/// system
Turn AI agents into
disciplined senior engineers.
57 composable skills wired into a single flow.
Use it now —
01 / the problem
AI agents are capable but undisciplined.
No tests before commit
They ship code that looks correct but breaks in production.
No review before push
They overwrite each other's work without noticing.
Generic output only
They produce the same patterns regardless of context.
When developers used AI, they took 19% longer. Developers estimated AI saved them 20% of the time. Experts predicted 38% faster.
— Becker et al., METR, July 2025
The gap between "capable" and "reliable" is not intelligence. It's process.
Why process matters →02 / how it works
Six phases. Every time.
Every project follows the same disciplined flow. No shortcuts. No skipped steps.
Write the spec before any code. Interview requirements. Lock scope.
spec-driven-development interview-me idea-refine
Break work into atomic tasks. Estimate effort. Identify dependencies.
planning-and-task-breakdown architecture-analysis
Implement one slice at a time. Test each before expanding.
incremental-implementation test-driven-development source-driven-development
Run tests. Check builds. Verify behavior matches spec.
test-driven-development debugging-and-error-recovery browser-testing-with-devtools
Code review. Security audit. Performance check. Quality gates.
code-review-and-quality security-and-hardening performance-optimization
Clean commits. CI/CD. Deployment. Documentation.
git-workflow-and-versioning ci-cd-and-automation shipping-and-launch
No phase is optional. Every step has a skill. Every skill has a gate.
03 / the harness
Rules that depend on memory fail.
Rules that depend on the harness succeed.
01: Instructions & Rules
Who the agent is, what it cares about, what it must never do. AGENTS.md, SOUL.md, STEERING-GUIDE.md.
02: Tools & Sandboxes
Task-specific capabilities loaded on demand. 57 skills, 54 guides, eval system, terminal, CI.
03: Orchestration & Guardrails
When tools fire, how agents coordinate, and deterministic hooks at lifecycle points. Pre-commit v8 (9 gates), commit-msg v5, commit-approval.sh.
04: Observability & Verification
Evidence it's working or quietly drifting. project-metrics, HEALTH-CHECK.md, validate-skill-table.sh, PROGRESS_STATUS.md.
It broke its own rules three times in two days. Then it built a harness that made breaking them impossible.
It learns from its own failures. Faster than competitors ship features.
04 / philosophy
Doesn't make agents faster.
Makes them reliable.
-
01
Quality emerges from explicit contracts, not implicit trust.
-
02
Rules that depend on memory fail. Rules that depend on visible blocks succeed.
-
03
A hook that accepts any token is a suggestion. A hook that verifies integrity is a gate.
-
04
Speed is a side effect of precision, never the primary goal.
-
05
Context is finite. Evict before you drown.
-
06
"No, that's overcomplicated" is more valuable than blind compliance.
-
07
Learns from its own failures faster than competitors ship features.
Read SOUL.md →Turns AI coding agents from obedient servants into disciplined senior engineers.
05 / compatible agents
Designed for OpenCode.
Portable to everything else.
One install. Any agent. Any stack.
- OpenCode
- Claude Code
- Cursor
- GitHub Copilot
- Kiro
- Gemini CLI
- Devin Desktop
- Any agent
Stack-agnostic.
Works with Node, Rust, Python, Go, Ruby, Dart, or any stack.
init-agents detects your stack automatically. Creates STACK_CONFIG.md.
06 / skills
Everything you need.
Nothing you don't.
Foundation
engineering-fundamentals frontend-ui-engineering context-engineering user-onboarding
Frontend
frontend-web frontend-mobile frontend-desktop frontend-pwa visual-frontend-mastery browser-testing-with-devtools
Backend
backend-api-mastery api-and-interface-design
DevOps
fullstack-shipping shipping-and-launch ci-cd-and-automation performance-optimization observability-and-instrumentation security-and-hardening
Process
spec-driven-development architecture-analysis planning-and-task-breakdown incremental-implementation documentation-and-adrs source-driven-development deprecation-and-migration doubt-driven-development multi-agent-orchestration customize-opencode
Quality
code-review-and-quality code-simplification project-health-check dev-environment-audit project-metrics
Design Review
critique-skill audit-skill clarify-skill hard-skill polish-skill typeset-skill adapt-skill optimize-skill delight-skill
Testing
test-driven-development
Meta-Skills
skill-creator skill-improver
Git
git-init-and-versioning git-workflow-and-versioning
Debug
debugging-and-error-recovery debugging-three-strikes
Ideation
interview-me idea-refine
Design Skins
industrial-brutalist-ui minimalist-ui soft-premium-ui output-skill redesign-skill
57 skills · 54 guides · 6 harness components · eval system
Browse all skills →07 / quick start
Start in seconds.
