convez.ai
SkillAutomaçãoIntermediário
liks79

Claude Skills — Comandos e Agentes para Produtividade

liks79/claude-skills

29 comandos + 4 skills + agente para pesquisa, carreira e automação

Ver no GitHub
InicialPluginScore · de 0 a 100

Coleção de 29 comandos slash, 4 skills e 1 agente especializado para Claude Code, cobrindo pesquisa, carreira, finanças, automação e produtividade. Inclui scripts bundled, templates de relatórios e integração com ferramentas como Gemini, PowerPoint e Gmail.

Instalação

$npx convez-ai liks79/claude-skills

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
2
Forks
1
Acompanhando
0
Saúde
42%
Último commit
há 4 meses
Versão
v1.5.8
Issues abertas
0
Contribuidores
2

Casos de uso

  • Automação de pesquisa e análise de mercado com templates estruturados
  • Gestão de carreira com agente dedicado e análise de oportunidades
  • Análise financeira e de portfólio com skill invest
  • Automação de workflows de escritório (PowerPoint, Gmail, relatórios)
  • Integração com APIs (Gemini, GitHub, ferramentas de job search)

Veredito

Ecossistema robusto e bem arquitetado para power users do Claude Code que precisam de automação cross-domain. Documentação clara, 29 comandos funcionais e templates prontos tornam viável, mas curva de instalação (marketplace em duas etapas) e dependência de Python/bash limitam para iniciantes.

README

A tradução para o português deste README ainda está na fila.

Claude Skills

A curated collection of slash commands, skills, and agents for Claude Code — covering research workflows, career development, productivity tools, and automation.

License: MIT Commands Skills Version


Architecture

flowchart TD
    User(["👤 User"])

    subgraph Plugin["📦  claude-skills  Plugin"]
        subgraph Entry["Entry Points"]
            CMDS["📋 Commands  ×29\n━━━━━━━━━━━━━━━━━━\nresearch · career · git\nAI · wiki · finance\nreal-estate · utils"]
            SKILLS["🧠 Skills  ×4\n━━━━━━━━━━━━━━━━━━\ngemini  ·  pptx  ·  invest\nnewsletter"]
        end

        AGENT["🤖 Agent\n━━━━━━━━━━━━━━━━━━\ncareer-researcher"]

        subgraph Assets["Bundled Assets"]
            SCRIPTS["⚙️ Scripts  ×10\n━━━━━━━━━━━━━━━━━━\napt_report · apt_watch\nemail_summary · presign\ngenerate_image · github-urls\nrecent · gmail_fetch_newsletter\nfindjob (package) · scan"]
            TMPLS["📄 Templates  ×7\n━━━━━━━━━━━━━━━━━━\nT1 Executive Brief\nT2 Tech Deep-Dive\nT3 Market Analysis\nT4 Comparative Eval\nT5 Strategic Roadmap\nT6 Investment Report\nT7 Newsletter Digest"]
        end
    end

    User -->|"invoke /command"| CMDS
    User -. "context keyword\n(slides · ask Gemini · …)" .-> SKILLS

    CMDS -->|"career topic or /career-*\n→ delegates"| AGENT
    AGENT -. "reads skill definitions\n(career commands)" .-> CMDS
    CMDS -->|"shell exec via\nplugin cache path"| SCRIPTS
    CMDS -->|"research commands\nread scaffold"| TMPLS
    SKILLS -->|"pptx office\nworkflow"| SCRIPTS

What's Included

TypeCountContents
Slash Commands29Research, career, git, AI tools, productivity, finance, real estate, newsletter, job finder, vault scan, tag management
Skills4gemini (Gemini CLI wrapper), pptx (PowerPoint toolkit), invest (portfolio analytics), newsletter (Gmail newsletter curation)
Agent1career-researcher (dedicated career research sub-agent)
Scripts10Python/shell scripts bundled with commands (findjob is a full Python package with 11 company parsers; scan is a bash+Python multi-file package)
Templates7T1–T7 research and curation report templates

Installation

Note: Installation requires two steps — first register the marketplace, then install the plugin. Running claude plugin install liks79/claude-skills directly will fail with "not found in any configured marketplace" because liks79/claude-skills is a GitHub repo path, not a marketplace plugin identifier.

