Tancoai launched its free tier this week—50 local skills, zero API keys required. Your tasks run locally on your machine using your own agent and model. Your task content never leaves your system.
This privacy-first approach is compelling. But there's a critical prerequisite that determines how much value you'll actually get: your LLM's capability level.
In this guide, I'll categorize these 50 skills by model requirements so you can make an informed decision about which ones to install now, and which to wait on.
The Ground Rules
Free skills come as standard SKILL.md packages. The flow is: download ZIP → validate → place in your local skill directory per your agent's documentation → refresh → copy the task instruction and run it with your agent.
No API keys needed. Your platform never sees your task content. But the quality of results depends entirely on your model. That same skill will produce polished output with GPT-4o or Claude Sonnet, and a basic template with a weaker model.
Skills for Weaker Models (~15)
These 15 skills focus on structured information rearrangement rather than deep understanding. Even basic models can follow fixed rules effectively:
- Meeting Summary: Extract agenda items, conclusions, action items, owners, deadlines, and open questions from meeting notes
- Weekly Report: Convert chat logs and TODOs into completed tasks, impact, blockers, and next week's priorities
- Expense Categorization: Sort transactions into categories with amounts and flagged anomalies
- Calendar Conflict Detection: Find overlapping time slots and hidden scheduling friction (commute time, prep buffer)
- Travel Checklist: Generate destination-aware packing lists accounting for weather, duration, and activities
- Group Expense Split: Preserve payer, participants, currency, and calculate minimal transfers to settle
- Used Item Listing: Organize condition, defects, accessories, and listing terms
- Relocation Plan: Sequence pre-move, moving day, and post-move tasks
- Personal Introduction: Select relevant experiences for the occasion and time constraints
- Email Drafting: Adjust tone based on recipient relationship and communication goal
- Commit Message: Normalize code changes to Conventional Commits format
- Flashcard Generation: Convert material into active recall cards with fixed structure
- Task Prioritization: Rank TODOs by impact, deadline, dependencies, and effort
- Birth Month Flowers: Look up and generate month-specific flower meanings
- Color Palette: Convert color names into harmonized palette cards
These work because they're formula-driven, not generation-heavy.
Skills for Mid-Tier Models (~15)
These require genuine understanding but work well with GPT-4o or Claude Sonnet:
- Project Handoff: Generate transfer packages with checklists, checksums, quality gates, and relay chains
- Change Risk Analysis: Define affected files, track versions, detect boundary violations, run regression checks
- 8-Type Personality Test: Role-play 8 personas evaluating your skill or idea
- Concept Naming: Draw from 8 sources and apply era and aesthetic filters to suggest names
- Knowledge Decomposition: Find prerequisites and break topics into achievable sub-goals
- Pricing Breakdown: Separate costs, margins, tax, refund rate into detailed line items
- Reading Notes: Parse highlights into assertions, evidence, questions, and action items
- Presentation Structure: Confirm audience decision, then outline page-by-page narrative
- Xiaohongshu Copy: Write post drafts based on verified product facts (no fabrication)
- Spaced Repetition: Build a study schedule accounting for subject and weak areas
- Resume Review: Diagnose resume gaps and suggest edits against job requirements
- Argument Outline: State thesis, then recursively decompose
- Error Analysis: Categorize mistakes by knowledge gap, reading error, computation, or strategy
- Interview Prep: Deep-dive follow-up questions on real past projects
- Year in Review: Reflect on the year through key milestones and connections
These are solid with modern mid-tier models. Lighter models will produce vaguer results.
Skills for Strong Models (~8)
These demand high language sophistication, reasoning depth, or domain understanding:
- Five-Dimension Review: Check for AI tone, marketing voice, persona consistency, technical depth, and authenticity—while avoiding duplication with past posts
- API Contract Review: Map resource names, method semantics, status codes, error structures, auth, pagination, idempotency, and versioning
- 16 Personality Types Coach: Generate observation questions, counterexamples, and week-long micro-experiments per type
- Nine Behavior Mirrors: Score 18 scenario questions across 9 behavioral dimensions
- Parent-Child Game Generator: Produce genuinely playable games within safety and age-appropriateness bounds
Strong models (Claude Opus/Sonnet, GPT-4) unlock the real value here. Weaker models will feel surface-level.
Free Tier Boundaries
The free tier handles local, lightweight tasks. If you need fact-checking, source verification, or provable claims—the paid tier (辩真, 灼见, 连珠) handles those. Those tools can verify against external sources and are built for precision.
New users get 5 free premium tasks to test. Use them on something you genuinely need verification for, not routine free skill testing.
How to Get Started
Free skills: Download ZIP → place in skill directory → refresh agent → copy instruction and submit
Paid skills: Add MCP config: https://mcp.tancoai.com/mcp, then run discover_skills to confirm
Links:
- GitHub: https://github.com/tancoai/lianzhu-skill
- Platform: tancoai.com
The Bottom Line
Free skills don't take your API keys or task content. But they do depend on your model. Start by fixing your LLM layer. Once you have a capable model, the skill library multiplies its value.
Skill strength isn't about platforms—it's about the human asking the questions and the model answering them.
Top comments (0)