Agent Skill Evolution v2.1.0

Agent Skill Evolution v2.1.0

This release moves the public abstraction from infrastructure-shaped runtime management to the native Agent Skill lifecycle.

Highlights

• Adds evolve-skills for portfolio inventory, canonical-source selection, updates, deduplication, evaluation handoff, and removal.
• Retires the agent-runtime Skill. The internal agent_runtime package and launchd labels remain stable for v2 compatibility.
• Clearly separates Google ADK, native Agent Skills, the optional support runtime, Buildomator task state, ReMe memory, and policy authorization.
• Adds original discover/test/evolve/remove artwork and an explicit Riot/Google non-affiliation boundary.
• Preserves the existing Hook, deterministic policy, bounded ReMe, immutable release, backup, and rollback implementation.

Upgrade

npx skills@latest add hanzw/agent-skill-evolution \
  --skill evolve-skills skill-governance first-principles-checkpoint \
  --global --agent codex claude-code --yes

npx skills@latest remove agent-runtime \
  --global --agent codex claude-code --yes

Verification

• 40 Python tests passed locally and in GitHub Actions.
• All three native Skills passed quick_validate.py.
• Native Skill discovery found exactly three Skills.
• Promptfoo configuration validation passed.

SHA-256 checksums are attached alongside the source archive.

Agent Runtime v2.0.0

Complete macOS-first runtime for Codex + Claude.

Included:

• shared lifecycle Hooks and deterministic side-effect policy
• bounded local ReMe memory with project isolation
• atomic install, source provenance, backups, rollback, and health heartbeat
• agent-runtime, first-principles-checkpoint, and skill-governance Skills
• Promptfoo ablation template, CI, architecture, security policy, and banner

The former v1 skill-only repository has been renamed and expanded. Personal project rules, legacy log migration, copied upstream Skills, model routing, and shadow registries are deliberately excluded.

Install dependencies are pulled directly from PyPI, GitHub, and the native Skills manager as documented in README.

Skill Governance v1.0.0

Minimal Agent Skill lifecycle governance.

Install pulls directly from canonical sources:

• promptfoo/promptfoo: eval authoring and provider setup
• hanzw/skill-governance: keep/update/remove policy and ablation template

Architecture: native Skill discovery is canonical; ReMe stores durable history; Buildomator/HANDOFF stores long-task state; Promptfoo performs controlled ablation. No shadow registry or custom telemetry stack.