My public skill stack for Codex, npx skills, and agent workflows that need more discipline than a giant prompt.

This is not a universal skill registry. It is the installable public surface for skills I actually use, maintain, or curate. Drafts, risky experiments, generated docs, raw upstream mirrors, and operator-specific material are kept out of this public repo unless they have been reviewed and deliberately promoted.

This is my agent operating stack:

• execution discipline so long work does not dissolve into noisy chat
• proof discipline so "done" means something specific
• skill hygiene so useful workflows become reusable capability instead of one-off conversation residue
• model and capability routing through portable, availability-aware presets instead of hard-coded model choices scattered across prompts
• design operators for frontend, visual systems, Penpot, and Illustrator-style work
• curated external mirrors when outside skills are useful but need prefixes, provenance, and install hygiene before sharing

The folder structure is part of the product. It keeps my owned workflow skills, design/operator skills, knowledge packs, generated docs, and external mirrors separate so they can be installed, reviewed, and updated without collisions.

Agents get better when useful workflow knowledge becomes durable:

• repeated failures become rules, scripts, or skills
• repeated checks become deterministic tools
• long work gets checkpoints instead of noisy chat
• skill catalogs stay reviewable instead of becoming a junk drawer
• public installs stay narrow, attributed, and security-reviewed

The point is not to collect every skill. The point is to keep a working stack that can be installed, audited, and improved without losing provenance.

TheAngrySkills is organized by how a skill behaves in real work, not just by who wrote it.

This matters because skill repos collide fast. Many packs eventually contain their own handoff, review, browser, plan, or ingest skill. Prefixes and categories make it clear what is mine, what is mirrored, what is generated, and what is safe to install by default.

npx skills add https://github.com/TheAngryPit/TheAngrySkills.git --list

Include nested mirror packs:

npx skills add https://github.com/TheAngryPit/TheAngrySkills.git --list --full-depth

npx skills add https://github.com/TheAngryPit/TheAngrySkills.git --skill skill-catalog-curator

npx skills add https://github.com/TheAngryPit/TheAngrySkills.git --skill skill-catalog-curator -g -a codex

Install the HyperFrames router only:

npx skills add https://github.com/TheAngryPit/TheAngrySkills.git --skill hyperframes

Install model-capability-router for this operator's recommended native palette: GPT-6.1 Sol Medium for coordination and open-ended execution, Astra Low for planning/review and difficult decisions with XHigh reserved for exceptional reasoning, and Luna High/XHigh/Max for bounded work. Preserve an explicit native selection, including Sol Low; the recommendation does not change a home configuration or active thread. Choose directly for the task; use one visible task per coherent outcome and native subagents for useful bounded work. User-owned tasks require an explicit request and are for independent or durable outcomes, schedules, or distinct host/cloud boundaries. There is no mandatory escalation ladder, resolver or coordinator stage. Respect the selected main model and live channel availability. The optional model-routing-preset-builder customises this same policy, not a second runtime. Both skills are installed only from this repository, never Workbench.

npx skills add https://github.com/TheAngryPit/TheAngrySkills.git -g -a codex -y -s model-capability-router -s model-routing-preset-builder

Install continuity-handoff to create a durable packet under .traverse/continuity/, move between primary and secondary profiles without forking a bloated transcript, and perform one bounded recall against the exact recorded source. Packet creation never commits, archives, deletes, or mutates the source task automatically.

Install the custom-build validation skill:

npx skills add https://github.com/TheAngryPit/TheAngrySkills.git --skill openclaw-custom-build-validation

Ask your agent to validate a specific PR head, compose an ordered patch stack, or refresh a custom build onto a chosen upstream commit. The skill records the exact candidate, uses OpenClaw Manager (ocm) for local test environments, and uses openclaw-crabbox when remote reproduction adds useful evidence. Install that companion separately if remote testing is needed.

You need Git, the candidate's build tools, and OCM for local validation. The bundled helpers use Node.js; source checks and approved PR comments use an authenticated GitHub CLI. Remote tests also require an authorized provider. Installing this skill does not install those tools or start any services.

Validation preserves the personal installation. Adoption, optional telemetry, and publishing comments require the user's authorization. Reports distinguish source inspection, automated tests, runtime checks, and human confirmation. This is a TheAngrySkills workflow, not an official OpenClaw release campaign.