Option 1 — Via Marketplace slash commands (Recommended)

/plugin marketplace add liks79/claude-skills
/plugin install claude-skills@liks79-skills

Option 2 — CLI (two steps)

# Step 1: Register the marketplace (required before install)
claude plugin marketplace add liks79/claude-skills

# Step 2: Install the plugin using the marketplace plugin identifier
claude plugin install claude-skills@liks79-skills --scope user

Option 3 — Manual (settings.json)

Add the following to ~/.claude/settings.json:

{
  "extraKnownMarketplaces": {
    "liks79-skills": {
      "source": {
        "source": "github",
        "repo": "liks79/claude-skills"
      }
    }
  },
  "enabledPlugins": {
    "claude-skills@liks79-skills": true
  }
}

Commands

Research & Knowledge Management

CommandDescription
/claude-skills:new-research <topic>Create a structured research note. Auto-selects one of five templates (T1–T5) based on topic keywords. Delegates career topics to the career-researcher agent.
/claude-skills:apply-research-template <file> [Template N] [depth]Restructure an existing markdown file into a research template. Supports --inplace to overwrite the original.
/claude-skills:scan [options]Scan the vault and build/update a research index. Extracts YAML frontmatter, titles, tags, and dates from all Markdown files. Generates a browsable index with year × month matrix and tag cloud. Incremental cache keeps subsequent runs fast. Tested and verified with Quartz v5.0+.
/claude-skills:add-tags [options]Scan an Obsidian vault and assign tags to untagged markdown files using Claude Code natively (no API key required). Builds and maintains a Tag_Dictionary.md with tag statistics and untagged file list. Incremental cache with surgical --paths rescan for fast updates. Quartz v5 compatible.
/claude-skills:wiki-ingest <file-or-url>Ingest a file, web URL, or YouTube video into an LLM-readable wiki. Extracts concepts and entities, creates cross-linked pages under wiki/compiled/.
/claude-skills:wiki-query <question>Search the local wiki and synthesize an answer with [[wikilink]] citations. Optionally saves the result as a new synthesis page.
/claude-skills:wiki-lintAudit wiki health: broken links, orphaned pages, missing frontmatter, stale entries (>90 days). Generates a report with action items.

Career Development

CommandDescription
/claude-skills:findjob [--output-dir PATH] [--db-path PATH]Scan 11 company career sites (AWS, Google, Microsoft, Anthropic, OpenAI, Databricks, Datadog, Cloudflare, Palantir, Redis, Coupang) for openings matching your configured positions and locations. Tracks changes via SQLite and generates a ranked Markdown report with new/removed positions delta.
/claude-skills:findjob listDisplay current configuration — wanted locations, positions, score threshold, and a table of all registered companies with their location filters. No network requests made.
/claude-skills:career-company-analysis <company>Web research on a company's tech stack, culture, interview process, and compensation. Saves a structured report to career/companies/.
/claude-skills:career-job-analysis <URL-or-text>Analyze a job posting. Extracts requirements, performs gap analysis against your background, and lists resume keywords.
/claude-skills:career-interview-prep <company> <role>Generate a structured interview prep guide covering coding, system design, behavioral, and technical deep-dive questions.
/claude-skills:career-salary-research <role> [region] [years]Research market salary data from Blind, LinkedIn, Levels.fyi, and job boards. Produces a distribution table by experience level.
/claude-skills:career-to-pptx <md-path>Convert a career markdown file (company analysis, job analysis, etc.) into a PowerPoint presentation using python-pptx.

