A curated collection of skills and agents for Claude Code CLI, covering software engineering, DevOps, data engineering, infrastructure administration, and workflow orchestration.
Version: 1.3.2 · Last published: 2026-06-24 · 184 skills · 5 agents · 9 workflows · 2 commands See the Changelog for what changed in each release.
- 184 skills — domain knowledge Claude loads on demand to help with specific tasks. Highlights include the legacy-comprehension family (
lineage-extract-static,mainframe-lineage-parsers,legacy-code-intel,structure-recovery,intent-extract— data/process lineage, deterministic COBOL/JCL/DB2 + Control-M → OpenLineage, SCIP-style symbol graphs, and record-structure recovery for legacy estates),env-readiness(read-only environment readiness doctor),publish-to-github(releasing this whole tree to public github safely),visual-companion(browser-based mockup/diagram review during design), andcross-project-mail(AI-agent messaging across sibling projects on one host) - 5 agents — specialized sub-agents for execution, review/evolution, evergreening, task routing, and knowledge management
- 9 saved workflows — committed
.jsorchestration scripts under~/.claude/workflows/(e.g.bob-serial-exec,design-tournament,alf-sweep) that parameterize the existing agents; main-loop-only on Claude Code, with a host-neutral serial fallback for every other host - Slash commands — short user-invoked workflows under
~/.claude/commands/(e.g./exit-with-docsto wrap up a session and update project docs)
Skills are auto-discovered by Claude Code from frontmatter descriptions — you don't call them explicitly. Agents are invoked by name (forge, bob, alf, pa, wiki). Commands are invoked with a leading slash (e.g. /exit-with-docs).
# 1. Clone the repo
git clone https://github.com/<your-username>/<repo-name>.git
cd <repo-name>
# 2. Bootstrap a full environment (recommended for first-time setup)
python3 bootstrap-environment.py
# or, on Windows: pwsh -NoProfile -NonInteractive -File bootstrap-environment.ps1The bootstrap orchestrator runs install.py plus a series of idempotent post-install steps so a fresh machine reaches a complete Claude Code environment in one command. Re-run any time — every step skips work already done.
| Step | What happens |
|---|---|
| 1 | install.py places skills / agents / commands under ~/.claude/ |
| 2 | Places ~/.claude/CLAUDE.md (global instructions) |
| 3 | Symlinks ~/.claude/AGENTS.md → CLAUDE.md (GitHub Copilot bridge) |
| 4 | Mirrors pa-server/ into ~/.claude/pa-server/ (MCP server for the pa agent) |
| 5 | Merges canonical SessionStart hooks into ~/.claude/settings.json (preserves any you already have; backs the old file up to .bak) |
| 6 | Writes ~/.claude/policy-limits.json with conservative defaults, mode 0600 |
| 6b | Seeds ~/.claude/memory/ — the global memory tier + preference profiles (user-preferences); never clobbers a populated profile |
| 6c | Restores ~/.claude/publish-config.json from installer/ if present (your private publish scrub list) — else points you at init-config |
| 7 | Symlinks ~/.claude/bin/claude-observe for the process-observation skill |
| 8 | Mirrors skills into ~/.codex/skills/* (only if codex CLI is on PATH) |
| 9 | Pins Gemini model in ~/.gemini/settings.json (legacy — only if file exists) |
| 10 | Prints next-step guidance |
After bootstrap, two more steps inside Claude Code finish the setup:
# Inside Claude Code:
/setup # interactive permissions upgrade
# (conservative -> autonomous)
# Per working repo you care about:
bash scripts/install-pre-push-hook.sh /path/to/repo # POSIX
# or any OS (incl. enterprise Windows where PowerShell is blocked):
python3 scripts/install-pre-push-hook.py --target-repo /path/to/repoThe pre-push hook runs a secrets scanner (live API keys, PEM private keys, AWS credentials, JWTs, inline tokens, etc.) before every git push. Critical/high findings block the push; medium/low advisory hits print but don't block. Override per-push with git push --no-verify.
