agent-skill-repository-index
Find and compare Daniel's reviewed GitHub skill sources. Use for high-star skill discovery, link checks, or safe installation, update, and removal of one selected skill.
Install
npx skills add https://github.com/aAAaqwq/AGI-Super-Team/tree/main/skills/agent-skill-repository-index
claude plugin marketplace add https://llmmart.ai/marketplace.json && claude plugin install aaaaqwq-agi-super-team@llmmart
git clone https://github.com/aAAaqwq/AGI-Super-Team.git
The skills CLI installs just this skill, for any of its supported agents. Claude Code installs the whole aaaaqwq/agi-super-team collection as a plugin from our marketplace. Git is the plain clone.
Skill manifest
Daniel's Recommended High-Star GitHub Skills
Route agents to reviewed upstream repositories without globally activating every repository. Treat popularity as a discovery signal, never as proof of quality or safety.
Load Only What You Need
- Read references/repositories.md to find and compare sources.
- Read references/star-snapshot.md only when the user asks about popularity or ranking.
- Read references/installing.md before installing, updating, linking, migrating, or removing a skill.
- Run
python3 scripts/verify_repository_index.py --repo-root /absolute/path/to/reposto verify index-to-clone identity. Add--remotewhen current GitHub reachability matters.
Fast Workflow
- Resolve intent: discovery, comparison, installation, update, removal, or link verification.
- Search already installed skills first. Reuse an equivalent Codex-native skill instead of adding a duplicate.
- Shortlist at most three sources by capability, runtime compatibility, maintenance, license, and risk class. Use stars only as a tie-breaker.
- If a repository is an
awesome-*catalog or has no rootSKILL.md, use it for discovery only. Trace the selected item to its original upstream. - Inspect one candidate directory completely:
SKILL.md, adjacent resources, license, Git remote, commit, worktree state, scripts, hooks, dependencies, credentials, network actions, and persistent state. - Classify it as
DAILY,LIBRARY,QUARANTINE, orDUPLICATE. Do not installQUARANTINEcapabilities without explicit authorization for their runtime effects. - Follow references/installing.md. Install only the reviewed skill subdirectory, not the entire aggregator.
- Validate metadata and resources, then test one positive trigger and one negative trigger.
- Return an installation receipt containing upstream URL, local source, subpath, commit, license, class, target, install method, adaptations, checks, update method, and rollback method.
Efficient Search
Search metadata before opening files:
rg -n --glob 'SKILL.md' '^(name|description):.*KEYWORD' /absolute/path/to/repos
Then inspect only the chosen skill directory. Scan risky behavior without printing secret values:
rg -l --hidden '(curl.+\|\s*(sh|bash)|sudo\b|rm\s+-rf|cookies?|auth|token|secret|hook|auto.?update|deploy|tunnel|daemon|memory)' /absolute/path/to/candidate
Decision Rules
- Prefer the original upstream, a Codex-native plugin, or a single passive instruction skill.
- Prefer a symlink for a clean reviewed local source that should update with its repository; prefer a normalized copy when Codex-specific adaptation is required.
- Keep catalogs, full runtimes, hooks, MCP servers, auto-updaters, deployers, tunnels, browser profiles, and persistent-memory systems inactive during discovery.
- Preserve dirty repositories and existing targets. Never overwrite or delete them to complete an installation.
- Fetch live star counts only when the user asks for a current ranking. Label the bundled figures as Daniel's July 2026 snapshot.
For “redesign this landing page,” invoke the selected design skill. For “which high-star design skill repository does Daniel recommend, and install the safest option,” invoke this skill first.
Files (agi-super-team)
-
agents
-
openai.yaml 301 B
interface: display_name: "Daniel's High-Star GitHub Skills" short_description: "Discover and safely install Daniel's GitHub skill picks" default_prompt: "Use $agent-skill-repository-index to choose, verify, and safely install the best-fit skill from Daniel’s recommended GitHub repositories."
