Changelog

All notable changes to Firmis are documented here. Format follows Keep a Changelog.

[2.1.0] - 2026-03-31

Added

• firmis accept - risk acceptance with audit trail, expiry dates, version pinning, and AI-suggested dispositions
• Finding states - every finding now has a state (detected, confirmed, dismissed, accepted) that drives security grades
• Security grades - A-F grading computed after deep scan, with grade comparison showing impact of accepted risks
• MCP server mode (firmis mcp) - run Firmis as an MCP server for agent integration via stdio transport
• Interactive fix flow - per-finding approval with Y/n/s/q/a prompts
• AI disposition - deep scan now suggests accept/fix/investigate for each finding with confidence scores
• Monitor .env protection - new rule blocks .env file writes
• JSON v2.0 schema - new schema_version, scan_mode, state, attack_technique, reasoning fields
• ISO 42001 and MITRE ATLAS compliance frameworks (now 7 total)

Changed

• Terminology standardized: “pattern”, “threat”, and “hotspot” are now “finding” everywhere
• firmis scan now scans current directory by default. Use --global for full project + home scan
• Auth flow: server-side OAuth via Cloudflare Worker. No localhost or Supabase URLs visible
• All pricing amounts and tier names removed from CLI output and reports
• Pentest moved from Pro to Business-only
• license command removed (replaced by account-based tier check via firmis login)
• HTML report: grade badge, filter tabs, visual threat map, expandable findings, risk register section

Fixed

• Confidence display bug (was showing 9000%)
• SARIF output now includes hardening findings as note level
• CI exit code returns 0 when --fail-on not set
• Badge command hides raw HTTP status codes

[2.0.0] - 2026-03-27

Breaking Changes

• Package renamed from firmis-scanner to firmis-cli
• JSON output schema updated to v2.0 (new fields, removed deprecated interfaces)
• buildBucketedOutput alias removed
• ConfirmedThreatEntry and ExpectedBehaviorEntry types removed

Added

• Deep scan - AI-powered finding verification with server-side analysis (5 credits per component)
• Workspace - complete dashboard rewrite with projects, scan history, billing, and onboarding
• LemonSqueezy billing - account-based subscriptions and credit packs
• Cloud sync - scan results, fix logs, and monitor events sync to workspace

Migration

• Update imports: firmis-scanner to firmis-cli
• Update npx commands: npx firmis-scanner to npx firmis-cli
• JSON consumers: handle new state field on findings, schema_version: "2.0"

[1.7.0] - 2026-03-11

Added

• firmis init - one-command project setup: detects AI tools, runs first scan, generates .firmisrc.json, shows next steps with contextual upgrade path
• GitHub Action (firmislabs/firmis-scanner@v1) - composite action with PR grade badge comments, HTML report artifacts, and optional workspace sync
• 8 new detection rules across agent-memory-poisoning, credential-harvesting, insecure-config, known-malicious, network-abuse, prompt-injection, supply-chain, and tool-poisoning categories
• Total rules: 269 across 26 categories
• Behavioral scoring wired into runtime monitor decision engine
• Training data pipeline: auto-export labeled sessions, synthetic data generation, weight calibration via grid search

Fixed

• Behavioral scoring was not triggering decisions (score scale mismatch: 0-1 vs 0-100)
• Slowdrip attack template missing exfiltration phase (100% false negative rate)

[1.6.1] - 2026-03-10

Added

• Cloud workspace Phase 1 MVP (Cloudflare Pages deployment)
• firmis login, firmis logout, firmis whoami commands
• --sync flag for scan result upload to firmislabs.com
• firmis badge for README security grade badge
• Post-scan CTA for lead generation
• Scan history with local storage

[1.3.0] - 2026-03-01

Added

• 34 new detection rules: access-control (3 rules), insecure-config (3 rules), expanded credential-harvesting, prompt-injection, supply-chain, and suspicious-behavior categories
• Nanobot platform analyzer
• Total rules: 209 across 26 categories

Fixed

• False positive reduction in secret detection for test fixtures
• Discovery timeout on large monorepos (500-file limit per component)
• Cross-platform finding deduplication

Security

• Input validation on component names (path traversal prevention)
• MAX_FILES_PER_COMPONENT=500 limit (DoS prevention)

[1.2.0] - 2026-02-18

Added

• Credential harvesting and prompt injection rule hardening (Sprint B)
• Supply chain detection improvements
• Cross-platform dedup engine (src/scanner/dedup.ts)

Fixed

• YAML escaping issues ('\' → '' for literal quotes)
• PlatformRegistry singleton state persistence across tests
• Broad regex false positives in secret detection

Security

• Component name validation against path traversal and XSS

[1.1.0] - 2026-02-16

Added

• 8 platform analyzers: Claude, MCP, Codex, Cursor, CrewAI, AutoGPT, OpenClaw, Nanobot
• 175+ YAML detection rules across 12 threat categories
• YARA-like pattern matching engine
• Secret detection (60 rules)
• OSV vulnerability scanning
• Discovery + Agent BOM (CycloneDX 1.7)
• CI pipeline command (firmis ci)
• SARIF 2.1.0 and HTML report output

[1.0.0] - 2026-02-12

Added

• Initial release
• Core scan engine with regex pattern matching
• Terminal, JSON output formats
• scan, list, validate commands