Changelog

Changelog

Every release, every change. Full transparency.

v5.3.3 Fix March 6, 2026
Dependency Updates
  • zod — upgraded from 3.25.76 to 4.3.6
  • @types/node — upgraded from 22.19.11 to 25.3.5 (dev)
  • CI actions — setup-node v4 → v6, codeql-action v3 → v4, github-script v7 → v8
v5.3.2 Fix March 6, 2026
Deep Audit & Enterprise Console Fix
  • Enterprise Console audit — Fixed nav structure to match standard pattern, corrected audit file path (audit.jsonl → audit.log), added missing save_custom_rules event type to audit filter, standardized footer
  • CI command fix — Corrected wrong npx pcp-engine pcp check commands in docs (should be pcp check after global install)
  • Provider count fix — features.html SVG corrected from 5 to 4 providers
  • Pricing date fix — models.html updated to March 2026 (was February)
  • Video v2 — Scene transitions, peppy background music, honest cost comparison, removed test count display
v5.3.1 Fix March 6, 2026
Accuracy & Docs Audit
  • Honest cost comparison — Fixed savings display that clamped negative values to 0%. When routing to a pricier model (e.g., Sonnet vs GPT-4o baseline), the CLI and MCP now honestly show “X% more expensive” instead of “Comparable cost.”
  • Model count corrected — 11 → 10 across all pages (matches actual PRICING_DATA: 3 Anthropic + 3 OpenAI + 2 Google + 2 Perplexity)
  • Free tier updated — why.html and terms.html now correctly reflect 50 optimizations/month (was showing stale “10 lifetime” from pre-v5.1.0)
  • Nav standardized — All 15 pages now share the same canonical nav: Features | Models | Docs | Pricing | Changelog. Removed stale Home and Enterprise Console links.
  • Docs expanded — Added Validate API, Benchmark, and PR Comment Bot sections to docs.html. Added 4 missing CLI subcommands to reference table.
  • Pricing version updated to 2026-03
v5.3.0 New March 6, 2026
CI-First Release
  • GitHub Action PR Comment Bot — comment: 'true' posts a formatted markdown table to every PR with per-file scores, pass/fail status, and top issues. Idempotent updates via <!-- pcp-bot --> marker.
  • Lightweight Validate API — import { validate } from 'pcp-engine/validate' for any Node.js backend. Zero MCP dependency.
  • Benchmark Dataset + CLI — 15 curated prompts with locked expected scores (±3 tolerance). pcp benchmark for regression testing.
  • 4 new CLI subcommands — benchmark, demo, report, badge (11 → 15 total)
  • Product explainer video — 5-minute enterprise-ready walkthrough embedded on landing page
  • 832 tests across 38 test files
v5.2.1 Fix March 4, 2026
GitHub Format & Docs Fixes
  • GitHub format fix — --format github: confidence now reads from lint result instead of hardcoded n/a
  • Severity fix — Passing prompts now emit ::notice instead of ::warning (correct severity level)
  • Annotation fix — Confidence included in error annotations too
  • README — Added pcp demo, pcp badge, pcp report to CLI examples; demo/badge/report now listed under “All subcommands”
  • README — Added --format github, --warn-only, --output CI flags documentation
  • README — Architecture diagram upgraded from ASCII art to Mermaid flowchart
v5.2.0 New March 4, 2026
CLI Expansion + Modular Architecture
  • pcp demo — Guided first-run experience with example prompts
  • pcp badge — Generates shields.io PQS badge markdown for README
  • pcp report — Produces prompt-quality.json + prompt-quality.md artifacts
  • --format github — Outputs ::warning/::error annotations for PR inline display
  • --warn-only — Advisory CI mode (exit 0 even on failures)
  • --output <dir> — Custom output directory for report files
  • GitHub Action step summary — Markdown quality table in $GITHUB_STEP_SUMMARY
  • Confidence levels — Shown in check, score, and preflight human output
  • PQS branding — “Prompt Quality Score” consistent branding throughout CLI output
  • Modular tools architecture — Split 1957-line tools.ts into 6 focused files in src/tools/
  • MIGRATION.md — Step-by-step migration guide from old package names
  • PQS_SPEC.md — Formal scoring specification
  • Examples directory — 6 real-world prompt examples covering various quality levels
  • 8 golden snapshot tests for CLI output format stability
v5.1.0 Breaking March 4, 2026
Major Rebrand & Launch Release
  • Package renamed from claude-prompt-optimizer-mcp to pcp-engine
  • Free tier expanded from 10 lifetime to 50 optimizations/month
  • CI pipeline added with Node 18/20/22 matrix testing
  • ReDoS protection for custom regex rules
  • Fail-closed enforcement in enterprise enforce mode
  • Scoring honesty — quality score reflects input prompt only, compiled output gets structural checklist
  • Documentation accuracy — fixed dependency count (3, not 2), model count (10, not 11), version alignment
  • USD pricing added alongside INR
  • CLAUDE.md and CONTRIBUTING.md added
v5.0.2 Fix March 3, 2026
Domain Migration
  • Domain migration to getpcp.site
  • Install script updates
v5.0.1 Fix March 3, 2026
Hook Compatibility
  • Hook compatibility fixes
  • CLI hook install improvements
v5.0.0 New March 2, 2026
CLI Auto-Check Hooks
  • CLI auto-check hooks — pcp hook install for UserPromptSubmit integration
  • pcp doctor — installation health checker
  • pcp hook subcommands for managing prompt hooks