Git & GitHub

CommandDescription
/claude-skills:ship [hint]Full git workflow: assess changes → create claude/* branch → stage → commit (Conventional Commits) → push → open PR with gh.
/claude-skills:github-urls [N] [--pr N|<sha>]Print GitHub URLs for the N most recently changed files. Use --pr N for files changed in a specific PR, or pass a commit SHA.
/claude-skills:grass-tracker [username]Show GitHub contribution graph status using grass-tracker. Falls back to basic stats via gh api if the CLI is not installed.

AI Tools

CommandDescription
/claude-skills:gemini <prompt> [--model] [--file] [--diff] [--summary]Run a prompt through Google Gemini CLI. Supports code review (--diff), file summarization (--summary), and model selection.
/claude-skills:image-gen <prompt> [--output] [--model]Generate images via Google Gemini API. Supports NanoBanana, NanoBanana 2 (default), NanoBanana Pro, Imagen 4, and Imagen 4 Fast models.

Productivity

CommandDescription
/claude-skills:email-summary [days]Fetch and classify Gmail messages by importance: 🔴 urgent, 🟡 important, 🔵 informational, ⚪ ads. Default: last 7 days.
/claude-skills:email-archive [N] [--dry-run]Fetch unread inbox messages, assign labels via AI (newsletter, career, finance, security, etc.), and archive. Default: 50 messages. Use --dry-run to preview without changes.
/claude-skills:newsletter --label <name|id> [--days N]Fetch Gmail messages from a label, classify by topic (AI/BigTech/Startup/Tools), and generate a premium T7 intelligence digest report. Saves to notes/newsletters/. Default: last 7 days (max 90).
/claude-skills:cal <event>Create or view Google Calendar events using natural language input (KST timezone). Supports Google Meet links and attendees.
/claude-skills:presign <file> [hours]Upload a file to Cloudflare R2 or AWS S3 and return a presigned URL. Default expiry: 24 hours.
/claude-skills:recent [N]List the N most recently modified files in the current directory tree. Default: 10.

Finance & Investment

CommandDescription
/invest [sheet_url]Read a Google Sheets investment portfolio via GWS CLI, fetch live market data, and generate a premium T6-template report covering allocation, performance, market intelligence, risk matrix, and action plan. Saves to reports/finance/.
/share [md_path] [hours]Convert the most recent research .md to PDF and upload it to Cloudflare R2, returning a presigned URL (and optionally a Quartz URL). Defaults to the most recently modified file under notes/ or reports/.

Korea Real Estate

CommandDescription
/claude-skills:apt <region> [--months N] [--type] [--forecast N] [--pdf]Generate an apartment price report for Seoul/metropolitan districts using MOLIT official transaction data (data.go.kr). Includes trend charts and 6-month forecast.
/claude-skills:apt-watch <complex-number|URL> [--name] [--location] [--type] [--no-db] [--pdf]Track active listings for a specific apartment complex on Naver Real Estate. Accepts a complex number or full Naver URL. Detects new and removed listings via SQLite snapshot comparison. Use --no-db for a one-shot snapshot without writing to the change-tracking DB.

Meta

CommandDescription
/claude-skills:cmdsList all commands provided by this plugin, grouped by category.

/add-tags — Obsidian Vault Tag Manager

No API key required — Claude Code handles all tagging via its native Read/Edit tools. No Anthropic API calls. Quartz v5 compatible: uses standard [title](path) Markdown links (not wikilinks).

Scans an Obsidian vault and assigns tags to untagged Markdown files using Claude Code natively. Outputs a Tag_Dictionary.md with a tag statistics table and a list of untagged files.

