v2.0.0 — Deep-Harness

This release promotes the former prompt-upgrade package to Deep-Harness: a Codex skill for rigorous artifact improvement and AI-agent harness engineering. It replaces mode-based routing with one evidence-bounded, prompt-mediated lifecycle.

Highlights

• Renamed the skill to deep-harness / Deep-Harness.
• Narrowed activation: explicit invocation, substantive improvement, explicitly requested rigorous reasoning, or AI-agent harness engineering.
• Removed normal/thorough mode selection and universal default-on behavior.
• Added the evidence-bounded Intent Brief and direction-changing clarification threshold.
• Added AI-authored, linted Review Prompt Packets and Edit Prompt Packets.
• Added end-to-end traceability, Work Filter convergence, delegated implementation safeguards, ledger discipline, and independent verification.
• Removed the obsolete zero-skip companion after merging its safeguards into the current contract.
• Replaced the repository documentation with Codex/Deep-Harness installation and upgrade guidance.

Upgrade

cd ~/.codex/skills/deep-harness
git pull origin main

If your Codex installation uses the personal agent-skill directory, update from ~/.agents/skills/deep-harness instead.

v1.3.0 — zero-skip thorough doctrine

Summary

This release makes thorough and complete skill compliance paths blocking law, not optional depth. Agents can no longer skip orchestra, in-session ledger, or Gate 3 delegation to ship faster.

Highlights

• New reference-zero-skip-compliance.md — pinned execution path, pre-write tool guard, violation recovery, parent parity
• SKILL.md zero-skip thorough doctrine — restrictions, not recommendations; parent implement ban when Task exists
• Complete skill compliance triggers — “follow skill completely”, “zero skip”, “execute all processes”, “skill compliance architect”
• Expanded gate checklist, failure modes #6–8, PRE-FLIGHT item #8 COMPLIANCE
• Seven markdown files ship together (update install/copy instructions)

Upgrade

cd ~/.cursor/skills/prompt-upgrade
git pull origin main

v1.2.0 — Slot D Skill Orchestration

Summary

This release adds Slot D support for installed Cursor Agent Skills, strengthens runtime prompt packets, and updates the package docs for v1.2.0.

What's New

• Added Slot D as an optional fourth composed-role slot for installed, fully-read Agent Skills.
• Added read-before-use rules: a Slot D skill must be installed, fully read, guideline-recorded, and relevant before use.
• Clarified Slot C vs Slot D:
  • Slot C = tools, MCP servers, plugins, hooks
  • Slot D = installed skill instruction packages
• Strengthened runtime prompt packets with purpose, capability boundary, Snapshot context, relevant skill instructions, evidence hooks, output contract, and Slot D guidelines when applicable.
• Added independent acceptance scoring governance with a >=90/100 per-dimension target, bounded by Work Filter and saturation stop rules.
• Expanded word-bank coverage for Laravel/Livewire/Filament admin, smart contracts, codebase onboarding, and agentic optimization.

Notes

prompt-upgrade remains the governing orchestration contract. Slot C tools and Slot D skills are managed capabilities and do not override gates, Work Filter, stop rules, system/developer instructions, or user constraints.

v1.1.0 — Execution gates & PRE-FLIGHT

Summary

Hardens agent compliance with the prompt-upgrade contract. Orchestra and runtime subagents are required before Implement—not optional background context.

Highlights

• NON-NEGOTIABLE EXECUTION GATES (Gate 0–4 + blocking checklist)
• PRE-FLIGHT mandatory before in-scope artifact file writes
• Common agent failure modes (hard stops) — five named drift patterns with recovery protocol
• New reference-execution-gates.md — Task patterns, wave ledger template, Gate 3 delegation
• In-session wave ledger required before Implement (not deferred to final reply)
• Runtime-first probe — Task tool must be used when available

Upgrade

cd ~/.cursor/skills/prompt-upgrade
git pull origin main

v1.0.0 — stable

Stable release — no further updates planned