skill-self-improvement-verificator

name	skill-self-improvement-verificator
description	Meta-skill for verifying Framework Upgrades (Gemini Self-Improvement).
tier	3
version	1

Skill: Self-Improvement Verificator

[!IMPORTANT] TIER 3 (Meta-Skill): Use ONLY when creating or modifying System/Agents, System/Docs, or GEMINI.md.

1. Purpose

This skill acts as a Guardian for the Framework itself. Before any changes are applied to the "Brain" of the IDE (Prompts, Skills, Docs), this skill verifies that the proposed changes are safe, consistent, and do not break the "Constitutional" rules (Tier 0 skills).

2. Modes

This skill operates in two distinct modes depending on the current Stage.

Mode A: SPECIFICATION AUDIT (Analysis Phase)

Trigger: When docs/TASK.md proposes a Framework Upgrade. Action: Verify the Specification.

Checklist:

1. Root Integrity: Does the update respect skill-core-principles (Stub-First, Atomicity)?
2. Skill Compatibility: Do new Agents/Prompts explicitly load TIER 0 skills?
3. Documentation: Does the task include updating System/Docs/ to reflect changes?
4. Migration: Does the task describe how to migrate existing sessions (if applicable)?

Mode B: PLAN AUDIT (Planning Phase)

Trigger: When docs/PLAN.md details the implementation steps. Action: Verify the Execution Plan.

Checklist:

1. Verification Step: Is there an explicit step to run pytest or validation_scripts?
2. Rollback: Is there a backup step (e.g., cp GEMINI.md GEMINI.bak)?
3. Atomic Updates: Are changes broken down into safe, verifiable chunks?
4. Test Coverage: Does the plan include adding/updating tests for new framework features?

3. Usage (Workflow Integration)

This skill is primarily called by the /framework-upgrade workflow.

# Example Usage in Workflow
- Validator: `spec-validator` (Standard)
- **Meta-Validator**: `skill-self-improvement-verificator` (Framework Safety)

4. Failure Conditions (Blocking)

If ANY of the following are found, STOP and return a CRITICAL failure:

• Removing skill-core-principles or skill-safe-commands from any Agent.
• Modifying GEMINI.md without a corresponding update to System/Docs.
• Creating a new Workflow without defining its Trigger in GEMINI.md.

5. Output

The output of this skill is a Critique Artifact:

• docs/reviews/framework-audit-[ID].md

Resources

• Template: assets/audit_template.md
• examples: examples/audit_examples.md (See Good/Bad patterns)

6. Emergency Bypass Protocols

In rare cases (e.g., hotfixes, core refactoring), checks may be bypassed. REQUIREMENT: ANY bypass must be explicitly justified in the Audit Artifact.

Flag	Effect	Use Case
[BYPASS_TIER_PROTECTION]	Allows modifying TIER 0 skills (Core Principles).	Upgrading the Core Framework.
[BYPASS_DOCS_CHECK]	Skips documentation verification.	Emergency Hotfixes.
[OVERRIDE_VERIFICATION]	Force-Approves the Audit despite failures.	False positives or manual override.