python3 bootstrap-environment.py --dry-run # preview without writing
python3 bootstrap-environment.py --force # overwrite existing CLAUDE.md / pa-server / policy-limits
python3 bootstrap-environment.py --skip-install # don't run install.py
python3 bootstrap-environment.py --skip-codex # don't touch ~/.codex/
python3 bootstrap-environment.py --skip-gemini # don't touch ~/.gemini/settings.json
python3 bootstrap-environment.py --claude-home /opt/claude # custom config root
python3 bootstrap-environment.py --helpIf you already have a CLAUDE.md, hooks, and other host config and just want the skill content placed:
python3 install.py # interactive
python3 install.py --noninteractive # Claude + link
python3 install.py --target claude,gemini --mode link
python3 install.py --target all --mode move --force # full copy, overwrite
python3 install.py --claude-home /opt/claude # custom path
python3 install.py --help # all flagsinstall.py is what bootstrap-environment.py calls under the hood for step 1, so the flags are pass-through-able via --install-arg.
bootstrap-environment.ps1is enterprise-hardened —#Requires -Version 5.1,[CmdletBinding()]with named switch parameters, no-ExecutionPolicy Bypass, no dot-source, no-Command. Invoke withpwsh -NoProfile -NonInteractive -File bootstrap-environment.ps1.- The bootstrap script probes
py(PEP-397 launcher) →python3→pythonin that order and runs the orchestrator with whichever resolves first. - For backward compatibility, the older
install.cmd/install.ps1entry points are still present. They use-ExecutionPolicy Bypassfor compatibility with very locked-down machines; preferbootstrap-environment.ps1on modern setups. - Symbolic links on Windows require admin or Developer Mode. If symlinks fail, the bootstrap falls back to copy and prints a warning (drift risk if you later edit
CLAUDE.md— theAGENTS.mdcopy won't update).
If you'd rather not run a script:
cp -r skills/* ~/.claude/skills/
cp -r agents/*.md ~/.claude/agents/
cp -r commands/*.md ~/.claude/commands/
cp CLAUDE.md ~/.claude/CLAUDE.md
ln -s ~/.claude/CLAUDE.md ~/.claude/AGENTS.md
cp -r pa-server ~/.claude/pa-serverYou'll then need to add the SessionStart hooks to ~/.claude/settings.json by hand — see the bootstrap script's source for the canonical entries. For the full multi-model experience with Codex and Gemini as second-opinion models, see Dependencies.
| Tier | What it gives you | What to install |
|---|---|---|
| Minimal | All 150+ domain skills work as Claude reference material. wiki agent works fully. |
Claude Code CLI only |
| Standard | + forge / bob / alf agents with multi-model reviews (Codex + Gemini) |
Add Codex CLI + Codex plugin + Gemini CLI MCP |
| Full | + pa agent with persistent task tracking, browser-based product reviews |
Add pa-server MCP (custom) + claude-in-chrome MCP |
See docs/dependencies/ for complete install instructions.
| Agent | Purpose | Run standalone? |
|---|---|---|
| forge | Design exploration with multi-model challengers (Claude + Codex + Gemini). Produces approved design docs. | Yes |
| bob | Autonomous implementation executor. Reads approved design docs, decomposes into work packages, delegates to team-based orchestration, verifies output. | Yes (needs a design doc) |
| alf | Evolution & improvement reviewer. Audits skills/agents/code/products for staleness, drift, capability gaps, security, performance. Produces evidence-cited reports. | Yes |
| pa | Task router and persistent workspace manager. Classifies intent, routes to specialist agents, tracks state across sessions. | Yes (stateless mode without pa-server MCP) |
| wiki | Knowledge base builder. Ingests sources into cited, interlinked markdown wiki pages. Compile-once, read-cheaply model. | Yes (no external dependencies) |
Architecture: pa routes → forge designs → bob implements → alf reviews → all can query wiki. See Agent Graph for the full interaction diagram.
Orchestration & Meta (click to expand)
forge · agent-teams · team-manager · challenger · qa-reviewer · ux-reviewer · research-for-skills · codex-orchestration · web-research · development-lifecycle · project-documentation · simplify · publish-to-github · visual-companion · env-adoption · env-readiness · smart-config · knowledge-grounding
Build and manage agent teams, run multi-model reviews, decompose work.
Knowledge Management
wiki · obsidian · large-file-analysis · confluence-documentation · confluence-content-creator · confluence-rest-api · research-vectorization
Persistent knowledge bases, large document analysis, Confluence integration, vector search.
Python Development
python-flask-developer · python-auth-security · python-parallelism · python-data-engineer · python-enterprise-connectors
Flask, auth/OAuth/OIDC, asyncio/multiprocessing, SQLAlchemy, enterprise database connectors.