Key features:

  • No API key — Claude Code's Read/Edit tools process files directly (no external API)
  • Incremental cache — only re-scans changed files; repeated runs stay fast
  • Surgical --paths rescan — after tagging N files, re-reads only those N files O(N) instead of O(total)
  • Quartz v5 compatible — [tag](tags/tagname) link format for tag page navigation

Usage

/add-tags                               → default scan + auto-assign tags (up to 50 files)
/add-tags --force                       → full rescan, ignore cache
/add-tags --no-assign                   → scan + build Tag Dictionary only (skip assignment)
/add-tags --delta                       → fast scan: only files newer than cache
/add-tags --paths "a.md,b.md"           → surgical rescan of specific files only
/add-tags --max-files 20                → limit files processed per run
/add-tags --output-dir notes            → override Tag Dictionary output directory
/add-tags --include-dirs 10_PARA,notes  → restrict scan to specific directories (default: all)

Scan Mode Reference

ModeWhen to useSpeed
--paths "f1,f2"Re-scan exactly the N files you just tagged⚡ Fastest
(auto, cache exists)Default — delta scan via find -newer cache🚀 Fast
--forceFull rebuild or purge entries for deleted files🐢 Slow
(auto, first run)No cache yet — full walk to initialize🐢 Slow

Environment Variables

Configure in ~/.claude/settings.local.json → "env" block:

VariableDefaultDescription
ADD_TAGS_OUTPUT_DIR00_INBOXOutput directory for Tag_Dictionary.md
ADD_TAGS_FILENAMETag_Dictionary.mdOutput filename
ADD_TAGS_CACHE_DIR<script_dir>/.cacheTag cache directory

Plugin configuration (installed via plugin)

// ~/.claude/settings.local.json
{
  "env": {
    "BASE_DIR":           "/absolute/path/to/your/vault",
    "ADD_TAGS_OUTPUT_DIR": "00_INBOX",
    "ADD_TAGS_FILENAME":   "Tag_Dictionary.md",
    "ADD_TAGS_CACHE_DIR":  "/absolute/path/to/your/vault/.cache/add-tags"
  }
}

ADD_TAGS_CACHE_DIR recommendation: when installed as a plugin, scripts live inside the plugin cache directory. Set an absolute path inside your vault to keep the cache co-located with your content. If unset, the cache is written inside the plugin cache directory.

Output

  • $ADD_TAGS_OUTPUT_DIR/Tag_Dictionary.md — tag statistics + untagged file list (default: 00_INBOX/Tag_Dictionary.md)
  • $ADD_TAGS_CACHE_DIR/tag_cache.json — incremental tag cache (auto-created; do not commit)

Requirements

  • uv in PATH (used to run Python scripts)
  • The vault must be a git repository (used for repo root auto-detection)

/scan — Vault Research Index Builder

Compatibility: Tested and verified with Quartz v5.0+ (Debian 13, Node.js v26). Uses standard [title](path) Markdown links (not [[wikilinks]]) to avoid the | table-cell parsing issue in Quartz v5's remark pipeline. Tag links use #tag (OFM-processed) with a fallback to `code span` for emoji-prefixed tags incompatible with the v5 OFM regex (\p{L} only, no \p{Emoji}).

Scans a Markdown vault and generates a browsable research index. Extracts YAML frontmatter, first-H1 title fallback, tags, creation dates, and categories. Outputs a structured report with:

  • Year × Month matrix table — clickable cell counts linked to monthly sections
  • Tag cloud — top 60 tags sorted by frequency with occurrence counts
  • Document list — grouped by year → month, newest first
  • Undated files — separate section for files without a created: date

An incremental cache (meta_cache.json) means repeated /scan calls only re-parse changed files, keeping the command fast on large vaults.

Usage

/scan                                    → 00_INBOX/index.md (all .md in repo)
/scan --force                            → Full rescan, ignore cache
/scan -o notes/index                     → Output to notes/index/index.md
/scan -f my_index.md                     → Custom filename
/scan --dirs 20_AREAS/ai-ml              → Scan only ai-ml subdirectory
/scan --file-include md,mdx              → Include .md and .mdx files
/scan --file-exclude pdf,png             → Exclude by extension

Environment Variables

Configure in ~/.claude/settings.local.json → "env" block:

