skillZs
★ LIVE SKILL TAGS ★
>>> LIVE SKILLS INDEX <<<
* OPEN SOURCE *
NO LOGIN, NO TRACKING
※ REAL INSTALL DATA ※
← back to all skills
hunter-leo/llm-agent-oop-coding-skill28 installs

spec-coding-skill

Guides LLM agents through large-scale coding tasks using a spec-driven, phase-by-phase methodology with round-based iterative execution. Covers requirement definition, planning, algorithm design, implementation, deviation handling, and round-to-round issue tracking. Includes OOP/SOLID principles, language-specific coding standards, a two-layer state machine (requirement lifecycle + round execution) for flow decisions, and trigger-driven extended methodologies for TDD, structured debugging, architecture deepening, terminology alignment, and dual-axis review. Keywords: spec-driven development, round-based execution, iterative development, deviation protocol, state machine, coding methodology, software engineering workflow, requirements-to-code pipeline, project blueprint, TDD, debugging, architecture review, method selection.

How do I install this agent skill?

npx skills add https://github.com/hunter-leo/llm-agent-oop-coding-skill --skill spec-coding-skill
view source ↗

Is this agent skill safe to install?

  • Gen Agent Trust Hubpass

    This skill defines a structured software development methodology for AI agents, covering requirements, planning, and execution. It incorporates security best practices such as prohibiting hardcoded credentials and enforcing testing standards. The primary security consideration is the inherent risk of indirect prompt injection, as the agent is instructed to read and interpret instructions from external project files and source code without explicit sanitization or boundary markers.

  • Socketpass

    No alerts

  • Snykpass

    Risk: LOW · No issues

What does this agent skill do?

Spec Coding Skill

A spec-driven methodology for LLM agents to define, plan, and implement coding work with quality and consistency. Uses round-based iterative execution — when problems arise during implementation, they are logged and resolved in subsequent rounds rather than silently deviating from the plan. Includes automatic task sizing — trivial fixes skip document ceremony, while complex features get the full structured flow.

STOP — Mandatory Startup Sequence

Do each step in order. Do not skip any step. Each step is a hard gate — the next step depends on information from the previous one.

Step 1 — Read Global Agent Rules (DO NOT SKIP)

MUST read 00-agent-execution.md before doing anything else. This file defines rules that override all phase-specific instructions: Task Sizing, Phase Transition Re-read Discipline, Self-Check, Deviation Protocol, and Round-based Execution Model. If you skip this file, you will miss mandatory rules and violate the process.

Step 2 — Check Project State

Check if .dev/blueprint.md exists:

Blueprint exists?Action
YesRun the Session Bootstrap. Present full project status, then resume the active requirement.
NoProceed to Step 3 (new requirement).

Step 3 — Task Sizing (MANDATORY)

Run Task Sizing to determine workflow depth. Announce the size to the user before proceeding.

SizeWorkflow
XS (1 file, trivial)Brief init.md with inline plan → implement → test → commit
S (1-3 files, well-defined)Lightweight init.md (Spec + Requirements + Plan) → implement → test → commit
M (multi-module)Full Phase 01–07 flow
L/XL (cross-requirement)Full flow + Blueprint layer + Phase 02 mandatory

Step 4 — Phase Transition Rules (applies to ALL phases)

Before entering ANY phase, you MUST:

  1. Re-read the phase reference file — see Phase Transition Re-read Discipline for the exact file per phase. Never enter a phase cold.
  2. Run Method Selection — evaluate every method tagged for that phase against trigger conditions. Apply if triggers match; provide specific factual justification if they don't. See Method Selection below. Generic "not needed" is invalid.

Step 5 — Enter Phase 01

Follow 00-initialization.md. XS/S: Pre-flight Checks then lightweight init.md. M+: full init.md with Spec, Requirements, Action Items, Constitution.

Step 6 — Work Through Phases in Order

Proceed through Phase 02–07. At each phase entry, repeat Step 4 (re-read phase doc + run Method Selection). After Phase 07, perform Round Review (Step 7).

Step 7 — Round Review (MANDATORY after Phase 07)

Check issues.md. Open issues → re-enter Phase 01 for Round N+1. No open issues → done.

Execution is round-based: problems found during a round are captured in issues.md and resolved in the next round, not silently fixed mid-flight. See 01-round-mechanism.md for the state machine and decision matrix.

All generated documents go under .dev/[NNN]-[req-name]/ in the project root.

