Skill category
Use this category when the work is about improving how you instruct Claude — not just using Claude, but getting it to behave correctly and consistently across your project.
Profiles for writing better Claude Code instructions, CLAUDE.md configuration, and system prompt patterns.
Use the hub when you need a wider choice set, review methodology, or a different workflow path. Return to the skills overview
Use this profile when you are comparing Claude Code skill collections and need a clear mental model for what a package is, how SKILL.md-based bundles work, and how to evaluate them before installation.
3 source signals • Last reviewed 2026-04-01
Do not blindly install unreviewed skills from SkillsMP or any other public directory.
Use this profile when Claude Code keeps missing project conventions, architecture context, or operating rules and you need to structure CLAUDE.md files deliberately instead of relying on repeated prompts.
3 source signals • Last reviewed 2026-04-01
Do not put API keys, passwords, or other secrets in any CLAUDE.md file.
Use this profile when the quality bottleneck is not installation but instruction design: project rules, system framing, few-shot examples, and stable prompt patterns that Claude Code can reuse.
3 source signals • Last reviewed 2026-04-01
Do not encode secrets, private credentials, or sensitive customer data into reusable system prompts or CLAUDE.md files.
Use this profile when the problem is less about installation and more about getting consistently strong Claude Code output. It fits operators who want better prompt framing, review habits, and session control.
1 source signals • Last reviewed 2026-04-02
Do not mistake confident model output for verified results; expert workflows still need repo-level checks and manual review.
Use this profile when a repository already has Claude instructions but they are outdated, incomplete, or drifting from reality. It is especially useful after toolchain changes, workflow shifts, or repeated assistant mistakes.
1 source signals • Last reviewed 2026-04-02
Do not add secrets, tokens, or customer-specific data to CLAUDE.md just because the assistant benefits from more context.
Use this profile when the challenge is not creating CLAUDE.md but deciding how to format persistent working memory inside it. It fits teams that want repo instructions, plans, and context blocks to stay readable for both humans and Claude Code.
1 source signals • Last reviewed 2026-04-02
Keep working memory concise so Claude Code is not forced to parse a bloated instruction file full of stale context.
Use this profile when Claude-oriented workflows need shared memory or persistent context beyond the current session. It is useful for teams experimenting with memory-backed agent tooling and prompt persistence.
1 source signals • Last reviewed 2026-04-02
Persistent memory layers can quietly accumulate sensitive data, so define what is allowed into memory before connecting the service broadly.
Use this profile when subagent quality problems come from context overload, missing inputs, or fuzzy task boundaries. It is useful for teams that already use delegated workflows and now need sharper context discipline.
1 source signals • Last reviewed 2026-04-02
Do not pass broad repo or secret context to every subagent by default; include only what the delegated task truly needs.
Need broader discovery? Skills overview
Need OpenClaw context? OpenClaw workflow