VariableDefaultDescription
SCAN_OUTPUT_DIR00_INBOXOutput directory (relative to repo root or absolute)
TARGET_FILENAMEindex.mdOutput filename
BASE_DIR(git auto-detect)Repository root override
SCAN_INCLUDE_DIRS.Comma-separated directories to scan
SCAN_FILE_INCLUDEmdExtensions to include (no dot; empty = all)
SCAN_FILE_EXCLUDE(empty)Extensions to exclude (no dot)
SCAN_EXCLUDE_PATTERNS(built-in)Additional path exclude patterns (Python regex, comma-separated)
SCAN_CACHE_DIR<repo>/.claude/scripts/scan/.cacheMetadata cache directory

Built-in path excludes (always applied): .git/, .claude/, .obsidian/, TEMPLATES/, WIKI/, Clippings/, Excalidraw/, node_modules/, 40_ARCHIVES/

Default SCAN_EXCLUDE_PATTERNS: findjob/,invest-report-,investment-report-,apt-watch-,newsletter-,resume-portfolio/ — skips files generated by other claude-skills commands (/findjob, /invest, /apt-watch, /newsletter). Override by setting SCAN_EXCLUDE_PATTERNS="" to disable, or append your own patterns.

Frontmatter title

The generated index uses the output directory name as the YAML title: (e.g., 00_INBOX). This is intentional: Quartz v5's explorer widget sorts pages by their frontmatter title, so using the folder name keeps the index file grouped with its siblings rather than appearing under "R" for "Research Index".

Plugin configuration (installed via plugin)

When installed as a plugin, BASE_DIR should point to your vault root (it defaults to the git root of whatever directory Claude Code is running in):

// ~/.claude/settings.local.json
{
  "env": {
    "BASE_DIR":              "/absolute/path/to/your/vault",
    "SCAN_OUTPUT_DIR":       "00_INBOX",
    "TARGET_FILENAME":       "index.md",
    "SCAN_EXCLUDE_PATTERNS": "findjob/,reports/"
  }
}

Output

  • $SCAN_OUTPUT_DIR/$TARGET_FILENAME — the generated Markdown index (default: 00_INBOX/index.md)
  • $SCAN_CACHE_DIR/meta_cache.json — incremental metadata cache (auto-created; do not commit)

Requirements

  • python3 or uv in PATH (no additional packages needed — stdlib only)
  • The vault must be a git repository (used for BASE_DIR auto-detection)

Skills

Skills are context-loaded automatically by Claude Code based on triggers.

gemini

Trigger: User mentions "ask Gemini", "use Gemini CLI", or invokes /claude-skills:gemini.

A wrapper around the Gemini CLI for non-interactive use inside Claude Code sessions. Covers single-shot Q&A, stdin piping, code review via git diff, and file summarization.

Model aliasModel IDNotes
(default)gemini-2.5-flashFast, general purpose
gemini-2.5-progemini-2.5-proHigh quality, complex reasoning
gemini-2.0-flashgemini-2.0-flashLightweight

invest

Trigger: /invest command, or context involving "investment portfolio", "Google Sheets portfolio", "GWS portfolio", or T6 investment report generation.

A portfolio analytics skill that reads holdings from Google Sheets via GWS CLI, computes per-owner and per-asset-class aggregates, flags concentration risks, and pre-computes all Mermaid chart numeric arrays for the T6 report template.

StepWhat it does
1–3Resolve spreadsheet ID, fetch sheet names, read all sheets in parallel
4Parse Holdings: owner, asset class, ticker, avg cost, current price, P&L
5Compute aggregates: portfolio summary, owner-level, asset class %, rankings, risk flags
6Load T6 template via plugin-cache path resolution
7Apply Mermaid rendering guidelines (English-only labels, chart type constraints)

/findjob — Job Opening Tracker

Scans 11 company career pages in one shot, scores results by position relevance, stores history in SQLite, and generates a ranked Markdown report.