JavaScript / Frontend
content-writer · plus JS/TS patterns from upstream plugins
Frontend design, content creation, modern web frameworks.
Java Enterprise
java-backend · java-frontend
Spring Boot, Spring Security, Spring Data JPA, Angular 17+, React 18+, NgRx/Redux.
Docker
docker-fundamentals · docker-networking · docker-storage · docker-security · docker-cicd · docker-compose-patterns · docker-admin
Container lifecycle, networking, volumes, security hardening, CI/CD, multi-stage builds.
Linux — RHEL 9 / AlmaLinux / Rocky
rhel-server-admin · rhel-web-servers · rhel-databases · rhel-docker-host · rhel-file-storage · rhel-network-infra · rhel-monitoring · rhel-ollama-nvidia
Full RHEL 9 administration stack including NVIDIA/CUDA/Ollama for AI workloads.
Linux — Ubuntu 24.04 LTS
ubuntu-server-admin · ubuntu-web-servers · ubuntu-databases · ubuntu-docker-host · ubuntu-file-storage · ubuntu-network-infra · ubuntu-monitoring · ubuntu-ollama-nvidia
Full Ubuntu 24.04 administration stack, parallel to the RHEL set.
Linux — Auth & Identity
linux-centrify
Centrify/Delinea Server Suite, Active Directory integration, PAM/NSS.
Windows Server
windows-powershell · windows-cmd · windows-ps-server-admin · windows-ps-security · windows-ad-admin · windows-sso
PowerShell 7.x + 5.1, batch scripting, AD administration, security hardening, SSO (AD FS, Azure AD, SAML/OAuth/OIDC).
Databases
mongodb · db2-rhel · db2-mainframe
MongoDB (replica sets, sharding), DB2 LUW on RHEL, DB2 for z/OS mainframe.
IBM Mainframe & Middleware
ibm-mainframe · cobol-developer · db2-mainframe · ibm-mq · ibm-websphere · cognos-admin · cognos-user · datastage-developer · pega-robotics · control-m · pick-developer · lineage-extract-static · mainframe-lineage-parsers · legacy-code-intel · structure-recovery · intent-extract · intent-map-render
z/OS, JCL, COBOL, DB2 z/OS, IBM MQ, WebSphere, Cognos Analytics, DataStage, Pega RPA, Control-M workload automation, Pick/MultiValue BASIC. The legacy-comprehension family extracts data/process lineage (LLM-as-parser + a deterministic COBOL/JCL/DB2 + Control-M → OpenLineage 2.0.2 engine with multi-abstraction views), SCIP-style symbol graphs, record-structure/DDL recovery, and per-component functional intent from legacy estates.
Data Engineering
data-lake · data-warehouse · data-mart · datastage-developer · market-data-engineering · research-vectorization
Lakes (Parquet/Iceberg/Delta), warehouses (Kimball/Inmon), marts (star/snowflake), ETL with DataStage, financial data pipelines.
SaaS Architecture
saas-architecture · saas-developer
Multi-tenant patterns, subscription billing, onboarding flows, tenant isolation, feature flags.
Trading & Finance
trading-strategy-backtester · trading-risk-management · crypto-exchange-integration · day-trading-patterns · financial-sentiment-analysis · geopolitical-market-impact · social-trading-signals · market-data-engineering · trader-psychology-analysis
Backtesting with vectorbt/backtrader, kill switches, CCXT, intraday patterns, FinBERT sentiment, event-driven signals.
SEO & Marketing
seo-content-strategist · seo-keyword-strategist · seo-structure-architect · seo-meta-optimizer · seo-serp-optimizer · seo-authority-builder · seo-data-analyst · ai-search-optimizer · conversion-psychology · ecommerce-growth
Full SEO stack including AI search optimization (ChatGPT/Gemini/Perplexity citations), schema markup, topic clusters, GSC/GA4 analytics.
WordPress & WooCommerce
wordpress-developer · wordpress-admin · woocommerce-developer · hostinger-hosting
Theme/plugin development, block editor, WooCommerce checkout, hosting-specific patterns.
Career Development
career-coach · career-assessment · career-planning · career-positioning · career-leadership · career-storytelling · career-transition
Career strategy, self-assessment, interview prep, promotion cases, leadership development.
Presentations
presentation-builder · presentation-narrative · presentation-datavis · presentation-diagrams · presentation-styling · presentation-renderer
Orchestrated presentation creation — narrative frameworks, charts, diagrams, styling, PPTX/HTML export.