The helpers have regression tests and a secrets scan. The complete OCM/Crabbox workflow has not yet been validated end to end; a passing helper test is not evidence that a particular OpenClaw build works.

Use this repo like a guided setup, not like a blind package install:

The workflow is simple: stage the procedure, show what will happen, confirm the risky step, then make the result durable.

See docs/skill-categories.md for category rules.

Documentation skills are not published here by default. They are still part of the operating model: when a tool or project matters, turn its docs into a searchable, attributed reference pack instead of forcing an agent to improvise from memory. Public documentation packs should be promoted deliberately, with clear provenance and safety review.

External mirrors are different. They are curated external skills, not generated docs. Only selected mirror families are promoted for public install. The public stack should not become a blind dump of every mirrored skill from every source.

Operator-specific packs, unpublished workflow experiments, private knowledge workflows, personal voice machinery, company/client strategy, and local maintenance helpers stay out of this public repo unless a sanitized public rewrite is reviewed and promoted deliberately.

The currently accepted mirror families for public curation are:

• OpenClaw skills, with openclaw- prefixes
• Taste skills, with taste- prefixes
• Vercel Agent Skills, with vercel-agent- prefixes
• Marketing skills, with marketing- prefixes
• Looper skills, with looper- prefixes
• Effective HTML skills, with effective- prefixes
• HyperFrames by HeyGen skills, preserving natural names and using hyperframes- for generic names
• Emil Kowalski skills, preserving existing emil- names and using emil- for generic names
• Matt Pocock skills, curated under mirrors-mattpocock/ with approved overlays and provenance
• Cursor plugin skills, selectively indexed under mirrors-cursor/ with a pinned source snapshot and per-skill overlays

The Emil family follows the same one-to-one curated mirror workflow as OpenClaw. It is available as the separate mirrors-emilkowalski family for explicit project selection; it is not added to the owned default install sets.

When mirrors are published here, they should:

• keep source-specific prefixes such as openclaw-, taste-, vercel-, or hyperframes-, or remain under an explicit source mirror family
• preserve upstream README, license, copyright, and notices
• avoid mixing owned TheAngrySkills workflow logic into upstream material
• stay clearly marked as curated mirrors, not original work

Families configured in the curated mirror manifest are refreshed daily by the Sync Curated Mirrors workflow. It discovers valid upstream skill directories, preserves the family naming policy, rewrites only frontmatter names and internal skill paths needed by the prefix, validates the complete catalog, and opens or updates one review PR when upstream changes exist. New or removed upstream skills and changed executable surfaces remain visible in that PR and are never merged automatically.

The Cursor family uses a separate pinned snapshot and reviewed per-skill overlays. It is not in that daily workflow. Its source, selected candidates, holds, and manual update checks are documented in sources/cursor-plugins/README.md.

This matters because different skill packs often collide on names like handoff, review, or browser. Prefixes keep installs legible.

This repo is open source, but still curated.

Good issues and pull requests usually:

• fix a broken npx skills install or update path
• improve one existing skill without changing its role
• add missing provenance, license, or safety notes
• improve tests, CI, scanner behavior, or docs
• propose a new skill with clear scope and no private/local leakage

main is protected. Community PRs are welcome as review inputs, but arbitrary skills, mirrors, generated docs, and install surfaces will not be merged just because they pass CI.

Every PR must clear the minimum safety gate in docs/pr-safety-minimums.md.

Read CONTRIBUTING.md before proposing changes.

Use Discussions for broader questions, setup feedback, and ideas that are not yet concrete issues or PRs.

Skills are executable instructions for agents. Scripts, workflows, install commands, mirrors, generated docs, lifecycle hooks, and package managers are supply-chain surfaces.

Before trusting a skill, check:

• what files it asks an agent to read or write
• whether it runs scripts or shell commands
• whether it changes host/global state
• whether it touches secrets, credentials, hooks, MCP servers, or CI
• whether its source and license are clear

Read SECURITY.md before reporting security-sensitive findings.

Owned TheAngrySkills content is released under the MIT License.

Future mirrored or generated third-party content must preserve its upstream copyright, license, notices, and provenance. The MIT license for this repo does not relicense third-party material.