v1.0.0

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog,
and this project adheres to Semantic Versioning.

1.0.0 - 2026-05-22

Added

• Initial public release of the grokplan adversarial planning skill.
• Mechanical Compliance Requirements in launch.md (the contract copied into every run):
  • ROW_RE_READ — must re-read the current tracker row and quote a unique ≥12-word phrase from a high-value column before starting work on it.
  • TEMPTATION_LOG — must quote the exact literal text being reinterpreted before any deviation or reinterpretation.
  • Anti-Backfill Rule — all compliance outputs must appear before any implementation edits in the same response.
  • Periodic Contract Re-Engagement — re-read the core rules + Row Quality Standard and cite concrete proof from the live run state every 5–7 rows and at every session start/resumption.
  • Initial Execution Commitment ritual required in the very first response after receiving launch.md.
• Row Quality Standard (7-item checklist + explicit Anti-Gaming Criteria). Medium and Deep runs now require the Composer to produce rows that survive this standard and a structured self-assessment.
• Mandatory Row Quality Review artifact produced by the Orchestrator before Compose can close; surfaced to the user at the final Phase 5 Approval Gate.
• grokplan/launchers/verify-row.sh — minimal, copyable template for strong verification commands.

Changed

• Significantly strengthened planner-side requirements in 02-compose.md and orchestrator.md so that weak rows are rejected early.
• Discovery phase now explicitly pressures toward work that can be expressed as rows meeting the Row Quality Standard from the start.
• Public documentation (README, SKILL.md, plugin.json, examples) completely overhauled for honesty: direct tone, realistic example, clear limitations, no dev scaffolding language.
• Companion TUI source moved to experimental/companion/ and excluded from the public payload.

Fixed

• Removed cartoonish examples. Replaced with one grounded scenario ("Adding an export command to an existing CLI data processing tool") showing before/after.

Notes

This is the first public release. The focus is raising the cost of weak planning and execution drift for long-horizon LLM (and human) work. It does not claim to eliminate fidelity problems — it makes them materially more expensive while staying practical to follow.

The Companion TUI is experimental, not required for value, and not included in this release. A production TUI (ratatui on stable Rust) is planned for later.