(ls "specs/${padded_num}_${project_name}/plans/"*[0-9][0-9]*.md 2>/dev/null | wc -l)
artifact_number=$((count + 1))
fi
artifact_padded=$(printf "%02d" "$artifact_number")
Note: Revised plan does NOT increment next_artifact_number. Only research advances the sequence.
Stage 4: Research Discovery
Discover existing plan and new research reports:
4a. Find existing plan:
padded_num=$(printf "%03d" "$task_number")
plan_dir="specs/${padded_num}_${project_name}/plans"
existing_plan=$(ls -1t "$plan_dir"/*.md 2>/dev/null | head -1)
4b. Find new research reports:
Use BOTH reports_integrated from state.json and file modification time comparison:
reports_dir="specs/${padded_num}_${project_name}/reports"
new_reports=()
if [ -n "$existing_plan" ]; then
plan_mtime=$(stat -c %Y "$existing_plan")
# Check reports_integrated from state.json (primary)
integrated=$(jq -r --argjson num "$task_number" \
'.active_projects[] | select(.project_number == $num) | .plan_metadata.reports_integrated // [] | .[]' \
specs/state.json 2>/dev/null)
for report in "$reports_dir"/*.md; do
if [ -f "$report" ]; then
report_basename=$(basename "$report")
# Check if already integrated
if echo "$integrated" | grep -qF "$report_basename"; then
continue
fi
# Fallback: check modification time
report_mtime=$(stat -c %Y "$report")
if [ "$report_mtime" -gt "$plan_mtime" ]; then
new_reports+=("$report")
fi
fi
done
else
# No plan exists -- all reports are "new"
for report in "$reports_dir"/*.md; do
if [ -f "$report" ]; then
new_reports+=("$report")
fi
done
fi
Stage 4b: Read and Inject Format Specification
Read the plan format file and prepare it for injection into the subagent prompt. This ensures the subagent always has the full format specification in its context.
format_content=$(cat .claude/context/formats/plan-format.md)
The format content will be included as a delimited section in the Stage 5 prompt.
Stage 5: Prepare Delegation Context and Invoke Subagent
CRITICAL: You MUST use the Agent tool to spawn the subagent.
Prepare delegation context:
{
"session_id": "sess_{timestamp}_{random}",
"delegation_depth": 1,
"delegation_path": ["orchestrator", "revise", "skill-reviser"],
"timeout": 1800,
"task_context": {
"task_number": N,
"task_name": "{project_name}",
"description": "{description}",
"task_type": "{task_type}"
},
"artifact_number": "{artifact_padded from Stage 3a}",
"existing_plan_path": "{path to existing plan or null}",
"new_research_paths": ["{path to report1}", "{path to report2}"],
"revision_reason": "{optional user reason}",
"roadmap_path": "specs/ROADMAP.md",
"metadata_file_path": "specs/{NNN}_{SLUG}/.return-meta.json"
}
Required Tool Invocation:
Tool: Agent (NOT Skill, NOT Plan)
Parameters:
- subagent_type: "reviser-agent"
- prompt: [Include task_context, delegation_context, existing_plan_path, new_research_paths,
revision_reason, metadata_file_path,
AND the format specification from Stage 4b as shown below]
- description: "Execute plan revision for task {N}"
Format Injection: Include the format specification from Stage 4b in the prompt as a clearly-delimited section:
<artifact-format-specification>
## CRITICAL: Plan Format Requirements
You MUST follow this format specification exactly when writing the plan artifact.
Non-compliance will be caught by postflight validation.
{format_content from Stage 4b}
</artifact-format-specification>
Place this section AFTER the delegation context JSON and BEFORE any other instructions.
DO NOT use Skill(reviser-agent) - this will FAIL.
The subagent will:
- Load revision context files
- Determine revision mode (plan revision or description update)
- Load existing plan and new research reports
- Synthesize revised plan or update description
- Write metadata to
specs/{NNN}_{SLUG}/.return-meta.json - Return a brief text summary (NOT JSON)
Stage 5b: Self-Execution Fallback
CRITICAL: If you performed the work above WITHOUT using the Agent tool (i.e., you read files,
wrote artifacts, or updated metadata directly instead of spawning a subagent), you MUST write a
.return-meta.json file now before proceeding to postflight. Use the schema from
return-metadata-file.md with the appropriate status value for this operation.
If you DID use the Agent tool, skip this stage -- the subagent already wrote the metadata.
Postflight (ALWAYS EXECUTE)
The following stages MUST execute after work is complete, whether the work was done by a subagent or inline (Stage 5b). Do NOT skip these stages for any reason.
Stage 6: Read Metadata File
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")
artifact_type=$(jq -r '.artifacts[0].type // ""' "$metadata_file")
artifact_summary=$(jq -r '.artifacts[0].summary // ""' "$metadata_file")
new_description=$(jq -r '.metadata.new_description // ""' "$metadata_file")
else
echo "Error: Invalid or missing metadata file"
status="failed"
fi
Stage 6a: Validate Artifact Content
If subagent status is "planned" and artifact_path is non-empty, validate the plan artifact against format requirements. This is non-blocking -- warnings are logged but do not prevent postflight from completing.
if [ "$status" = "planned" ] && [ -n "$artifact_path" ] && [ -f "$artifact_path" ]; then
echo "Validating plan artifact..."
if ! bash .claude/scripts/validate-artifact.sh "$artifact_path" plan --fix; then
echo "WARNING: Plan artifact has format issues (non-blocking). Review output above."
fi
fi
Stage 7: Postflight Status Update
For Plan Revision (status == "planned"):
Update task status to "planned" using the centralized script:
bash .claude/scripts/update-task-status.sh postflight $task_number plan $session_id
If the script exits non-zero, log error but continue (status update is best-effort for revise).
