Release v0.5.9 - #12
Merged
Merged
Conversation
Watch a path for file changes and re-scan on each change. Prints findings inline without exiting; logs secrets to audit.jsonl in real time. Ctrl+C exits watch mode. - Node: uses chokidar (new dep) - Python: uses watchdog (new dep, optional with helpful error) - CLI_SPEC.md updated with --watch documentation Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
…isting configs Old default pattern treated the pipe as regex alternation, matching any command containing "sh" as a substring—flagging git push, grep commands with shell patterns, .sh filenames, etc. DEFAULT_REQUIRE_APPROVAL was already updated to use word-bounded shell names, but existing ~/.rafter/config.json files written by older installs kept the bad pattern. Add migration logic to ConfigManager.migrate() (Node) and ConfigManager._migrate() (Python) to silently upgrade the old pattern on first load. Also update rafter-agent-security SKILL.md: clarify that command blocking is handled automatically by the PreToolUse hook, and that /rafter-bash is only needed in environments without the hook installed. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
F005: gitleaks v8.x outputs "v8.18.2" (no prefix), not "gitleaks version
v8.18.2". The old check `stdout.includes("gitleaks version")` and
`"gitleaks version" in result.stdout` always returned False, causing
`rafter agent verify` to report a working binary as failed. Fix: accept
any successful exit (code 0) with non-empty stdout. Also surface stdout
in the error detail string so the actual output is visible when things do
go wrong.
F006: `rafter agent init` printed "Restart OpenClaw to load skill" even
when the skill install had just failed. Track install success separately
(openclawOk / claudeCodeOk) and gate the Next Steps suggestions on actual
success. Both Python and Node fixed.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Bump version 0.4.0 → 0.5.6 in all three skill files - rafter/SKILL.md: sharpen description to SAST/code analysis; remove CVE/dependency language that implied SCA (Rafter does SAST, not SCA) - rafter-security-skill.md (OpenClaw): update last_updated, add hook-first caveat to /rafter-bash, fix pipe-to-shell display text formatting - rafter-agent-security/SKILL.md: version bump only (language already updated in previous session) Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
…ands (v0.5.7) - `rafter scan` (no subcommand) → remote backend scan (same as `rafter run`) - `rafter scan remote` → explicit alias for remote backend scan - `rafter scan local [path]` → local secret scanner (formerly `rafter agent scan`) - `rafter agent scan` → still works, prints deprecation warning to stderr - All hook scripts, CI templates, and skill files updated to `rafter scan local` - Both Node (TypeScript) and Python implementations in parity - All tests updated and passing (142 Node, 252 Python) - Version bumped to 0.5.7 Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
binary_manager.py: gitleaks v8.x outputs just the version string (e.g. "v8.18.2"), not "gitleaks version v8.18.2". Check exit code 0 + non-empty output instead of string matching. README: add concrete quickstart showing scan → protect → audit loop. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
audit --share: generates a redacted diagnostic excerpt safe for pasting into GitHub issues—CLI version, OS/arch, policy fingerprint (via loadWithPolicy for correct .rafter.yml merge), and last 5 events with truncated commands. Node + Python with full test coverage. pretool hooks: replace flat error strings with structured multi-line messages showing matched rule, risk level with description, and actionable next steps (approve command, adjust config). Node + Python parity maintained. Closes rc-pv014, rc-hjmmw Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Updates assertions to match new multi-line message format. No new test secrets added—existing AKIA/sk_live fixtures unchanged. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…es (rc-0as) Generic API Key and Generic Secret patterns now: - Require a negative lookbehind to avoid matching when the keyword (api_key, password, etc.) is part of a longer identifier (e.g. anthropic_api_key, user_password) - Require the value to be quoted, filtering out variable references and function calls that aren't actual hardcoded secrets Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Add Codex CLI detection (~/.codex) to agent init (Node + Python) - Add --skip-codex flag to agent init (Node + Python) - Add Codex skill installation to ~/.agents/skills/ (Node + Python) - Add codex environment to config schema and defaults (Node + Python) - Add Codex CLI status to agent status command (Node + Python) - Update Codex policy export with improved usage instructions - Bump all skill versions to 0.5.8 (rafter, rafter-agent-security, OpenClaw) - Update shell completions (bash, zsh, fish) with --skip-codex
