137 lines (104 loc) · 7 KB

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.

0.6.0 - 2026-03-20

Added

/forge-cancel as the primary Claude Code stop command, with /cancel-ralph kept as a legacy alias

Changed

Forge now uses FORGE_COMPLETE and FORGE_PAUSE as the primary control markers while still accepting legacy Ralph markers
Claude driver docs and install flow now use Forge-first naming and forge-loop.*.local.md as the primary loop-state file name
README positioning now keeps Ralph and other influences in an explicit lineage section instead of mixing them into the main product surface

0.5.1 - 2026-03-20

Fixed

Claude stop hook now resolves its real source path before loading shared helpers, so symlinked installs under ~/.claude/hooks/stop-hook.sh no longer fail with missing helper-library errors
Added regression coverage for running the stop hook through a symlinked install path

0.5.0 - 2026-03-20

Added

Optional explicit --done-when "TEXT" success contract for Forge task scopes

Changed

Forge now treats open-text task completion as the primary objective and KPI targets as guardrails
Codex driver status and prompt rendering now surface success mode and done_when text
README and driver docs now describe task-derived completion checks when no explicit success override is provided

0.4.2 - 2026-03-20

Added

forge-status surfaces for both Claude Code and Codex/manual drivers
Shared shell state helpers in scripts/forge-state-lib.sh

Fixed

Codex driver tests now cover malformed state and max-iteration edge cases

Changed

Claude stop hook and Codex driver now share the same frontmatter parsing helpers

0.4.1 - 2026-03-20

Fixed

Codex driver now derives the next iteration from recorded Forge state instead of blindly incrementing loop metadata
Codex driver now fails closed when multiple active sessions exist and no explicit session id is provided

Changed

Extracted shared Codex driver state helpers into drivers/codex/lib.sh
Expanded Codex driver regression tests to cover repeated continue calls and multi-session ambiguity

0.4.0 - 2026-03-20

Added

First-class Codex/manual driver with forge-init, forge-continue, and forge-cancel
install-codex.sh for linking Forge Core and Codex driver entrypoints into ~/.codex
Codex driver docs, prompt template, and regression tests

Changed

README and Forge Core docs now describe two shipped drivers instead of one
Support matrix now treats Codex as a bundled manual driver rather than protocol-only reuse
State docs now describe both .claude/ and .codex/forge/ driver roots

0.3.0 - 2026-03-20

Changed

Reframed Forge as Forge Core plus a first-class Claude Code driver instead of implying universal runtime support
Added an explicit support matrix covering Claude Code, Codex/manual protocol reuse, and other runtimes
Clarified claims across README, skill, hook docs, and agent surfaces so portability and support are described honestly

Added

Core-vs-driver architecture explanation in the protocol and README
Manual/protocol-only usage guidance for Codex and other non-Claude runtimes
Contribution guidance for new runtime drivers

0.2.1 - 2026-03-20

Added

/cancel-ralph command for stopping the active loop in the current project without deleting forge-state
tests/stop-hook.test.sh regression coverage for completion markers, pause markers, paused state handling, and completion promises

Fixed

Stop hook now requires exact control markers on their own line instead of matching substrings inside normal prose
Paused loops are ignored until they are explicitly resumed, instead of being silently reclaimed by the next session
Installer now links the /cancel-ralph command alongside /forge

Changed

Forge docs and skill instructions now describe exact control-marker semantics instead of implying the hook validates KPI targets itself
Fresh-context audit guidance now refers only to agents/personas available in the caller's environment
Failure recovery guidance now forbids whole-worktree reverts and limits cleanup to files changed in the current iteration

0.2.0 - 2026-03-20

Added

Simplicity criterion in DECIDE phase — code deletion at same KPIs is always a win; marginal gains from complexity are rejected
simplification strategy — dedicated strategy for reducing code complexity
Clean revert on failure — explicit git checkout to restore clean state between iterations
Ideas backlog in forge-state — captures deferred opportunities for future iterations
Getting Unstuck protocol — re-read scope, review backlog, combine near-misses, try the inverse, simplification pass
"Never stop to ask" rule — agent thinks harder instead of pausing for user input
Design Principles section in README — 8 principles distilled from studying autoresearch, Ralph Wiggum, pi-autoresearch, SICA, and forks
Deep research — 6 research documents in .research/ analyzing primary sources

Changed

Expanded credits with proper links and specific contributions from each influence
Strategy table now includes simplification (8 strategies total)
Critical Rules expanded from 7 to 10

0.1.0 - 2026-03-20

Added

The Forge Protocol — eight-phase iteration cycle (Orient, Measure, Evaluate, Decide, Execute, Verify, Record, Complete)
7 named strategies with automatic selection based on normalized KPI gaps
Stagnation detection and automatic strategy rotation after 3 low-delta iterations
Fresh-context evaluation via subagents every 3rd iteration (prevents anchoring bias)
Autoregressive state file (.claude/forge-state.SESSION.md) that persists KPIs, strategies, and lessons across iterations
Stop hook for iteration engine (compatible with Ralph Wiggum loops)
/forge command with --coverage, --speed, --quality, and --max-iterations options
Forge agent for spawning as a subagent on subsystems
Installer script with symlink-based setup
Multi-language support in MEASURE phase (Elixir, Python, JavaScript, Ruby, Go)
Simultaneous multi-KPI completion gate