Project & Delivery Management
project-manager · delivery-manager · project-finance · jira-rest-api
WBS, critical path, EVM, Agile ceremonies, flow metrics, Jira integration.
DevOps & CI/CD
jenkins · ansible · docker-cicd · gcp-workstations
Pipeline development, infrastructure as code, CI/CD Docker builds, cloud dev environments.
Observability
splunk-developer · rhel-monitoring · ubuntu-monitoring
SPL queries, dashboards, Prometheus/Grafana/Loki, ELK stack, alerting.
CLI Tools & Integrations
claude-code-cli · gemini-cli · gh-copilot-cli · codex-orchestration · mcp-server-creator
Claude Code CLI reference, Gemini CLI, GitHub Copilot CLI, Codex delegation, building MCP servers.
Image Generation
nano-banana · vertex-banana
Image generation via Gemini API (primary) and Vertex AI (fallback).
Performance
performance
Profiling, load testing, query optimization, bottleneck analysis.
- docs/dependencies/README.md — install tiers, component matrix, quick start
- docs/dependencies/local-tools.md — Claude Code CLI, Codex CLI, Gemini CLI, plugins, environment variables
- docs/dependencies/mcp-servers.md — Gemini CLI MCP, claude-in-chrome MCP, pa-server MCP
- docs/dependencies/agent-graph.md — how agents interconnect, data contracts, minimal subsets
- feedback-loop.md — known compatibility issues + reporting template. Read this first if you're installing on Windows / macOS / WSL / a container / an air-gapped host — this collection was developed on a single Linux host and your environment may behave differently. Your bug report becomes the next user's starting point.
Skills and agents in this repository follow these principles:
- Documentation-first — Most skills are reference material Claude reads on demand, not executable code
- Graceful degradation — Missing optional dependencies produce gap notes, never hard failures
- Evidence before assertions — Agents (especially
alf) require sources, tiers, confidence levels for all external claims - Model-agnostic language — Skills avoid Claude-specific constructs where possible; most work in Codex CLI and Gemini CLI too
- No secrets, no tokens — Skills reference environment variable names, never values. The
paagent explicitly refuses to store credentials - Standalone-capable — Every agent can run without the others; the full graph is opt-in
| Installing | You get | Time |
|---|---|---|
| Skills only | 150+ domain knowledge modules | <1 min |
| Skills + agents | + forge/bob/alf/wiki/pa orchestration | <2 min |
| + Codex CLI + plugin | + GPT-5.4 challenger reviews | ~10 min (Codex auth) |
| + Gemini CLI + MCP | + Gemini 3 third-model verification, 1M-context analysis | ~10 min (Gemini auth) |
| + claude-in-chrome | + browser-based product reviews for alf | ~15 min (Chrome extension) |
| + pa-server (custom) | + persistent task tracking across sessions | Build your own |
Contribution guidelines — to be added.
The publish-to-github skill in skills/publish-to-github/ is itself the workflow that produced this repo. It scrubs private content, runs security checks, and produces a clean staging directory. To publish your own modified copy:
- Customize
skills/andagents/for your needs - Create
~/.claude/publish-config.json(template atskills/publish-to-github/templates/publish-config.example.json) - Run
python3 ~/.claude/skills/publish-to-github/scripts/publish_prep.py --extended-scan - Inspect the staging dir, then
git init / add / commit / pushfrom there
See skills/publish-to-github/SKILL.md for the full workflow.
- Privacy scrub: the maintainer's GitHub handle is genericized to
your-gh-userin published examples (a schema sample, a SARIFinformationUri, clone/issue URLs) and added as a scrubberforbidden_pattern, so the publish gate blocks it going forward. (Emails + web addresses audited — all were already placeholders/examples.)
- Fix: trimmed the
legacy-code-intel(1038→1000) andmainframe-lineage-parsers(1227→990) descriptions back under the 1024-char SKILL.md limit — they had grown past it with the Pick/Java/Control-M trigger additions and were being skipped at load. All capabilities/triggers preserved. - Discoverability: cross-linked
vscode-agentsfromclaude-code-cli(its VS Code GUI counterpart) andmcp-server-creator(the consumer side — using a built MCP server inside a VS Code agent).
