Phase Context Loading Protocol
This skill defines skill loading tiers to optimize token consumption while preserving automation capabilities.
[!IMPORTANT]
Single Source of Truth for Skill Loading Strategy.
Reference this skill to understand WHEN to load WHICH skills.
TIER 0: System Foundation — ALWAYS LOADED
[!CAUTION]
NEVER lazy-load TIER 0 skills. They enable core automation.
These skills MUST be available in EVERY session at bootstrap.
| Skill |
Tokens |
Why Always Load |
core-principles |
~519 |
Anti-hallucination rules, Stub-First, Documentation First |
skill-safe-commands |
~927 |
Enables automation — mv, ls, git, tests auto-run |
artifact-management |
~636 |
Archiving protocol, file management, dual state tracking |
| TOTAL |
~2,082 |
Non-negotiable system foundation |
TIER 0 Loading Rule
- Load at session bootstrap (via
GEMINI.md / .cursorrules)
- These skills are loaded BEFORE any phase begins
- Never remove or defer loading of TIER 0 skills
TIER 1: Phase-Triggered — Load on Phase Entry
These skills are phase-specific and loaded when entering the corresponding phase.
| Phase |
Skills to Load |
Approx. Tokens |
| Analysis |
requirements-analysis, skill-task-model, artifact-formalizer (contract only), (skill-archive-task if new task) |
~3,100-4,000 |
| Architecture |
architecture-design, artifact-formalizer (contract only), (architecture-format-core OR -extended) |
~2,850-4,200 |
| Planning |
planning-decision-tree, skill-planning-format, tdd-stub-first, artifact-formalizer (contract only) |
~2,525 |
| Development |
developer-guidelines, documentation-standards |
~1,100 |
| Review |
Phase-specific -review-checklist skill |
~300-400 |
| Security |
security-audit |
~500 |
| VDD |
vdd-adversarial, (adversarial-* per workflow) |
~1,000-2,500 |
TIER 1 Loading Conditions
Analysis Phase:
requirements-analysis — When 02_analyst activatedskill-task-model — When creating/updating TASK.mdskill-archive-task — When existing TASK.md detected AND new (unrelated) task
Architecture Phase:
architecture-design — When 04_architect activatedarchitecture-format-core — Default for most updatesarchitecture-format-extended — Load ONLY for:
- Creating NEW system from scratch
- Major refactor (>3 components affected)
- User explicitly requests "full architecture template"
Planning Phase:
planning-decision-tree — When 06_planner activatedskill-planning-format — When creating PLAN.mdtdd-stub-first — When stub creation starts
Development Phase:
developer-guidelines — When 08_developer activateddocumentation-standards — When updating docs
Review Phase:
- Load the appropriate
*-review-checklist for the current review type
Every authoring phase (Analysis, Architecture, Planning):
artifact-formalizer — load references/authoring-contract.md before the first sentence of
any artifact is written. Load the SKILL.md body and the scanner only when auditing an artifact
that already exists.- Why. The contract is ~1,100 tokens; auditing an artifact afterwards costs a full re-read of
that artifact plus the rewrite. Measured cost of the late order:
artifact-formalizer/references/measurement-baseline.md §5.
TIER 2: Extended — Load Only When Explicitly Requested
These skills are specialized and loaded only when explicitly invoked by user or workflow.