-
-
references
-
installing.md 4.7 KB
# Safe Skill Installation and Lifecycle ## Table of Contents - [Choose the installation mode](#choose-the-installation-mode) - [Preflight](#preflight) - [Codex](#codex) - [Claude Code and OpenClaw](#claude-code-and-openclaw) - [Validate](#validate) - [Update](#update) - [Rollback or remove](#rollback-or-remove) - [Installation receipt](#installation-receipt) ## Choose the Installation Mode | Candidate | Action | |---|---| | Catalog with no runnable root skill | Do not install; search it and follow the chosen item to upstream. | | Repository containing many skills | Select and review one subdirectory. Never expose the repository root globally. | | Passive single skill with compatible metadata | Link the reviewed directory or use the client's native installer. | | Skill requiring Codex-specific wording or tools | Copy to a personal Codex skill root, normalize, and record provenance. | | Plugin, hook suite, MCP server, daemon, memory system, browser profile, deployer, or connector | Keep quarantined until the user explicitly authorizes those runtime effects. | ## Preflight 1. Resolve the exact upstream URL and local source directory. 2. Confirm `git remote get-url origin`, commit, license, and worktree status. 3. Read the complete candidate directory and identify every executable or external action. 4. Compare its name and trigger semantics with installed skills. 5. Select an explicit destination and verify it is absent. Never overwrite an existing file, directory, or symlink. 6. Obtain explicit approval for any quarantined behavior; ordinary passive local skill installation needs no extra approval when the user already requested it. ## Codex Prefer these methods in order: 1. Use the built-in `$skill-installer` for a supported GitHub repository or subpath. 2. For a clean reviewed local source, create one symlink in the documented shared personal skill root, represented here as `<agent-skills-dir>/<skill-name>`. 3. When adaptation is required, copy the complete skill into `<codex-home>/skills/<skill-name>`, normalize frontmatter to only `name` and `description`, and generate `agents/openai.yaml`. 4. For project-only behavior, use the project's documented `.agents/skills/<skill-name>` location instead of a personal-global target. Before linking, resolve both source and target to explicit absolute paths. Refuse if the target already exists. Link only the directory containing the reviewed `SKILL.md`. ## Claude Code and OpenClaw - Prefer the upstream repository's documented marketplace or plugin installation when the candidate is a plugin rather than a plain skill. - For a plain Claude Code skill, use the client-documented `<claude-config>/skills/<skill-name>` location only after adapting unsupported tool names when necessary. - For a plain OpenClaw skill, use the documented `<openclaw-config>/skills/<skill-name>` location and review OpenClaw-specific hooks, services, and credentials separately. - Do not copy one client's hooks, permissions, slash commands, or tool metadata literally into another client. ## Validate For Codex-compatible skills, run: ```bash python3 <codex-home>/skills/.system/skill-creator/scripts/quick_validate.py <installed-skill-dir> ``` Also verify: - every referenced relative resource resolves; - folder name equals frontmatter `name`; - the global or project link resolves to the intended source; - one representative request triggers the skill; - one nearby but out-of-scope request does not trigger it; - no duplicate name or substantially overlapping description was introduced. ## Update 1. Inspect local changes before fetching. Preserve dirty clones. 2. Fetch the upstream and compare commits. 3. Fast-forward only a clean, non-diverged clone. 4. Re-run the candidate audit, risk scan, validator, and trigger tests before accepting the update. 