- New skill
vscode-agents: build & use AI agents and agentic flows in VS Code — agent mode, custom agents (.agent.mdpersonas with their own tools/model/handoffs), MCP servers (.vscode/mcp.json), prompt files, instructions, hooks, and the Copilot cloud/coding agent. Includes multi-model orchestration (a main driver + second-opinion/verifier subagents each on a different model — the VS Code analogue of the Claude-CLI Claude+Codex+agy pattern, incl. the subagent cost-tier rule and BYOK for exact versions) and a security-first treatment (MCP trust prompts, themcp.jsontrust-bypass, sandboxing, least-privilege tools, hooks as guardrails). Researched against official VS Code + GitHub docs (source-verified, content treated as untrusted).
- Publish-config reproducibility + docs:
bootstrap-environment.pygains step 6c that restores your privatepublish-config.json(the scrub list) on a fresh machine if you version it underinstaller/— so a redeploy rebuilds the publish setup, not just the scripts (public users without one are pointed atinit-config). The Quick-Start step table now reflects steps 6b (memory seed) and 6c.
publish-to-githubships a generic gated scrubber framework: a self-containedstage_to_public.shdriver that hard-gates on the forbidden-pattern scrub verify (a leak can never reach the public repo) before mirroring + pushing. The framework (engine + gate + example config) is now reusable by anyone for their own public publishing — you bring your own private scrub list (publish-config.json:forbidden_patterns/scrubs/exclusions), which is never published.
- Memory + preferences subsystem: formalizes the two-tier file memory (global
~/.claude/memory/, loaded every session, layered with per-project memory) and adds a newuser-preferencesskill — durable, cross-project preference profiles (coding/presentations/email/tone) the agent loads before domain work and moulds to over time. Capture is explicit-only (recorded when you state a preference, never inferred). A_meta/memory_primer.pySessionStart hook reports which memory tiers loaded + the liveskills · agents · gatescount, andbootstrap-environment.pyseeds the memory tier + profiles on a fresh machine (never clobbering a populated profile).
publish-to-githubgainssync-metadata: a one-source reconciler that derives this catalog's**N skills**/**M agents**/**K workflows**counts, the Version/Last-published header, and the GitHub repo About (description + topics) from the live skill tree — so the published counts and the repo description can no longer silently drift.
- legacy-comprehension family — lineage v1:
mainframe-lineage-parsersgains a deterministic Control-M Automation-API jobs-as-code extractor (scheduler→program→data, job→job event DAG) alongside the existing COBOL/JCL/DB2 path; new OpenLineage facets (standardcolumnLineage 1-2-0, acontrolmDependenciesjob facet, and asourceCodeLocation.contentSha256cross-engine join key).lineage-extract-staticgains multi-abstraction views — L1 file-interaction (job-retained) + L2 table/column — as a singlereport.html3-tab switcher. Pick/MultiValue + Java coverage added across the family (LLM-path);legacy-code-intelgains a Java symbol addendum. - New skill
env-readiness: a read-only environment readiness doctor — reviews~/.claude+ per-project settings (skills/agents/workflows, SessionStart hooks, config, gates, identity, tooling) and prints a READY / READY-WITH-WARNINGS / NOT-READY verdict with repair pointers. Never mutates. - Dependency ergonomics: the lineage skills ship
requirements-optional.txt+ acheck_deps.pydoctor (--check-deps) + a "Dependencies & environments" guide (venv / PEP-668 / air-gapped / VS Code) — core stays pure-stdlib, enhancers are opt-in and never auto-installed. - AMY (pa-server) M0b: routine engine, role-lens, review-scopes, nudge-lifecycle, and the AMY brief hook with integration tests.
- Counts: 159 → 182 skills, 4 → 5 agents, 8 → 9 workflows.
- Established the curated skill/agent/workflow tree, the bootstrap installer, the security keystone (
llm-security,secret-scanning,sast-tooling,threat-modeling), the career family, and the legacy-comprehension siblings (lineage-extract-static,legacy-code-intel,structure-recovery).
This repository is a collection of prompts, documentation, and configuration designed to work with Claude Code CLI. It does not contain executable programs (beyond small helper scripts in some skills). Running the skills and agents requires Claude Code CLI and, for the full experience, additional third-party tools documented in docs/dependencies/.
The skills encode how to perform tasks — Claude reads them and acts on your behalf. You are responsible for reviewing Claude's actions before confirming destructive operations (deletes, force pushes, production deploys, etc.).