$ git clone https://github.com/juandelossantos/another-agent-skills.git
$ cd another-agent-skills
$ bash install.shThen in any project:
$ init-agentsYour agent now has 57 skills + 54 guides. Every project. Every time.
08 / faq
Questions.
What is Another Agent Skills?
A framework of 57 composable skills that turn AI coding agents into disciplined senior engineers. It adds mechanical enforcement, not just prompts. Skills follow a 6-phase lifecycle: Define, Plan, Build, Verify, Review, Ship.
How is it different from other skill frameworks?
Three things: (1) Mechanical enforcement via time-window approval and 9 pre-commit gates. (2) The Guardian Pattern: DECISION POINT before every mutation. (3) Context engineering: 258 lines always-loaded (1.9% of 200K context). Other frameworks give you prompts. This gives you process.
Which agents does it support?
Designed for OpenCode. Portable to Claude Code, Cursor, GitHub Copilot, Kiro, Gemini CLI, Devin Desktop, and any agent that reads AGENTS.md. Stack-agnostic: works with Node, Rust, Python, Go, Ruby, Dart, or any stack.
Is it free?
Yes. MIT License. Open source. No subscriptions. No paid tiers. The installer is free, the skills are free, the enforcement is free.
How does mechanical enforcement work?
Three levels: (1) Process rules: agent remembers. (2) Visible manifest: agent presents DECISION POINT. (3) Time-window approval: commit-msg hook v5 verifies freshness (<5 min). No tokens, no hash, no friction.
Can I customize the skills?
Yes. Skills are Markdown files in the `skills/` directory. Edit them, add new ones, remove ones you don't need. The framework is modular: use all 55 or pick the ones that matter to your workflow.
Can an agent still break the rules despite enforcement?
Yes. The hooks enforce git operations — they require tokens, and tokens require manifests. But no system can fully prevent an agent from acting against protocol. The agent could generate a token without presenting the manifest, interpret ambiguous responses as approval, or push without asking. The human must stay alert. These gates create friction, not guarantees.
What is the Harness?
The Harness is everything around the AI model that turns raw intelligence into reliable output. Six components: Instructions & Rules, Tools, Sandboxes, Orchestration, Guardrails & Hooks, and Observability. This project is a complete open-source Harness. See docs/HARNESS.md.
What is the Factory Model?
The Factory Model means the developer's primary output is not code — it's the system that produces code. You design the specifications, agents, tests, and quality gates. The agents handle implementation. You handle architecture, verification, and judgment.
What's the theoretical foundation?
The concepts are based on 'The New SDLC With Vibe Coding' by Addy Osmani, Shubham Saboo, and Sokratis Kartakis (May 2026) and 'Agent Skills' by Singhal et al. (Google, May 2026). This project is an open-source implementation of those principles.
How does the eval system work?
Each skill has eval files that test 4 failure modes: trigger, execution, token budget, and regression. Run bash scripts/eval/run-evals.sh --all to validate. See Evaluation Guide for full documentation.
references
Sources
- Becker, J., Rush, N., Barnes, E., & Rein, D. (2025). Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity. METR. arXiv:2507.09089
- Osmani, A., Saboo, S., & Kartakis, S. (2026). The New SDLC With Vibe Coding: From ad-hoc prompting to Agentic Engineering. Google. Google Drive
- Singhal, T., Hernandez Larios, G., Dus, D., Nigam, L., & Kolan, S. (2026). Agent Skills: From Prototype to Production. Google. Google Drive