📦 ai-retrospective — ai-retrospectiv...工具

# AI Collaboration Retrospective Post-session systematic review tool. Eight-dimension deep analysis drives a continuous improvement loop for AI-assisted development.

Core Principles

• Conversation context data source: complete conversation history 的 current 会话 已经 在...中 context — 否 external data fetching needed
• Progressive loading: Detailed evaluation criteria live 在...中 references/analysis_dimensions.md — 加载 在...上 demand
• Self-reflection 第一个: Examine AI's own shortcomings 之前 analyzing 用户-side improvements. 不 关于 criticizing 用户 — 's 关于 finding efficiency gains 在...中 "AI + Human" collaboration
• Quantify everything: Every finding 必须 reference specific conversation turns, wasted operations, 和 include counterfactual reasoning ("如果 X 有过 已 已完成, Y turns could 有 已 saved")
• Dig deep: Don't settle 对于 "否 findings." Complete self-check 列表 对于 每个 dimension 之前 declaring clean

Execution 模型

This skill is pure LLM instruction-driven — no scripts, no external dependencies. It works on any AI assistant that can:

• Access current conversation history
• 读取 reference files 从 skill's directory
• 写入 输出 files 到 workspace

Capability adaptation: workflow 下面 references file operations 和 memory updates. 如果 AI tool doesn't support specific capability, skip step 和 note 在...中 举报. analysis itself 仅 requires conversation context access.

Workflow (Six Steps)

Step 1: Conversation Review — Extract 键 Events + Tag Waste Points

Scan the entire conversation context and extract these key events into a timeline: | Event Type | Recognition Signal | |-----------|-------------------| | Tool invocations | Command execution, file reading/writing, web searches, code generation | | File changes | Files created, modified, or deleted | | Errors & fixes | Error messages, lint failures, debugging cycles | | Repeated modifications | Same file/feature modified multiple times, user providing multiple clarifications | | Decision points | Technology choices, architecture decisions, trade-offs | | Automation/plugin usage | Any skill, agent, plugin, or extension triggered during the session | | User clarifications | User adding context because the AI misunderstood intent | | Verification rounds | User providing test data/feedback, AI analyzing verification results | | AI misjudgments | AI providing wrong conclusions, missing critical issues, or jumping to premature conclusions | 过滤 rule: System initialization events (bootstrap files, identity setup, etc.) excluded 从 analysis. Critical step — Waste point tagging: After building the timeline, interrogate each event in reverse:

• Could step 有 已 avoided? 如果 something 有过 已 已完成 earlier, would step unnecessary?
• Could step 有 happened sooner? 做过 AI 延迟 something 应该 有 proactively 已完成?
• 做过 step duplicate prior work? 是 AI hand-writing logic could 有 已 reused?

Tag events where the answer is "yes" with [⚠ Optimizable] and record the reason. These tags are the core input for Step 2. Output format: Chronological event list with type labels and brief descriptions. Waste points tagged separately.

Step 2: Eight-Dimension Deep Analysis

Load references/analysis_dimensions.md for detailed evaluation criteria, self-check lists, and common patterns per dimension. Analyze conversation events dimension by dimension to identify improvement opportunities. Eight dimensions overview:

• AI Self-Reflection ⭐ — AI's mistakes, delayed reactions, missed judgments 在...中 会话 (highest priority, 必须 analyzed 第一个)
• Verification Strategy — 做过 AI proactively define verification criteria 和 expected outcomes, 或 passively wait 对于 用户 feedback?
• Automation Opportunities — Repetitive workflows 或 hand-written scripts could encapsulated 进入 reusable automations
• Existing Automation Tuning — 是 任何 existing automations/skills/templates used? 做过 它们 有 gaps, unclear instructions, 或 输出 issues?
• Tool Integration Opportunities — Operations would benefit 从 dedicated tool integrations, plugins, 或 API connections
• Knowledge Persistence — Preferences, conventions, 和 technical decisions 从 会话 应该 persisted 对于 future sessions
• Documentation Updates — Project docs, coding standards, 或 architecture notes 需要 updating
• Workflow Efficiency — Sequential steps could parallel, repeated labor, suboptimal tool choices