Phase Overview

              ┌──────────────────────────┐
              │    Task Sizing           │  ← mandatory first step
              │  XS → direct implement   │     (see 00-agent-execution.md)
              │  S  → minimal init.md    │
              │  M+ → Phase 01-07 below  │
              └──────────────────────────┘
                          │ (M+ only)
                          ▼
Phase 01 — Initialization              (required for M+)
              ↓
Phase 02 — Prerequisite Tasks          (optional, LLM judges)
Phase 03 — Algorithm Design            (optional, LLM judges)
              ↓
Phase 04 — Implementation Plan         (required for M+)
              ↓
Phase 05 — Task Planning               (required for M+)
              ↓
Phase 06 — Create start-and-resume.md  (required for M+)
              ↓
Phase 07 — Execution                   (required for M+)
              │
              ▼
     Round Review                      (check issues.md)
         │
    ┌────┴────────┐
    │             │
  open          no open
  issues        issues
    │             │
    ▼             ▼
 Phase 01*     ✅ Done
 (Round N+1)
    │
    └──→ back to Phase 01-07
         (next round)

Phases 02 and 03 are independent (parallel, not sequential) — both, one, or neither may be needed.

Phase Reference

PhaseOutputReference
Blueprint layer.dev/blueprint.md05-blueprint-management.md
01 Initializationinit.md00-initialization.md
02 Prerequisite Tasksinspect.md / research.md / profiling.md / diagnosis.md01-prerequisite-tasks.md
03 Algorithm Designalgorithm-design.md02-algorithm-design.md
04 Implementation Planplan.md03-implementation-plan.md
05 Task Planningtasks.md04-task-planning.md
06 Create start-and-resume.mdstart-and-resume.md00-start-and-resume.md § Step 0
07 Executioncode00-start-and-resume.md
Round Reviewissues.md + Round History + N+1 loop01-round-mechanism.md

Blueprint: .dev/blueprint.md tracks all requirements, phases, and dependencies. Updated at every phase transition. See 05-blueprint-management.md.

Coding Reference

TopicReference
OOP & SOLID Principles01-oop-principles.md
Coding Standards (all languages)02-coding-standards.md
Git Workflow03-git-workflow.md
Round Mechanism01-round-mechanism.md
Reader's GuideREADME.md

Methods

MethodPhaseWhen to UseReference
Term Grilling + ADR01 / 02 / 04Fuzzy terminology, hard-to-reverse decisions00-term-grilling-and-adr.md
Vertical Slice TDD07Tasks with clear public interfaces01-vertical-slice-tdd.md
Dual-Axis ReviewRound Complete / PRSignificant code reviews02-dual-axis-review.md
Architecture Deepening02Modifying code with shallow modules03-architecture-deepening.md
Structured Debugging02Non-trivial bug fixes04-structured-debugging.md

Method Selection — Trigger-Driven

Methods are gate-checked by trigger conditions, not optional suggestions. At each phase entry, the agent MUST:

  1. Identify all methods tagged for that phase (see the table above)
  2. Read each method's ## When to Use / ## Do Not Use sections from its standalone file
  3. Evaluate trigger conditions against current context
  4. Apply if triggers match; produce a specific factual justification if they do not

The Method Selection table is produced as a planning artifact at every applicable phase entry. Skipping without justification is not permitted. The trigger conditions are defined in each method's documentation — the phase docs do not duplicate them.

Core Rules

Rules unique to this skill. See 00-agent-execution.md for: § File Reading Discipline (read before edit), § Self-Check (tests/quality), § Phase Transition Re-read Discipline (always re-read phase doc before starting), § Round-based Execution Model (issues.md gating), § Design Principle Compliance (OOP/SOLID).

  • Minimize changes when project live; breaking changes OK for pre-launch + new modules
  • All identifiers, comments, docs in English
  • Maintain the blueprint — update .dev/blueprint.md at every phase transition
  • Start with references/README.md — reader's guide before diving into phase docs
  • Run Method Selection at every phase entry — evaluate tagged methods against trigger conditions. Apply if triggers match; justify if they don't.
  • ⚠ Follow the Deviation Protocol — when implementation differs from plan.md: stop, assess severity, log to issues.md, present options to user. Never silently modify plan.md, add unplanned files, or change interfaces without logging the deviation. See execution/00-start-and-resume.md § Deviation Protocol.

Add the canonical catalog link to the repository README so users can inspect current installs and available audits. The publishing guide covers the complete discovery path.

<a href="https://skillzs.dev/skills/hunter-leo/llm-agent-oop-coding-skill/spec-coding-skill">View spec-coding-skill on skillZs</a>