5. For normalized copies, review and port upstream changes deliberately; do not overwrite the adaptation wholesale. ## Rollback or Remove - For a symlink, verify `readlink` points to the expected source, then unlink only that exact link. - For a copied skill, move the exact directory to a dated quarantine or Trash location rather than deleting it immediately. - For a plugin or runtime, use its native disable/uninstall flow and confirm hooks, services, MCP entries, credentials, and background processes are removed or intentionally retained. - Restore the recorded previous commit or prior normalized copy, then rerun validation. ## Installation Receipt Return and record: ```text upstream_url: source_subpath: source_commit: license: classification: destination: install_method: native | symlink | normalized-copy | project-local adaptations: validation: positive_trigger: negative_trigger: update_method: rollback_method: ``` -
repositories.md 6.7 KB
# Daniel's Recommended GitHub Skill Sources This portable source index is used by AGI Super Team. A maintainer may keep reviewed mirrors under a private `<repository-mirror-root>/<local>` directory; verify each remote and commit before migration. See [star-snapshot.md](star-snapshot.md) for Daniel's supplied July 2026 popularity snapshot and [installing.md](installing.md) for the safe lifecycle workflow. Stars do not determine safety. ## Tier S | Repository | Local | Class | Best use | |---|---|---|---| | [anthropics/skills](https://github.com/anthropics/skills) | `skills` | LIBRARY | Official reference implementations and format conventions. | | [obra/superpowers](https://github.com/obra/superpowers) | `superpowers` | LIBRARY | Engineering methodology; use its Codex plugin only when explicitly wanted. | | [affaan-m/everything-claude-code](https://github.com/affaan-m/everything-claude-code) | `everything-claude-code` | LIBRARY; runtime QUARANTINE | Broad operator patterns; select passive skills only. | | [forrestchang/andrej-karpathy-skills](https://github.com/forrestchang/andrej-karpathy-skills) | `andrej-karpathy-skills` | DAILY | Compact coding-behavior guardrails. | | [mattpocock/skills](https://github.com/mattpocock/skills) | `mattpocock-skills` | LIBRARY with selected DAILY | TypeScript, teaching, writing, QA, planning, and issue workflows. | | [leonxlnx/taste-skill](https://github.com/leonxlnx/taste-skill) | `taste-skill` | LIBRARY with selected DAILY | Anti-slop interface design; `design-taste-frontend` is the reviewed global candidate. | | [shanraisshan/claude-code-best-practice](https://github.com/shanraisshan/claude-code-best-practice) | `claude-code-best-practice` | LIBRARY; examples QUARANTINE | Claude Code concepts, settings, hooks, and MCP examples. | ## Tier A | Repository | Local | Class | Best use | |---|---|---|---| | [hesreallyhim/awesome-claude-code](https://github.com/hesreallyhim/awesome-claude-code) | `awesome-claude-code` | LIBRARY | Claude Code resource discovery. | | [addyosmani/agent-skills](https://github.com/addyosmani/agent-skills) | `agent-skills` | LIBRARY with selected DAILY | Production SDLC workflows. | | [alchaincyf/nuwa-skill](https://github.com/alchaincyf/nuwa-skill) | `nuwa-skill` | LIBRARY | Researching and distilling people’s mental models. | | [ComposioHQ/awesome-claude-skills](https://github.com/ComposioHQ/awesome-claude-skills) | `awesome-claude-skills` | LIBRARY; connectors QUARANTINE | Large discovery catalog; review auth actions separately. | | [pbakaus/impeccable](https://github.com/pbakaus/impeccable) | `impeccable` | DAILY candidate | UI anti-pattern audit and design quality. | | [OthmanAdi/planning-with-files](https://github.com/OthmanAdi/planning-with-files) | `planning-with-files` | DAILY | Persistent file-backed planning; keep one canonical installation. | | [Lum1104/Understand-Anything](https://github.com/Lum1104/Understand-Anything) | `Understand-Anything` | LIBRARY; automation QUARANTINE | Knowledge graphs; services and hooks remain opt-in. | | [VoltAgent/awesome-openclaw-skills](https://github.com/VoltAgent/awesome-openclaw-skills) | `awesome-openclaw-skills` | LIBRARY | OpenClaw/ClawHub discovery. | | [LearnPrompt/ai-news-radar](https://github.com/LearnPrompt/ai-news-radar) | `ai-news-radar` | DAILY reader; pipeline LIBRARY | Zero-key AI news reader; publishing pipeline is