Two-part GitHub Issues integration: - `rafter issues create from-scan`: Create issues from backend scan results (--scan-id) or local scan JSON (--from-local). Includes dedup by file+pattern fingerprint hash, severity labels, and batch creation. - `rafter issues create from-text`: Create issues from natural language input via stdin pipe, --file, or --text. Extracts title, severity, security labels, and file references from text. Both node/ and python/ implementations with identical CLI surface. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…uos) Generic API Key and Generic Secret patterns now require the matched value to contain at least one digit (API Key) or digit/special char (Secret). This prevents false positives when the value is an env var name like ANTHROPIC_API_KEY or DATABASE_PASSWORD rather than an actual secret. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…ons (rc-9ag)
Extend agent support beyond Claude Code and OpenClaw. New MCP-based
integrations detect each AI engine's config directory and install the
rafter MCP server entry into their respective config files:
- Gemini CLI: ~/.gemini/settings.json
- Cursor: ~/.cursor/mcp.json
- Windsurf: ~/.codeium/windsurf/mcp_config.json
- Continue.dev: ~/.continue/config.json
- Aider: ~/.aider.conf.yml
Updates init command with --skip-{gemini,aider,cursor,windsurf,continue}
flags and status command to show all integrations.
# Conflicts: # node/src/commands/agent/init.ts # node/src/commands/agent/status.ts # node/src/core/config-defaults.ts # node/src/core/config-schema.ts
Add DOCS_SYNC_CHECKLIST.md to shared-docs/ with: - Source of truth file mapping - Per-change-type checklist of which docs files to update - Common pitfalls (deprecated commands, flag names, file paths) - Sync process steps Companion to Rome-1/docs PR #2 which syncs docs with CLI v0.5.8.
…sights (rc-q8c) Analyze Anthropic's claude-code-security-review GitHub Action to extract security patterns, false positive filtering strategies, and insights applicable to rafter-cli's agent system. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Analyzes the 17 filtering precedents from claude-code-security-review and proposes injecting 10 as CLAUDE.md guidance during `rafter agent init`. Includes draft content, injection mechanism, and implementation plan. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Review of CLI help text, agent init output, skills, command structure, and documentation. Proposes 8 improvements including project-level CLAUDE.md generation, post-init summary, command hierarchy cleanup, contextual scan prompts, policy surfacing, expanded skill triggers, agent status enhancement, and rafter doctor health check. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…ized rafter-cli (rc-s1b) Prioritized list of 15 strategies across 4 tiers with feasibility assessments. Covers proactive guidance (CLAUDE.md guardrails), active enforcement (dependency gates, SAST-lite), reactive detection (PostToolUse secret scanning), and semantic analysis (LLM review gates). Builds on rc-q8c research and rc-1qt guardrails proposal. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…peline (rc-6mb) 5-stage security pipeline: static analysis (semgrep + language linters), dependency scanning (npm/pip/cargo audit), OWASP Top 10 validation with framework-aware filtering, optional AI security review via Claude, and unified report/gate. Includes secret detection improvements, security test generation, sandboxing recommendations, and 8-phase implementation roadmap. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…(rc-crg) Per-platform integration spec covering Claude Code, Cursor, Codex CLI, Gemini CLI, Windsurf, Continue.dev, and Aider. Proposes project-level CLAUDE.md generation, rafter agent init-project command, capability manifest, agent session detection, and MCP server registration. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Cherry-picked from orphan polecat branch jasper/rc-4pxj1. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Cherry-picked from orphan polecat branch jasper/rc-5s1fw. Resolved pre-commit-hooks.yaml conflict by combining system + node + python entries. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Cherry-picked from orphan polecat branch opal/rc-9pt2y. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Cherry-picked from orphan polecat branch quartz/rc-m02tg. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…ons (rc-13c) - Add MCP install functions to Python agent.py for all 5 engines - Update Python init command with --with-gemini, --with-cursor, --with-windsurf, --with-continue, --with-aider options - Update Python status command to show MCP-native engine status - Add recipe docs for each new integration - Add 11 new tests for MCP install functions (all pass) - Node implementation was already complete Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
