Eniac

Eniac is a provider-neutral, agent-agnostic operating skill for token-aware AI agents. It helps agents handle simple tasks without overthinking, scale up to complex autonomous work, stay scoped, verify results, and avoid wasting tokens or credits.

Credit: ParamissionLab

What It Does

• Runs compact Perceive -> Reason -> Act -> Observe loops with cause, coverage, and documentation-impact checks on material signals.
• Scales effort from tiny edits (L0) to complex architecture (L3) and high-risk production work (L4).
• Keeps context use lean with cost guards, targeted reads, batching, and narrow verification first.
• Supports senior software engineering workflows: staged Greenfield launches, project-state detection, codebase audits, stack-risk analysis, restartable execution ledgers, evidence-backed project interfaces, CI proof graphs, full-cycle build/test/bug-hunt/polish/docs/ship, deep codebase audits, whole-repo code reads when requested, UI/UX/frontend quality, prompt or agent workflow design, and many-small-edit batches.
• Uses dependency-aware workstreams for difficult multi-surface work, while keeping one integration owner, explicit mutable scope, acceptance proof, and recovery gates.
• Requires a causal record before closing defects; emergency containment is clearly separated from the systemic correction.
• Provides ready-to-adapt CI templates for 8 stacks (Node.js, Python, Go, Rust, Zig, Ruby, Java, .NET) plus release workflows.
• Includes concrete code pitfalls and good patterns for 11 stacks with bad-to-good examples.
• Ships README and plan templates for consistent, high-quality documentation and planning.
• Handles Thai and other token-expensive languages with compact planning while preserving exact paths, commands, identifiers, and quotes.
• Loads deeper references only when needed (20 reference files, loaded on demand).

Structure

skills/
  Eniac/
    SKILL.md
    scripts/
      finalize_plan.py
    tests/
      test_finalize_plan.py
    references/
      ci-templates-by-stack.md
      architecture-evidence.md
      algorithm-workflow.md
      commands-by-stack.md
      delivery-proof-pipelines.md
      execution-ledger.md
      greenfield-launch-sequence.md
      loop-engineering.md
      multilingual-token-discipline.md
      plan-template-extended.md
      platform-portability.md
      product-ux.md
      project-interface-contract.md
      readme-template.md
      safety-and-delivery.md
      software-engineering.md
      stack-patterns-and-pitfalls.md
      stack-risk-matrix.md
      systematic-thinking.md
      workstream-orchestration.md

SKILL.md is the lightweight control plane. The 20 files in references/ are loaded only when deeper task-specific guidance is needed:

Category	Files	Loaded when
Core engineering	software-engineering.md, commands-by-stack.md	Non-trivial code work
Architecture	architecture-evidence.md	Existing-system architecture work, non-local changes, blast-radius analysis, or handoff
Complex workflow	workstream-orchestration.md	Difficult multi-workstream work, dependency waves, safe parallel execution, integration, or durable task context
Algorithmic quality	algorithm-workflow.md	L2+ execution loops requiring root-cause diagnosis, work weighting, coverage, or documentation-impact discipline
Actionable templates	ci-templates-by-stack.md, readme-template.md, plan-template-extended.md	Creating CI, README, or plans for broad work
Stack knowledge	stack-risk-matrix.md, stack-patterns-and-pitfalls.md	Stack-specific correctness risk
Greenfield	greenfield-launch-sequence.md, execution-ledger.md	New project or broad milestoned work
Safety	safety-and-delivery.md, delivery-proof-pipelines.md	L4, CI, dependencies, production
Specialized	product-ux.md, loop-engineering.md, systematic-thinking.md, multilingual-token-discipline.md	UI/UX, autonomy, complex decomposition, multilingual
Meta	project-interface-contract.md, platform-portability.md	README rewrites, platform installation

Installation

See skills/Eniac/references/platform-portability.md for current installation paths across supported platforms (OpenAI Codex, Anthropic Claude Code, Google Gemini CLI, GitHub Copilot, OpenCode, Cline, Windsurf, Zed, Augment, and others).

Quick approach: copy or link the entire skills/Eniac/ folder into your agent's native skills directory, or use the shared .agents/skills/eniac/ path where supported. If your runtime does not support skill loading, paste the core contract from SKILL.md into your agent instructions and point the agent at the reference files as optional material.

Usage

Use the eniac skill when you want:

• autonomous work that should continue until done or genuinely blocked
• strict token, credit, context, or scope control
• software engineering, debugging, review, planning, architecture, or research
• UI/UX/frontend work that should fit the product type, visual system, and real user workflow
• many small edits that should be batched safely
• Thai or multilingual work that should stay concise

Example prompts:

Use eniac to fix this bug, verify it, and keep changes scoped.

Use eniac to improve this dashboard UX for real daily operations without making it look like a landing page.

Use eniac to tighten this app UI: layout, controls, states, responsive behavior, and accessibility, without adding visual noise.

ใช้ eniac แก้หลายไฟล์ตาม pattern เดียวกันแบบประหยัด token และตรวจให้พอมั่นใจ

Token And Credit Discipline

Eniac is designed to reduce waste:

• simple work should not load extra references
• large files should be searched or read by targeted ranges first
• repeated edits should be batched by one safe rule
• concurrent work should begin only after contracts, mutable scopes, and integration proof are explicit
• L3 work should keep a compact durable execution context during execution so a later session can resume from verified facts; plans are deleted on completion by default unless retention is explicitly requested
• each material loop should reweight the next action and assess applicable flow, contract, failure, operations, and documentation coverage
• a green symptom check is not a final fix unless the causal record is proven or its uncertainty and owner are explicit
• installed references should stay specific so UI/UX, software, loop, and multilingual guidance load only when they can change the result
• broad tests, browser checks, web lookups, and subagents should run only when they can change the decision or risk report
• final reports should summarize changed behavior, verification, risks, and next steps without long reasoning traces

Updating

git -C Eniac pull
# Then copy the skills/Eniac folder to your agent's skill directory
# See platform-portability.md for your target platform path

Restart your agent or start a new session after updating.

Validation

python skills/Eniac/tests/test_finalize_plan.py

Versioning

This project follows semantic versioning where practical:

• Patch versions fix wording, structure, or small behavioral issues.
• Minor versions add compatible guidance or references.
• Major versions may change the skill contract or expected agent behavior.

Contributing

See CONTRIBUTING.md.

Security

See SECURITY.md.

Code of Conduct

See CODE_OF_CONDUCT.md.

License

MIT. See LICENSE.