Usage

/findjob                               # scan all companies and generate report
/findjob list                          # show current configuration — no network requests
/findjob --output-dir /path/to/dir     # override output directory
/findjob --db-path /path/to/jobs.db    # override SQLite DB path

/findjob list — Configuration Inspector

Displays the current configuration without running any scans:

Wanted Locations (4):
  • Seoul
  • South Korea
  • Seoul, South Korea
  • Korea

Wanted Positions (12):
  • Solutions Architect
  • Engineering Manager
  ...

Min Match Score: 0.40 — filters weak single-keyword overlaps

Registered Companies (11 total):
  #  Company                   Parser         Location Filter
  1  Amazon Web Services       aws            Seoul, South Korea
  2  Google                    google         Seoul, South Korea
  ...

To edit: commands/findjob.md → ## FindJob Configuration → findjob-config YAML block

Use this before running a full scan to verify your target positions and companies.

Configuration

Edit the embedded YAML block in commands/findjob.md (the block tagged # findjob-config):

wanted_locations:
  - "Seoul"
  - "South Korea"

wanted_positions:
  - "Solutions Architect"
  - "Engineering Manager"
  - "Staff Engineer"
  # ... add or remove as needed

min_match_score: 0.40   # 0.0 = all results, 0.4 = recommended, 1.0 = exact only

# Optional feature flags (both default to true)
enable_link_validation: true   # HTTP-check each job URL; remove 404/410/closed
enable_linkedin_search: true   # supplemental LinkedIn guest API search

companies:
  - name: "Anthropic"
    url: "https://www.anthropic.com/careers/..."
    parser: "anthropic"
    greenhouse_board: "anthropic"
  # ... add more companies

Supported ATS systems:

ATSCompaniesReliability
GreenhouseAnthropic, Datadog, Databricks, Cloudflare, Coupang, Redis✅ Stable JSON API
Ashby HQOpenAI✅ Stable JSON API
SmartRecruitersPalantir, Coupang✅ Stable JSON API
amazon.jobs APIAmazon Web Services✅ Stable JSON API
AF_initDataCallback scrapingGoogle✅ Embedded JS data extraction
Eightfold PCSX APIMicrosoft✅ Session cookie + internal API
LinkedIn guest API(supplemental)⚠️ Undocumented — may be blocked

Link validation: After each scan, all job URLs are HTTP-checked in parallel (20 concurrent, 12 s timeout). Jobs returning 404/410 or containing "job closed" phrases are removed from the report before saving to DB. Set enable_link_validation: false to skip (faster runs).

LinkedIn supplemental search: After the primary scan, each company is searched on LinkedIn's guest API. Unique results (not already found via the direct career page) are merged into the report. Set enable_linkedin_search: false to skip.

Outputs per run

  • career/job-search/findjob/findjob-YYYY-MM-DD.md — ranked report
  • career/job-search/findjob/jobs.db — SQLite database (run again to see delta)

Environment variables (optional)

VariablePurpose
FINDJOB_OUTPUT_DIROverride output directory
FINDJOB_DB_PATHOverride SQLite DB path
FINDJOB_CONFIG_FILEUse a custom config file instead of commands/findjob.md
BASE_DIRPrefix all output paths (same as other commands)

Plugin configuration (installed via plugin)

When installed as a plugin, the config file lives inside the plugin cache. To customize wanted_positions, wanted_locations, or add companies, set FINDJOB_CONFIG_FILE to point to your own copy:

// ~/.claude/settings.local.json
{
  "env": {
    "FINDJOB_CONFIG_FILE": "/absolute/path/to/my-findjob-config.md"
  }
}