separate. | | [composio-community/awesome-codex-skills](https://github.com/composio-community/awesome-codex-skills) | `awesome-codex-skills` | LIBRARY; connectors QUARANTINE | Codex-oriented skill discovery. | | [alirezarezvani/claude-skills](https://github.com/alirezarezvani/claude-skills) | `claude-skills` | LIBRARY | Broad production catalog; select one domain skill at a time. | | [wshobson/agents](https://github.com/wshobson/agents) | `agents` | LIBRARY with plugin candidates | Specialist agents and single-purpose plugins. | | [ruvnet/claude-flow](https://github.com/ruvnet/claude-flow) | `claude-flow` | QUARANTINE runtime; docs LIBRARY | Orchestration, hooks, daemon, federation, and persistent memory. | ## Tier B | Repository | Local | Class | Best use | |---|---|---|---| | [VoltAgent/awesome-agent-skills](https://github.com/VoltAgent/awesome-agent-skills) | `awesome-agent-skills` | LIBRARY | Cross-agent discovery; trace every item to upstream. | | [thedotmack/claude-mem](https://github.com/thedotmack/claude-mem) | `claude-mem` | QUARANTINE | Persistent memory, hooks, worker, database, and MCP. | | [supermemoryai/supermemory](https://github.com/supermemoryai/supermemory) | `supermemory` | QUARANTINE | Memory/RAG platform with external data and profiles. | | [garrytan/gstack](https://github.com/garrytan/gstack) | `gstack` | LIBRARY methods; runtime QUARANTINE | Founder workflows; full setup adds browser, daemon, deploy, and updater behavior. | | [ruvnet/ruflo](https://github.com/ruvnet/ruflo) | `ruflo` | DUPLICATE / QUARANTINE | Same current code lineage as `claude-flow`; choose one canonical source. | | [0xfurai/claude-code-subagents](https://github.com/0xfurai/claude-code-subagents) | `claude-code-subagents` | LIBRARY | Claude-specific specialist prompt library. | | [robwhite4/claude-memory](https://github.com/robwhite4/claude-memory) | `claude-memory` | QUARANTINE | Background session rotation, backups, and persistent memory. | ## Specialty | Repository | Local | Class | Best use | |---|---|---|---| | [zarazhangrui/frontend-slides](https://github.com/zarazhangrui/frontend-slides) | `frontend-slides` | LIBRARY / design candidate | HTML presentations; deployment and runtime downloads remain explicit. | | [nashsu/AutoCLI](https://github.com/nashsu/AutoCLI) | `AutoCLI` | QUARANTINE | Authenticated browser state, extension, adapters, and optional cloud sync. | | [ZhuLinsen/daily_stock_analysis](https://github.com/ZhuLinsen/daily_stock_analysis) | `daily_stock_analysis` | QUARANTINE application | Financial-analysis application, scheduled automation, and push channels. | ## General Discovery | Repository | Local | Class | Best use | |---|---|---|---| | [sindresorhus/awesome](https://github.com/sindresorhus/awesome) | `awesome` | LIBRARY | General curated-list index. It has no `SKILL.md` and does not overlap functionally with `taste-skill`. | ## Selection Rules 1. Search installed Codex-native skills before migrating anything. 2. Prefer a narrow upstream skill over an aggregator copy. 3. Read the candidate `SKILL.md` completely and inspect its adjacent resources. 4. Scan for secrets, hooks, persistence, external writes, deployment, tunnels, and auto-update behavior. 5. Record provenance, pinned commit, license, adaptations, trigger examples, and non-trigger examples. 6. Link or copy only the reviewed skill directory; never expose the entire repository by default. -
star-snapshot.md 1.7 KB
# Daniel's Star Snapshot Approximate GitHub stars supplied in Daniel's July 2026 source list. Values change continuously; fetch live counts before making a current ranking. A high count does not override license, compatibility, provenance, or safety review. | Repository | Approximate stars in supplied list | |---|---:| | `sindresorhus/awesome` | 487K | | `obra/superpowers` | 192K | | `affaan-m/everything-claude-code` | 182K | | `anthropics/skills` | 135K | | `forrestchang/andrej-karpathy-skills` | 131K | | `garrytan/gstack` | 97.4K | | `mattpocock/skills` | 84K | | `leonxlnx/taste-skill` | 61K | | `shanraisshan/claude-code-best-practice` | 53.6K | | `ruvnet/ruflo` | 51.4K | | `hesreallyhim/awesome-claude-code` | 43.4K | | `addyosmani/agent-skills` | 42K | | `ComposioHQ/awesome-claude-skills` | 34.5K | | `pbakaus/impeccable` | 27.6K | | `alchaincyf/nuwa-skill` | 21K | | `OthmanAdi/planning-with-files` | 20K | | `wshobson/agents` | 18.7K | | `zarazhangrui/frontend-slides` | 18K | | `supermemoryai/supermemory` | 16.7K | | `alirezarezvani/claude-skills` | 15.4K | | `composio-community/awesome-codex-skills` | 15.2K | | `thedotmack/claude-mem` | 12.8K | | `Lum1104/Understand-Anything` | 12K | | `ruvnet/claude-flow` | 11K | | `VoltAgent/awesome-openclaw-skills` | 8K | | `VoltAgent/awesome-agent-skills` | 5K | | `0xfurai/claude-code-subagents` | 2K | | `LearnPrompt/ai-news-radar` | Not supplied | | `robwhite4/claude-memory` | Not supplied | | `nashsu/AutoCLI` | Not supplied | | `ZhuLinsen/daily_stock_analysis` | Not supplied | When live ranking is required, prefer the GitHub API or GitHub CLI and cite the observation date. Do not rewrite this snapshot automatically.
-
-
scripts
-
verify_repository_index.py 4.1 KB
#!/usr/bin/env python3 """Verify repository-index links, local clone identities, and optional remotes.""" from __future__ import annotations import argparse import re import subprocess import sys from concurrent.futures import ThreadPoolExecutor from dataclasses import dataclass from pathlib import Path ROW_PATTERN = re.compile( r"^\| \[([^]]+)\]\((https://github\.com/[^)]+)\) \| `([^`]+)` \|" ) @dataclass(frozen=True) class Repository: slug: str url: str local: str def parse_index(reference: Path) -> list[Repository]: repositories: list[Repository] = [] for line in reference.read_text(encoding="utf-8").splitlines(): match = ROW_PATTERN.match(line) if match: repositories.append(Repository(*match.groups())) return repositories def github_slug(value: str) -> str | None: match = re.search(r"github\.com[/:]([^/]+/[^/]+?)(?:\.git)?$", value) return match.group(1).lower() if match else None def git_output(*args: str) -> str: return subprocess.run( ["git", *args], check=True, capture_output=True, text=True, timeout=30, ).stdout.strip() def check_local(repository: Repository, root: Path) -> str | None: clone = root / repository.local if not (clone / ".git").exists(): return f"missing clone: {clone}" try: remote = git_output("-C", str(clone), "remote", "get-url", "origin") except (subprocess.SubprocessError, OSError) as error: return f"cannot read origin: {error}" expected = repository.slug.lower() actual = github_slug(remote) if actual != expected: return f"origin mismatch: expected {expected}, got {remote}" return None def check_remote(repository: Repository) -> str | None: try: subprocess.run( ["git", "ls-remote", "--exit-code", repository.url, "HEAD"], check=True, capture_output=True, text=True, timeout=45, ) except (subprocess.SubprocessError, OSError) as error: return f"remote unreachable: {error}" return None def main() -> int: skill_root = Path(__file__).resolve().parent.parent parser = argparse.ArgumentParser(description=__doc__) parser.add_argument( "--reference", type=Path, default=skill_root / "references" / "repositories.md", ) parser.add_argument("--repo-root", type=Path) parser.add_argument("--remote", action="store_true") parser.add_argument("--expected-count", type=int) args = parser.parse_args() repositories = parse_index(args.reference) errors: list[str] = [] if args.expected_count is not None and len(repositories) != args.expected_count: errors.append( f"entry count mismatch: expected {args.expected_count}, got {len(repositories)}" ) if len({item.slug.lower() for item in repositories}) != len(repositories): errors.append("duplicate repository slug") if len({item.local for item in repositories}) != len(repositories): errors.append("duplicate local directory") for repository in repositories: if github_slug(repository.url) != repository.slug.lower(): errors.append( f"label/url mismatch: {repository.slug} points to {repository.url}" ) if args.repo_root: for repository in repositories: if error := check_local(repository, args.repo_root): errors.append(f"{repository.slug}: {error}") if args.remote: with ThreadPoolExecutor(max_workers=6) as executor: results = executor.map(check_remote, repositories) for repository, error in zip(repositories, results, strict=True): if error: errors.append(f"{repository.slug}: {error}") for error in errors: print(f"FAIL {error}") if errors: return 1 checks = [f"{len(repositories)} entries"] if args.repo_root: checks.append("local identities") if args.remote: checks.append("GitHub remotes") print("PASS " + ", ".join(checks)) return 0 if __name__ == "__main__": sys.exit(main())