| Skill |
Tokens |
Loading Condition |
architecture-format-extended |
~3,000 |
New system / major refactor / user requests full template |
skill-reverse-engineering |
~1,322 |
User requests "sync docs with code" / docs-code mismatch |
skill-update-memory |
~1,101 |
Post-development phase, .AGENTS.md updates needed |
skill-adversarial-security |
~838 |
VDD workflow with security focus |
skill-adversarial-performance |
~946 |
VDD workflow with performance focus |
vdd-sarcastic |
~145 |
VDD workflow with sarcastic mode |
Integration
How Agents Should Reference This
Agent prompts should:
- Always load TIER 0 skills at session start
- Load TIER 1 skills when entering their specific phase
- Load TIER 2 skills only when explicitly triggered
Referenced By
.gemini/GEMINI.md — TIER 0 at bootstrap.cursorrules — TIER 0 at bootstrap- All agent prompts — TIER 1 at phase entry
- VDD workflows — TIER 2 when invoked
1---2name: skill-phase-context3description: Skill loading tiers: TIER 0 (always), TIER 1 (phase-triggered), TIER 2 (extended). Defines when to load which skills.4---5# Phase Context Loading Protocol67This skill defines **skill loading tiers** to optimize token consumption while preserving automation capabilities.89> [!IMPORTANT]10> **Single Source of Truth for Skill Loading Strategy.**11> Reference this skill to understand WHEN to load WHICH skills.1213## TIER 0: System Foundation — ALWAYS LOADED1415> [!CAUTION]16> **NEVER lazy-load TIER 0 skills. They enable core automation.**17> These skills MUST be available in EVERY session at bootstrap.1819| Skill | Tokens | Why Always Load |20|-------|--------|-----------------|21| `core-principles` | ~519 | Anti-hallucination rules, Stub-First, Documentation First |22| `skill-safe-commands` | ~927 | **Enables automation** — `mv`, `ls`, `git`, tests auto-run |23| `artifact-management` | ~636 | Archiving protocol, file management, dual state tracking |24| **TOTAL** | **~2,082** | **Non-negotiable system foundation** |2526### TIER 0 Loading Rule27- Load at session bootstrap (via `GEMINI.md` / `.cursorrules`)28- These skills are loaded BEFORE any phase begins29- Never remove or defer loading of TIER 0 skills3031---3233## TIER 1: Phase-Triggered — Load on Phase Entry3435These skills are phase-specific and loaded when entering the corresponding phase.3637| Phase | Skills to Load | Approx. Tokens |38|-------|----------------|----------------|39| **Analysis** | `requirements-analysis`, `skill-task-model`, `artifact-formalizer` (contract only), (`skill-archive-task` if new task) | ~3,100-4,000 |40| **Architecture** | `architecture-design`, `artifact-formalizer` (contract only), (`architecture-format-core` OR `-extended`) | ~2,850-4,200 |41| **Planning** | `planning-decision-tree`, `skill-planning-format`, `tdd-stub-first`, `artifact-formalizer` (contract only) | ~2,525 |42| **Development** | `developer-guidelines`, `documentation-standards` | ~1,100 |43| **Review** | Phase-specific `-review-checklist` skill | ~300-400 |44| **Security** | `security-audit` | ~500 |45| **VDD** | `vdd-adversarial`, (`adversarial-*` per workflow) | ~1,000-2,500 |4647### TIER 1 Loading Conditions4849**Analysis Phase:**50- `requirements-analysis` — When 02_analyst activated51- `skill-task-model` — When creating/updating TASK.md52- `skill-archive-task` — When existing TASK.md detected AND new (unrelated) task5354**Architecture Phase:**55- `architecture-design` — When 04_architect activated56- `architecture-format-core` — **Default** for most updates57- `architecture-format-extended` — Load ONLY for:58 - Creating NEW system from scratch59 - Major refactor (>3 components affected)60 - User explicitly requests "full architecture template"6162**Planning Phase:**63- `planning-decision-tree` — When 06_planner activated64- `skill-planning-format` — When creating PLAN.md65- `tdd-stub-first` — When stub creation starts6667**Development Phase:**68- `developer-guidelines` — When 08_developer activated69- `documentation-standards` — When updating docs7071**Review Phase:**72- Load the appropriate `*-review-checklist` for the current review type7374**Every authoring phase (Analysis, Architecture, Planning):**75- `artifact-formalizer` — load `references/authoring-contract.md` **before** the first sentence of76 any artifact is written. Load the SKILL.md body and the scanner only when auditing an artifact77 that already exists.78- **Why.** The contract is ~1,100 tokens; auditing an artifact afterwards costs a full re-read of79 that artifact plus the rewrite. Measured cost of the late order:80 `artifact-formalizer/references/measurement-baseline.md` §5.8182---8384## TIER 2: Extended — Load Only When Explicitly Requested8586These skills are specialized and loaded only when explicitly invoked by user or workflow.8788| Skill | Tokens | Loading Condition |89|-------|--------|-------------------|90| `architecture-format-extended` | ~3,000 | New system / major refactor / user requests full template |91| `skill-reverse-engineering` | ~1,322 | User requests "sync docs with code" / docs-code mismatch |92| `skill-update-memory` | ~1,101 | Post-development phase, .AGENTS.md updates needed |93| `skill-adversarial-security` | ~838 | VDD workflow with security focus |94| `skill-adversarial-performance` | ~946 | VDD workflow with performance focus |95| `vdd-sarcastic` | ~145 | VDD workflow with sarcastic mode |9697---9899## Integration100101### How Agents Should Reference This102103Agent prompts should:1041. Always load TIER 0 skills at session start1052. Load TIER 1 skills when entering their specific phase1063. Load TIER 2 skills only when explicitly triggered107108### Referenced By109- `.gemini/GEMINI.md` — TIER 0 at bootstrap110- `.cursorrules` — TIER 0 at bootstrap111- All agent prompts — TIER 1 at phase entry112- VDD workflows — TIER 2 when invoked
MatrixFounder