For Description Update (status == "description_updated"):
Update state.json description and TODO.md directly:
jq --arg ts "$(date -u +%Y-%m-%dT%H:%M:%SZ)" --arg desc "$new_description" \
--argjson num "$task_number" \
'(.active_projects[] | select(.project_number == $num)) |= . + {
description: $desc,
last_updated: $ts
}' specs/state.json > specs/tmp/state.json && \
mv specs/tmp/state.json specs/state.json
Then use Edit tool to update the description in TODO.md.
On partial/failed: Keep status unchanged (do not call the script).
Stage 8: Artifact Linking (Plan Revision Only)
Add the new plan artifact to state.json.
IMPORTANT: Use two-step jq pattern to avoid Issue #1132 escaping bug. See jq-escaping-workarounds.md.
if [ -n "$artifact_path" ]; then
# Step 1: Filter out existing plan artifacts (use "| not" pattern to avoid != escaping - Issue #1132)
jq --argjson num "$task_number" \
'(.active_projects[] | select(.project_number == $num)).artifacts =
[(.active_projects[] | select(.project_number == $num)).artifacts // [] | .[] | select(.type == "plan" | not)]' \
specs/state.json > specs/tmp/state.json && mv specs/tmp/state.json specs/state.json
# Step 2: Add new plan artifact
jq --argjson num "$task_number" \
--arg path "$artifact_path" \
'(.active_projects[] | select(.project_number == $num)).artifacts += [{"path": $path, "type": "plan"}]' \
specs/state.json > specs/tmp/state.json && mv specs/tmp/state.json specs/state.json
fi
Update TODO.md: Link artifact using the automated script:
bash .claude/scripts/link-artifact-todo.sh $task_number '**Plan**' '**Description**' "$artifact_path"
If the script exits non-zero, log a warning but continue (linking errors are non-blocking).
Stage 8a: Lifecycle TTS Notification
Fire TTS and WezTerm tab coloring after artifact linking is complete:
lifecycle_script=".claude/scripts/lifecycle-notify.sh"
if [ -f "$lifecycle_script" ]; then
bash "$lifecycle_script" "$STATE_STATUS" &
fi
Non-blocking: called in background after artifacts are linked. Speaks "Tab N STATUS" (e.g., "Tab 3 planned") to announce the lifecycle transition.
Stage 9: Git Commit
For Plan Revision:
git add -A
git commit -m "$(cat <<'EOF'
task {N}: revise plan (v{NEW_VERSION})
Session: {session_id}
EOF
)"
For Description Update:
git add -A
git commit -m "$(cat <<'EOF'
task {N}: revise description
Session: {session_id}
EOF
)"
Commit failure is non-blocking (log and continue).
Stage 10: Cleanup
Remove marker and metadata files:
rm -f "specs/${padded_num}_${project_name}/.postflight-pending"
rm -f "specs/${padded_num}_${project_name}/.postflight-loop-guard"
rm -f "specs/${padded_num}_${project_name}/.return-meta.json"
Stage 11: Return Brief Summary
Return a brief text summary (NOT JSON). Example:
Plan Revision:
Plan revised for task {N}:
- Preserved {X} completed phases, revised {Y} phases
- Integrated {Z} new research reports
- Created revised plan at specs/{NNN}_{SLUG}/plans/MM_{short-slug}.md
- Status updated to [PLANNED]
- Changes committed with session {session_id}
Description Update:
Description updated for task {N}:
- Previous: {old_description}
- New: {new_description}
- Status: [{current_status}] (unchanged)
- Changes committed with session {session_id}
Error Handling
Input Validation Errors
Return immediately with error message if task not found.
Metadata File Missing
If subagent didn't write metadata file:
- Keep status unchanged
- Do not cleanup postflight marker
- Report error to user
Git Commit Failure
Non-blocking: Log failure but continue with success response.
jq Parse Failure
If jq commands fail with INVALID_CHARACTER or syntax error (Issue #1132):
- Log to errors.json
- Retry with two-step pattern (already implemented in Stage 8)
Subagent Timeout
Return partial status if subagent times out (default 1800s). Keep status unchanged for resume.
MUST NOT (Postflight Boundary)
After the agent returns, this skill MUST NOT:
- Edit source files - All revision work is done by agent
- Run build/test commands - Verification is done by agent
- Use research tools - Web/codebase search is for agent use only
- Analyze task requirements - Analysis is agent work
- Write plan files - Artifact creation is agent work
The postflight phase is LIMITED TO:
- Reading agent metadata file
- Updating status via
update-task-status.sh(handles state.json + TODO.md atomically) - Linking artifacts in state.json
- Updating description in state.json/TODO.md (description update path only)
- Git commit
- Cleanup of temp/marker files
Reference: @.claude/context/standards/postflight-tool-restrictions.md
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