Your config file only needs the findjob-config YAML block — it does not need to be a full copy of the command file. Example minimal config file:

```yaml
# findjob-config
wanted_locations:
  - "Seoul"
wanted_positions:
  - "Solutions Architect"
  - "Staff Engineer"
min_match_score: 0.40
companies:
  - name: "Anthropic"
    url: "https://www.anthropic.com/careers/jobs"
    parser: "anthropic"
    greenhouse_board: "anthropic"

---

### `newsletter`

**Trigger:** `/newsletter` command, or context involving "curate newsletters from Gmail", "newsletter digest", "Gmail label curation", or T7 newsletter report generation.

A Gmail newsletter curation skill that fetches messages from a specified label via gws CLI,
classifies them into topic buckets, and fills a T7 premium intelligence digest template.

| Step | What it does |
|------|-------------|
| 1–2 | Resolve script path (plugin cache), resolve label ID from name or ID |
| 3 | Fetch up to 40 messages from the label within the date window |
| 4 | Classify each message: AI & Engineering / Big Tech & Investment / Startup & Product / Tools & Infra / Other |
| 5 | Extract top keywords per category for Mermaid mindmap |
| 6 | Build Gantt milestones from event-dated messages |
| 7 | Format each item as analytical summary with filtered links |
| 8 | Compute aggregates: counts, top senders, key insights |
| 9 | Load T7 template via plugin-cache resolution and fill all placeholders |

**Output**: `notes/newsletters/newsletter-{label-slug}-YYYY-MM-DD.md`

**Requirements**: `gws` CLI authenticated, `uv` in PATH, `jq` installed.

---

### `pptx`

**Trigger:** Any `.pptx` file involved, or keywords like "deck", "slides", "presentation".

A complete PowerPoint toolkit with three workflows:

- **Read** — text extraction via `markitdown`, visual thumbnails, raw XML inspection
- **Edit** — unpack → manipulate XML → clean → repack, with subagent-parallel slide editing
- **Create** — from scratch using PptxGenJS with design guidance, color palettes, and typography rules

Includes full QA procedures: content validation, visual inspection via LibreOffice + `pdftoppm`, and a fix-and-verify loop.

---

## Agent

### `career-researcher`

A dedicated sub-agent for career research. Automatically delegated to when:
- `/claude-skills:new-research` detects career-related keywords (job search, interview, salary, etc.)
- Any `/claude-skills:career-*` command is invoked

**Scope:** Creates and updates files only within `career/`.

| Subdirectory | Responsibility |
|--------------|---------------|
| `interview/` | Coding, system design, behavioral prep |
| `job-search/` | Job posting analysis, application strategy |
| `companies/` | Company research, tech stack, culture |
| `skills-roadmap/` | Learning paths, technical roadmaps |
| `resume-portfolio/` | Resume strategy, portfolio structure |
| `salary/` | Salary research, negotiation strategy |
| `networking/` | Community, mentoring, outreach |

---

## Configuration

Some commands require API keys or external CLI tools.

### Environment Variables

Add to `~/.claude/settings.local.json` (never committed to git):

```json
{
  "env": {
    "BASE_DIR":                  "/absolute/path/to/your/workspace",
    "SCAN_OUTPUT_DIR":           "00_INBOX",
    "TARGET_FILENAME":           "index.md",
    "SCAN_EXCLUDE_PATTERNS":     "findjob/,reports/",
    "GEMINI_API_KEY":            "your-gemini-api-key",
    "DATA_GO_KR_API_KEY":        "your-data-go-kr-api-key",
    "STORAGE_PROVIDER":          "r2",
    "R2_ACCOUNT_ID":             "your-cloudflare-account-id",
    "R2_ACCESS_KEY_ID":          "your-r2-access-key-id",
    "R2_SECRET_ACCESS_KEY":      "your-r2-secret-access-key",
    "R2_BUCKET_NAME":            "presign-shared",
    "INVEST_DEFAULT_SHEET_URL":  "https://docs.google.com/spreadsheets/d/YOUR_SHEET_ID/edit",
    "QUARTZ_BASE_URL":           "http://your-quartz-host:7000"
  }
}

BASE_DIR is optional. When set, all commands that generate files (/claude-skills:new-research, /claude-skills:career-*, /claude-skills:wiki-*, /claude-skills:apt, /claude-skills:apt-watch, /claude-skills:image-gen, /claude-skills:invest, /claude-skills:scan) will write their output under that directory instead of the current working directory. For /scan, BASE_DIR also determines which vault is scanned — set it to your vault root when running from a different working directory. This is useful when you work across multiple projects but want all research notes and reports in one place.