Analysis requirements (mandatory): For each dimension:

• Run 通过 dimension's self-check 列表 (defined 在...中 references/analysis_dimensions.md)
• 对于 findings, 输出: Specific 事件 reference (哪个 turn, 什么 operation) + Counterfactual reasoning (如果 X 有过 已 已完成, Y could saved) + Recommendation + Priority
• 仅 之后 所有 self-check items pass 可以 dimension declared "否 findings" 和 skipped

Step 3: Generate Retrospective 举报

Load assets/report_template.md for the report template. Fill the template with results from Step 1 and Step 2 to produce a complete Markdown retrospective report. 举报 保存 path: {workspace}/retrospectives/{topic}_retrospective.md Naming rules:

• {topic} uses 2-4 English words joined 由 hyphens, summarizing 会话's core task (e.g., multithread-scope-collection, 登录-flow-refactor)
• Multiple retrospectives 在...上 相同 topic: 如果 file 已经 exists, append 新的 举报 在 end (separated 由 --- 和 新的 日期 heading) — don't 创建 新的 file

If the retrospectives/ directory doesn't exist, create it first.

Note: The save path above is a sensible default. Adapt to your project's conventions if they differ.

Step 4: Display 满 Analysis 在...中 Conversation

complete analysis 必须 shown directly 在...中 conversation — don't 只是 输出 summary 和 point 到 file. file 归档; primary reading experience 在...中 conversation. Output content (show in full, no trimming):

• 会话 summary: One-sentence overview
• Efficiency score: Optimizable turns / 总计 turns
• 事件 timeline: Complete 表 带有 waste point tags
• 所有 dimension findings: 每个 带有 事件 reference, problem, counterfactual reasoning, recommendation ( core content — never abbreviate 或 归约)
• 待处理 action 列表 (如果 任何)
• 举报 归档 location

Format: Use Markdown tables and headings for clear structure. Better to be thorough than to cut valuable analysis.

Step 5: Automatic Execution — Knowledge Persistence

For items identified in the "Knowledge Persistence" dimension (Dimension 6), execute persistence operations available in your AI tool:

• 如果 tool supports persistent memory (e.g., memory APIs, memory files, .memory directories), 写入 新的 preferences/conventions directly
• 如果 tool supports project-level notes 或 配置, 更新 those
• 如果 tool 有 否 persistence mechanism, 列表 items 应该 persisted 和 recommend 用户 保存 them manually

Briefly state what was updated after each operation. Skip this step if no knowledge needs persisting.

Step 6: 待处理 Action 列表

For the following types of improvement suggestions, do not auto-execute — list them for user selection: | Action Type | Examples | |------------|---------| | Create new automation | Reusable workflow, script template, custom command | | Tune existing automation | Modify instructions, parameters, or trigger conditions | | Create/update project rules | Coding standards, review checklists, conventions | | Update project documentation | Architecture docs, API references, onboarding guides | | Create tool integration | Custom plugin, API connection, webhook | List format: Numbered list, each item includes "Action type + Specific content + Expected benefit." User can reply with numbers to select which actions to execute. If no pending actions, skip this step and state "No additional actions needed for this session."

Edge Cases

Very short sessions: 如果 conversation 仅 few turns 带有 simple content, 输出 brief summary 和 state " 会话 是 brief — 否 significant improvement opportunities identified." Don't force analysis. Compressed/summarized history: 如果 conversation history appears compressed 或 truncated, analyze based 在...上 可用 context 和 note 在...中 举报: "一些 conversation history 是 compressed; analysis based 在...上 visible context." Tool capability limitations: 如果 AI tool 正在 used lacks certain capabilities referenced 在...中 workflow (e.g., 否 file writing, 否 memory persistence), adapt gracefully — perform analysis steps possible 和 clearly note 任何 skipped steps 带有 reason.