Skip to content

Releases

Anansi 0.2.0

Choose a tag to compare

@herson herson released this 18 Nov 21:00
34ab2f7

Changed (breaking)

  • Python 3.11 or later is required (ADR-0011). Python 3.10 is end of life, and the fixed releases of pillow and dnspython no longer support it, so 3.10 installs got versions with known vulnerabilities. The Docker image is unaffected.
  • Scan results are a list of findings (ADR-0003). reports/report.json, stored scan history and GET /api/scan/{id} now return a list of objects with host, port, protocol, service, product, version and cves, instead of an object keyed by port. Multi-host scans no longer overwrite each other. The CSV report gains Host, Protocol and Product columns. History saved before this change keeps its old shape.

Added

  • AI analysis with any OpenAI-compatible endpoint (ADR-0005, ADR-0006): new ai_base_url, ai_api_key_env and ai_token_parameter settings send --ai to OpenRouter or a local model such as Ollama, so scan data can stay on your machine. OpenAI remains the default, and its key is never sent to another endpoint. Invalid AI settings stop --ai before the scan starts.
  • Community Health: Added CODE_OF_CONDUCT.md, SECURITY.md, and FUNDING.yml.
  • Automation: Added GitHub Action for CI (.github/workflows/ci.yml) to run tests and linting on every push.
  • Templates: Added Issue templates for Bug Reports and Feature Requests, and a Pull Request template.
  • Documentation: Enhanced CONTRIBUTING.md with detailed setup and testing instructions.

Changed

  • Dependencies are locked with hashes (ADR-0010): requirements.txt is now generated from requirements.in by scripts/lock-requirements.sh and pins every package with hashes for every supported Python version. CI tools are locked in requirements-dev.txt, and mkdocs moved out of the runtime image to requirements-docs.txt. Edit the .in files, then re-lock.
  • CI fails on lint findings: flake8 now blocks the build, with its settings in .flake8 so local runs match CI. Actions are pinned to commit SHAs, and the workflow token can only read the repository.
  • CI tests Python 3.11 through 3.14, every version from the minimum to the newest release.
  • CI runs weekly on main as well as on every push and PR, so pip-audit flags a new advisory against a locked dependency without waiting for the next change.
  • The Docker image runs Python 3.14, the newest version CI tests, instead of 3.12.
  • Refactored CONTRIBUTING.md to .github/CONTRIBUTING.md for better GitHub integration.
  • Updated README.md with build status badges and sponsorship information.

Fixed

  • Readable JSON on the console: --compliance, --dns and --s3 printed \u2013 and \u00a7 instead of – and §.
  • The mkdocs pages describe the real code: they documented functions and flags that never existed (scan_network(), --enumerate). Pages that repeated the README are gone, the module pages show the actual classes and Finding records, and a test checks that every flag, import and link in docs/ exists.
  • The dashboard starts from any working directory: it found its templates relative to the directory it was started from, and importing web/app.py opened anansi.db there. The database now opens when the app starts, and the scheduler starts and stops with it through a lifespan handler instead of the deprecated on_event hooks.
  • PDF reports render any text: fpdf 1.7.2 is replaced by fpdf2 with an embedded DejaVu Sans font, so CVE descriptions, service banners and compliance controls outside Latin-1 no longer crash --pdf.
  • --pdf --compliance includes the compliance mapping: it was added after the file was saved, so it never appeared.
  • --pdf works on CIDR targets: the report for 10.0.0.0/24 is now report_10.0.0.0_24.pdf instead of a crash on the /.
  • --s3 no longer reports a bucket as clean when a check fails: failed checks are listed under errors and the command exits with status 1. It now checks all four public access block settings, grants to any AWS account, and whether the bucket policy is public, which needs the s3:GetBucketPolicyStatus permission.
  • Scheduled and manual scans no longer run at the same time: a scheduled scan that fires while another scan runs is skipped with a warning in the log, instead of overwriting the running scan's progress and results and letting a third scan start.
  • Scheduled scans survive restarts (ADR-0004): schedules are saved in the scan database and restored when the dashboard starts. Before, every restart or image update silently dropped them.
  • --ai works again with the default config: OpenAI shut down gpt-3.5-turbo on 2026-10-23. The default is now gpt-6-luna, and requests cap output with max_completion_tokens, which current reasoning models require instead of max_tokens. A custom ai_model in config.yaml must accept that parameter.
  • --ai analyses what the scan verified: the model now receives each service's product name and, with --cve, the CVEs found in the NVD. Before, it saw only http 2.4.49 and had to recall vulnerabilities from memory, so it could miss them or cite wrong ones.
  • An unknown report_format is an error: a value other than json or csv in config.yaml used to skip the report without a word. The CLI now says so and exits before the scan starts.