INVEST_DEFAULT_SHEET_URL is required by /claude-skills:invest when no sheet URL is passed as an argument.

QUARTZ_BASE_URL is optional. When set, /claude-skills:share appends a Quartz static-site URL alongside the presigned R2 URL. Omit it if you don't run a local Quartz instance.

For AWS S3 instead of R2, replace the R2_* keys with AWS_ACCESS_KEY_ID, AWS_SECRET_ACCESS_KEY, AWS_DEFAULT_REGION, and S3_BUCKET_NAME.

External CLI Requirements

Command(s)ToolInstall
/claude-skills:gemini, /claude-skills:image-genGemini CLInpm install -g @google/gemini-cli
/claude-skills:ship, /claude-skills:github-urls, /claude-skills:grass-trackerGitHub CLIbrew install gh
/claude-skills:grass-trackergrass-trackerSee repo for install
/claude-skills:cal, /claude-skills:email-summary, /claude-skills:email-archive, /claude-skills:newsletter, /claude-skills:investgws (Google Workspace CLI)See gws repo; gws auth login must be authenticated
/claude-skills:apt, /claude-skills:apt-watch, /claude-skills:presign, /claude-skills:share, /claude-skills:email-summary, /claude-skills:image-gen, /claude-skills:invest, /claude-skills:findjob, /claude-skills:newsletteruvcurl -LsSf https://astral.sh/uv/install.sh | sh
/claude-skills:career-to-pptxpython-pptxInstalled automatically via uv add python-pptx
pptx skillLibreOffice, Popplerapt install libreoffice poppler-utils

Research Templates

The /claude-skills:new-research and /claude-skills:apply-research-template commands use a seven-tier template system. Templates are auto-selected from topic keywords or can be specified explicitly (T1–T5 for general research; T6 and T7 are used exclusively by /invest and /newsletter).

TemplateNameBest ForAuto-trigger keywords
T1Executive BriefQuick summaries, overviews(default)
T2Tech Deep-DiveArchitecture, implementation"architecture", "deep dive", "analysis"
T3Market AnalysisTrends, competitive landscape"market", "trend", "landscape"
T4Comparative EvaluationSide-by-side comparisons"comparison", "vs", "evaluation"
T5Strategic RoadmapPlans, phases, milestones"strategy", "roadmap", "plan"
T6Investment ReportPortfolio analysis with chartsUsed exclusively by /invest
T7Newsletter DigestGmail newsletter intelligence digestUsed exclusively by /newsletter

Output Directory Layout

Commands that create files write to the following paths relative to your working directory:

Command groupOutput path
/claude-skills:new-researchnotes/<domain>/<topic>.md
/claude-skills:apply-research-templateSame directory as input file
/claude-skills:scan$SCAN_OUTPUT_DIR/$TARGET_FILENAME (default: 00_INBOX/index.md)
/claude-skills:career-*career/<subfolder>/
/claude-skills:wiki-*wiki/compiled/
/claude-skills:apt, /claude-skills:apt-watchreports/
/claude-skills:investreports/finance/investment-report-YYYY-MM-DD.md
/claude-skills:newsletternotes/newsletters/newsletter-{label-slug}-YYYY-MM-DD.md
/claude-skills:findjobcareer/job-search/findjob/findjob-YYYY-MM-DD.md + jobs.db
/claude-skills:image-gennotes/image-gen/ (or --output path)

Directories are created automatically on first use. Templates are bundled with the plugin under templates/research/ and resolved from the plugin cache at runtime — no project setup required.


License

MIT — see LICENSE

Ver o original no GitHub

Bom para

Relacionados