… (rc-1qd) Add post-match filtering in PatternEngine to reject matches where the quoted value looks like a variable/env var name rather than an actual secret. Filters values matching ALL_CAPS_UNDERSCORE (e.g. OPENAI_API_KEY_V2) and lowercase_identifier (e.g. my_app_secret_key_1) patterns. Applied consistently across scan, scanWithPosition, redactText, and hasMatches. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…ejd) Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Fix Python _install_codex_skills() broken source paths: was referencing non-existent python/.claude/skills/ directory. Now uses importlib.resources with bundled SKILL.md files in resources/skills/. - Fix OpenClaw skill install failing when ~/.openclaw exists but ~/.openclaw/skills/ doesn't: installRafterSkillVerbose() now checks the parent dir and creates skills/ subdir as needed (Node + Python). - Add Codex CLI check to verify command in both Node and Python, matching the status command's coverage. - Add tests for _check_codex and update verify command test mocks. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…q-xn95g) The catch block was deleting tmpReport before checking e.code===1, making the leaks-found recovery path unreachable. Moved cleanup after the exit code check so the report file is still available when leaks are detected. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
git diff/staged outputs paths relative to the repo root, but path.resolve(file) resolves relative to cwd. When running from a subdirectory, files are missed. Use git rev-parse --show-toplevel to get the repo root and resolve paths against it. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
… consistency (hq-nuctq) --staged/--diff with no files to scan now respects --format json/sarif instead of always outputting plain text. Both early return paths in scanDiffFiles and scanStagedFiles now call outputScanResults([], opts) which handles all output formats correctly. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…t content (rc-1qd) Node binary-manager.ts: verifyGitleaks() and verifyGitleaksVerbose() now accept any successful exit (code 0) instead of requiring non-empty stdout, since execAsync already rejects on non-zero exit. Python gitleaks.py: removed literal "gitleaks version" string check from stdout, relying solely on returncode == 0. Python binary_manager.py: removed requirement for non-empty stdout alongside returncode == 0, accepting exit code alone as sufficient. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…duce FPs (hq-oxdz2) - Add vendor, .venv, venv, __pycache__, .tox, .mypy_cache, .pytest_cache, results, .terraform, bower_components to default scan excludes - Generic Secret: raise min length 8→12, require both letters and digits - Bearer Token: raise min length to 20, require both letters and digits Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…hq-7vvvi) Validate both flags at the top of the action handler before any scan work begins. Invalid --engine values (e.g. --engine nope) now error with exit code 2 instead of silently falling through to auto mode. Invalid --format values (e.g. --format yaml) now error immediately instead of after the scan completes. Defense-in-depth validation also added to selectEngine() in both scan.ts and baseline.ts. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Raftersecurity
self-requested a review
March 8, 2026 03:08
Raftersecurity
approved these changes
Mar 8, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
v0.5.9 — Bug fixes from Codex end-to-end evaluation
--staged/--diffno-file paths now respect--format json/sarif--engine/--formatvalues now error before scan starts.venv,vendor,results, etc. + tightened Generic Secret (12+ chars, mixed) and Bearer Token (20+ chars) regexesv0.5.8
rafter agent audit --share— redacted diagnostic excerpt for issue reports (Node + Python)verify_gitleaks_verbosenow checks exit code, not string matchAlso includes
--with-*opt-in replaces--skip-*flagsrafter scan local,rafter agent scandeprecation)Test plan
rafter agent scan --engine gitleaks --format json <file-with-secret>returns findings (was returning[])rafter agent scan --staged --format jsonwith no staged files returns valid JSON (was plain text)rafter agent scan --stagedfrom subdirectory finds secrets in staged filesrafter agent verifypasses with valid gitleaks installrafter agent scan --engine nopereturns validation errorrafter agent scan --format yamlreturns validation error before scan starts.venv/vendordirs produces minimal false positives🤖 Generated with Claude Code