Skip to content

Latest commit

 

History

History
99 lines (75 loc) · 9.16 KB

File metadata and controls

99 lines (75 loc) · 9.16 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.

[Unreleased]

[1.3.0] - 2026-05-19

Added

  • Entry-point classification (delivery phase, section 3.1) — before writing a smoke test, the AI must classify the touched surface (GET-safe action / state-changing HTTP / webhook / CLI / cron / queue / event listener) and pick the matching entry-point shape. The curl-GET template only applies to read-only routes; everything else has a one-line rule for triggering through its real mechanism. Ties back to the testing phase's integration-vs-unit hard rule.
  • docs/prompt-engineering.md — full catalogue of prompt-engineering patterns used across phases (structure, reasoning gates, output discipline, severity/confidence/consistency tiers, cross-phase contracts). Extracted from the README so the README stays scannable; linked from README, Phases tab of the GitHub Pages site, and CONTRIBUTING.
  • docs/example-task/ — synthetic end-to-end walkthrough of one feature (csv-export-reports-listing) with every artifact from requirements through delivery, including a consensus synthesis, an architecture decomposition, a verification finding with file:line evidence, and executed smoke tests with side-effect checks. Linked prominently from the README and the docs site.
  • Laravel 11.x+ framework guide (prompts/frameworks/laravel.md) documented in README, CLAUDE.md, and the GitHub Pages Frameworks tab. Covers Eloquent, Form Requests, Policies, Events/Queues, Pest/PHPUnit, and artisan commands, with <framework_tool_affinity> rules and side-effect-asserting smoke-test templates.
  • VALIDATION.md cross-references added from CONTRIBUTING.md and PIPELINE_ORCHESTRATOR.md (the validation harness now has discoverable entry points from the two docs contributors actually start in).
  • .gitignore rule for .todo/ parking-lot directory (personal deferred-work staging, not shipped).

Changed

  • README opener rewritten to lead with the differentiator framing — "without the 'agent confidently broke prod' failure mode" — and the four pillars (hard-constraint phase gates, file:line-quoted verification, multi-model consensus, executed smoke tests). The Project Status disclaimer moved below the fold so first-time readers see the pitch, not the maintenance-mode caveat.
  • Pipeline phase count unified to 6 across all surfaces — README, CLAUDE.md, plugin manifests, marketplace listing, GitHub repo description, and docs/index.html. Consensus is now counted as phase 1 of 6 (previously framed as a separate pre-step); Requirements is explicitly "input you write" rather than a numbered phase. Previously the README said "five specialized AI agents" while plugin manifests said "6-phase" and the repo description said "5-phase" — three surfaces, three numbers.
  • Framework Support and Multi-Model Consensus copy reframed — removed hedging ("still works"/"enhancement") in favor of a contribution CTA and "fault-tolerant by design" language. The intent is the same; the tone now matches the actual posture.
  • GitHub Pages site (docs/index.html) brought current with the May 18 README rewrite — last edit before this batch was Apr 16. Hero now leads with the "broke prod" framing and links to docs/example-task/; phase cards renumbered so Consensus is 01 and Requirements is "Input" (file-text icon, not a numbered phase); a new "Install as a Claude Code Plugin (recommended)" section was added with /mots:* commands (the plugin install path had been entirely absent from the public site); AUTO_APPROVE default corrected from "none" to "all" and approval-gate copy reframed as optional throughout Overview / Workflow / Philosophy.

Removed

  • Codex plugin manifests (.codex-plugin/ and .agents/). They were added speculatively in v1.1.0, never linked from the README, never tested in CI, and cost a dual-version-bump at every release. If Codex compatibility comes back as a real demand, the manifests are recoverable from git history.
  • 35-bullet prompt-engineering enumeration in the README — moved to docs/prompt-engineering.md and replaced with a 3-sentence summary plus link, so the README stops scrolling past the techniques the reader is actually trying to find.

[1.2.1] - 2026-05-15

Changed

  • Consensus default Claude model bumped from Sonnet 4 to Sonnet 4.6 (claude-sonnet-4-20250514claude-sonnet-4-6). Sonnet 4 is being retired on the Anthropic API on 2026-06-15; this avoids service interruption for CONSENSUS_CLAUDE_MODE=api and the standalone bin/consensus.php script. session mode is unaffected — it routes Claude work through Claude Code Task subagents, which the harness resolves to current models.
  • motspilot.sh task-name validation now allows _ and . in the interior of task names (e.g. 01e_owasp-8.2.3-input-validation). Start/end must still be [a-z0-9].

