readme-skill

What it does

The skill reads local AI tool data (Claude Code, Codex CLI, Kiro, Trae, Gemini Antigravity, Cursor) to build a Markdown developer profile and a validated SVG poster under ./output/. It renders either a default history view or a specified month/date range, with anonymized external sharing by default. It produces a desensitized developer README and a corresponding poster file in a user-specified language (Chinese by default; English if requested). The output is written locally and not uploaded.

How it works

• Step 1: Prepare environment, create an output directory, and establish a reporting window if the user requests a time-bounded view. The tool sets variables like REPORT_START, REPORT_END_EXCL, REPORT_LABEL, and REPORT_SLUG for windowed reports; otherwise WINDOW_REQUESTED stays 0.
• Step 2 — 读取 Claude Code 数据: It reads ~/.claude/stats-cache.json to extract total sessions, total messages, first session date, longest session, hourly counts, model usage by model, daily activity, and daily model tokens. It derives claude_tokens_spent, claude_cache_read, and cache_to_spent_ratio. If a window is requested, it filters dailyActivity and dailyModelTokens to the reporting window; modelUsage remains global. If details are missing for a field, it notes the limitation rather than inventing data.
• Step 2.2 Slash-command 热度: It reads ~/.claude/history.jsonl to compute top slash commands and totals, filtering by the reporting window if applicable.
• Step 2.3 项目分布: It lists top projects from ~/.claude/projects and uses the first JSONL entry to recover canonical real paths for consistent later analysis.
• Step 2.4 计划与 skill 自研: It analyzes plan directories and skill inventories under ~/.claude and ~/.codex, counting plans and listing plan titles, as well as counting installed/unbuilt skills, and noting if a project path implies a self-built skill.
• Step 2.5 配置深度: It reads ~/.claude/settings.json to count hooks, mcpServers, and permissions.defaultMode.
• Step 3 — 读取 Codex 数据: It uses a read-only sqlite3 connection to ~/.codex/state_5.sqlite and runs queries to aggregate threads by time, model, reasoning, working directories, hour-of-day heatmaps, dates, and titles/messages; also analyzes Codex global history in ~./codex/history.jsonl and applies time window filtering if requested. It then gathers Codex-related artifacts and counts, as well as global skill/automation/rules under ~/.codex.
• Step 3b — 读取 Kiro 数据: If present, it gathers Kiro totals from a read-only SQLite store in ~/.local/share/kiro-cli/data.sqlite3 and adapts to the actual schema to extract kiro_sessions, kiro_messages, token breakdowns, and per-date/hour activity. If time columns are unavailable, it reports all-time context only. It also reads sessions in ~/.kiro/sessions/cli and summarizes per-project cwd, as well as a small keyword corpus from the first lines of jsonl session events. It enumerates Kiro assets such as agents, skills, steering, prompts, and settings.
• Step 3b.3 Kiro infrastructure: It enumerates Kiro agents, skills, steering, prompts, and settings/mcp, and instructs to parse frontmatter in each skill using the same YAML parsing logic described for Claude/Codex skills.
• Throughout: All data handling is local/read-only, with anonymization by default unless the user asks for private/real names. The skill produces an output Markdown profile and a validated SVG poster under output/ in the requested language.

When to use it

Use when you want to generate a desensitized, shareable AI-native developer portrait and a monthly or custom-date-range report across multiple AI programming tools. Triggers are phrases like the ones listed in the frontmatter, enabling on-demand generation of the profile and poster locally.

What it can touch

• Local data sources: Claude Code, Codex CLI, Kiro, Trae, Gemini Antigravity, Cursor
• Output directory: ./output/ for the generated Markdown profile and SVG poster
• Tools/files explicitly named: ~/.claude/stats-cache.json, ~/.claude/history.jsonl, ~/.claude/projects/, ~/.claude/plans, ~/.claude/settings.json, ~/.codex/state_5.sqlite, ~/.codex/history.jsonl, ~/.codex/skills/, ~/.kiro/, ~/.local/share/kiro-cli/data.sqlite3, ~/.kiro/sessions/cli, ~/.kiro/agents/, ~/.kiro/skills/, ~/.kiro/steering/, ~/.kiro/prompts/

Caveats

• Desensitization is the default; real names are only shown if the user expressly requests private/real names.
• If certain fields are missing from sources, the report may show a dash or a note indicating all-time aggregation where windowed data is unavailable.
• The workflow relies on local data layout; if a source changes, some queries may need adaptation but the instruction emphasizes non-modification of scripts and using Read/Jq for parsing frontmatter and data.