  • 787 tests passing
v4.1.0 New March 3, 2026
Enterprise Console Integration
  • Tool 20 — save_custom_rules: Deploy custom governance rules from the Enterprise Console to Prompt Control Plane. Validates, sorts deterministically, and returns a rule-set hash.
  • Enterprise Console → Product bridge: Build rules in the visual editor, click “Copy & Deploy”, paste into any MCP-connected AI assistant. Works with any LLM.
  • Consolidated Enterprise Features section in README with pricing table rows
  • Terminology updated: “Tamper-evident” → “Hash-chained” across all public docs
  • Content filter applied: internal paths, function names, and implementation details stripped from all public documentation
v4.0.3 Fix March 2, 2026
Contact & Privacy
  • Broadened contact page from Enterprise-only to general contact form with Subject dropdown
  • Replaced all exposed email addresses in privacy, terms, and security pages with contact form links
  • Added “Contact” link to footer Legal column across all pages
v4.0.2 Fix March 1, 2026
Cloudflare Pages Migration
  • Migrated hosting from GitHub Pages to Cloudflare Pages
  • Added security headers, custom 404, and redirect rules
  • Six new pages: changelog, privacy, terms, security, blog index, 404
  • Trust signals: Shields.io badges and three-column footer across all pages
v4.0.1 Security March 1, 2026
Security Hardening
  • Replaced MIT license with Elastic License 2.0 (ELv2) for stronger IP protection
  • Stripped source maps from the npm tarball, reducing package size from 256KB to 110KB
  • Scrubbed internal implementation details from public documentation
v4.0.0 Breaking March 1, 2026
Prompt Control Plane
  • Full rebrand from “Prompt Optimizer MCP” to “Prompt Control Plane”
  • Enterprise tier gates for policy enforcement, audit logging, and config lock features
  • New documentation and pricing comparison pages
  • License prefix change: po_pro_ → pcp_
  • Storage path migration: ~/.prompt-optimizer/ → ~/.prompt-control-plane/
v3.3.0 New March 1, 2026
Enterprise Operations
  • Policy enforcement mode with advisory and enforce configurations
  • Hash-chained audit logging with cryptographic hash chains
  • Config lock mode with passphrase protection for governance control
  • Two new tools: delete_session and purge_sessions for session lifecycle management
  • Comprehensive test coverage for enterprise workflows
v3.2.1 New March 1, 2026
Session History & Custom Rules
  • Reproducible session exports with auto-calculated integrity hashes
  • Custom rules support from local configuration directory
  • Rules versioning constant for drift detection across environments
v3.2.0 New March 1, 2026
Enterprise Unlock
  • Enterprise tier with 120 requests per minute and unlimited usage
  • Multi-page documentation site with dedicated pages for features, models, contact, and more
  • Shared design system with dark and light theme support
v3.1.0 New March 1, 2026
Smart Compression & Tool Pruning
  • Multi-stage context compression pipeline that strips irrelevant content and reports token savings
  • Zone scanner that protects structured content (code blocks, tables, lists) from modification
  • Tool pruning engine with relevance scoring to reduce context window usage
  • Four new ambiguity detection rules covering hallucination risk, underspecified agents, conflicting constraints, and model-scope mismatch
  • Pre-flight deltas showing estimated token savings before committing to optimization
v3.0.0 Breaking February 28, 2026
Decision Engine
  • Reasoning complexity classifier with six complexity types for intelligent model selection
  • Five optimization profiles: cost minimizer, balanced, quality first, creative, and enterprise safe
  • Deterministic model routing with full decision path audit trail
  • Dimensional risk scoring across four axes, scored 0–100
  • Perplexity provider support for research-oriented prompts
  • Three new tools: classify_task, route_model, and pre_flight
v2.3.0 New February 28, 2026
CLI Linter & GitHub Action
  • prompt-lint CLI binary for running prompt quality checks in CI/CD pipelines
  • GitHub Action for seamless workflow integration with configurable thresholds
  • Comprehensive CLI test suite
v2.2.0 New February 27, 2026
Programmatic API & Landing Page
  • Programmatic import { optimize } API for using the engine as a library
  • Curl-based installer for quick setup
  • GitHub Pages product site for documentation and marketing
v2.1.0 New February 27, 2026
Monetization
  • Three-tier pricing: Free, Pro, and Power with usage-based metering
  • Ed25519 offline license activation — no accounts, no network calls
  • Multi-LLM output targets: Claude XML, OpenAI system/user, and generic Markdown
  • Six new tools bringing the total to 11
v1.0.0 New February 26, 2026
Initial Release
  • Five core MCP tools for prompt analysis, scoring, compilation, refinement, and approval
  • Quality scoring on a 0–100 scale across five dimensions
  • Structured prompt compilation targeting Claude’s XML format
  • Session management for iterative prompt refinement
  • Zero LLM calls — fully deterministic, runs entirely offline

Want to see what’s next?

Follow the project on GitHub for release notes and updates.

View on GitHub →