-
-
SKILL.md 3.5 KB
--- name: agent-skill-repository-index description: "Find and compare Daniel's reviewed GitHub skill sources. Use for high-star skill discovery, link checks, or safe installation, update, and removal of one selected skill." --- # Daniel's Recommended High-Star GitHub Skills Route agents to reviewed upstream repositories without globally activating every repository. Treat popularity as a discovery signal, never as proof of quality or safety. ## Load Only What You Need - Read [references/repositories.md](references/repositories.md) to find and compare sources. - Read [references/star-snapshot.md](references/star-snapshot.md) only when the user asks about popularity or ranking. - Read [references/installing.md](references/installing.md) before installing, updating, linking, migrating, or removing a skill. - Run `python3 scripts/verify_repository_index.py --repo-root /absolute/path/to/repos` to verify index-to-clone identity. Add `--remote` when current GitHub reachability matters. ## Fast Workflow 1. Resolve intent: discovery, comparison, installation, update, removal, or link verification. 2. Search already installed skills first. Reuse an equivalent Codex-native skill instead of adding a duplicate. 3. Shortlist at most three sources by capability, runtime compatibility, maintenance, license, and risk class. Use stars only as a tie-breaker. 4. If a repository is an `awesome-*` catalog or has no root `SKILL.md`, use it for discovery only. Trace the selected item to its original upstream. 5. Inspect one candidate directory completely: `SKILL.md`, adjacent resources, license, Git remote, commit, worktree state, scripts, hooks, dependencies, credentials, network actions, and persistent state. 6. Classify it as `DAILY`, `LIBRARY`, `QUARANTINE`, or `DUPLICATE`. Do not install `QUARANTINE` capabilities without explicit authorization for their runtime effects. 7. Follow [references/installing.md](references/installing.md). Install only the reviewed skill subdirectory, not the entire aggregator. 8. Validate metadata and resources, then test one positive trigger and one negative trigger. 9. Return an installation receipt containing upstream URL, local source, subpath, commit, license, class, target, install method, adaptations, checks, update method, and rollback method. ## Efficient Search Search metadata before opening files: ```bash rg -n --glob 'SKILL.md' '^(name|description):.*KEYWORD' /absolute/path/to/repos ``` Then inspect only the chosen skill directory. Scan risky behavior without printing secret values: ```bash rg -l --hidden '(curl.+\|\s*(sh|bash)|sudo\b|rm\s+-rf|cookies?|auth|token|secret|hook|auto.?update|deploy|tunnel|daemon|memory)' /absolute/path/to/candidate ``` ## Decision Rules - Prefer the original upstream, a Codex-native plugin, or a single passive instruction skill. - Prefer a symlink for a clean reviewed local source that should update with its repository; prefer a normalized copy when Codex-specific adaptation is required. - Keep catalogs, full runtimes, hooks, MCP servers, auto-updaters, deployers, tunnels, browser profiles, and persistent-memory systems inactive during discovery. - Preserve dirty repositories and existing targets. Never overwrite or delete them to complete an installation. - Fetch live star counts only when the user asks for a current ranking. Label the bundled figures as Daniel's July 2026 snapshot. For “redesign this landing page,” invoke the selected design skill. For “which high-star design skill repository does Daniel recommend, and install the safest option,” invoke this skill first.
Comments (0)
Sign in to join the conversation.
Reviews (0)
No reviews yet.
No comments yet.