Added

  • CONSENSUS_CLAUDE_MODEL environment variable in bin/consensus.php — overrides the default Claude model used for both the fan-out call and the synthesis/differences judge calls. Lets you upgrade past future deprecations without a code edit.

[1.2.0] - 2026-04-17

Added

  • Per-phase model routing — phase prompts now declare a model: YAML frontmatter field. The orchestrator reads it with yq and passes it to the Task tool so each phase runs on the right model. Defaults: Architecture → opus (design trade-offs, blast-radius reasoning); Development / Testing / Verification / Delivery → sonnet (routine code generation + mechanical checks). Stretches Opus session quota without hurting output quality where it matters.
  • Session-mode consensus — new CONSENSUS_CLAUDE_MODE config key with two modes:
    • session (default inside Claude Code) — bin/consensus.php --external-only fans out to GPT-4o + Gemini only; the orchestrator spawns three Sonnet Task subagents for Claude's consensus roles (perspective → 01_claude.md, synthesis → 04_synthesis.md, differences → 05_differences.md). Claude work draws from session quota; ANTHROPIC_API_KEY is no longer required for consensus.
    • api (legacy) — bin/consensus.php handles everything via direct Anthropic API calls. Requires ANTHROPIC_API_KEY.
  • bin/consensus.php gains a --external-only flag that skips the Claude fan-out and both judge calls, drops ANTHROPIC_API_KEY from the required-key set, and reports === EXTERNAL-ONLY CONSENSUS COMPLETE ===.

Changed

  • /mots:init adapts required-key checks to CONSENSUS_CLAUDE_MODE: only OPENAI_API_KEY + GEMINI_API_KEY are required in session mode. Claude Code's ANTHROPIC_API_KEY stripping is now reported as "ready (mode: session)" instead of a problem.
  • Default .motspilot/config written by motspilot.sh now includes CONSENSUS_CLAUDE_MODE="session".
  • plugin.json userConfig description for ANTHROPIC_API_KEY clarifies it's only needed in api mode.

[1.1.0] - 2026-04-16

Added

  • Claude Code plugin support — install via /plugin marketplace add motsmanish/motspilot + /plugin install mots
  • 7 skills: /mots:pilot, /mots:pipeline (alias), /mots:init, /mots:status, /mots:archive, /mots:reactivate, /mots:view
  • Codex cross-compatibility (.codex-plugin/ + .agents/plugins/marketplace.json)
  • CONSENSUS config variable — graceful degradation when PHP/API keys unavailable
  • CONFIG_VERSION field for config migration tracking
  • Secret precedence in consensus.php: userConfig (keychain) > env vars > .env file

Changed

  • PIPELINE_ORCHESTRATOR.md paths are now plugin-aware (no motspilot/ prefix)
  • Consensus phase auto-skips when CONSENSUS=disabled
  • README installation options re-lettered (plugin is now Option A)

Added (previous)

  • WORKSPACE_DIR config option — store task artifacts in the project repo instead of inside motspilot's own directory. Enables version-controlling task data alongside project code.
  • Installation guide in README — symlink, submodule, and direct clone options with step-by-step instructions.

Fixed

  • Case-insensitive meta key matching in list_tasks() for compatibility with tasks created by older versions.

[1.0.2] - 2026-02-24

Changed

  • Default AUTO_APPROVE to all — pipeline phases run without pausing by default

[1.0.1] - 2026-02-23

Fixed

  • Use set -e safe arithmetic for counter increments in shell script

[1.0.0] - 2026-02-23

Added

  • Initial release of motspilot
  • 5-phase AI pipeline: Architecture, Development, Testing, Verification, Delivery
  • Shell script (motspilot.sh) for task management, archiving, and state tracking
  • Thinking framework prompts for each pipeline phase
  • CakePHP 4.x framework guide (prompts/frameworks/cakephp.md)
  • Auto-archive on pipeline completion
  • Task lifecycle: create, prepare, run, archive, reactivate
  • Checkpoint system for phase state tracking
  • Configurable approval gates between phases (AUTO_APPROVE setting)
  • Claude Code as native orchestrator via PIPELINE_ORCHESTRATOR.md
  • Multi-task support with concurrent task management
  • Color-coded logging to console and file