Skip to content

Roadmap

AION is intentionally small and focused: scan a Python repository, generate deterministic patches for a set of high-confidence security issues, verify them, and open pull requests only for fixes that pass.

Current status

The project ships:

  • context-aware Python scanning with repository profiling and Semgrep triage
  • deterministic repair for seven security issue types, each with an AST-level verification assertion
  • a verification gate (syntax + assertion + optional Semgrep re-scan) that gates every pull request
  • auto-update to scan, verify, and open pull requests for verified fixes
  • a reusable GitHub Action
  • drift detection (snapshot / drift), a continuous watch loop, and a status view over snapshots and repair history

Design principles

  • Trustworthy over broad. A small set of fixes that are always verified beats a large set that sometimes corrupts code.
  • Report, don't guess. When a fix cannot be proven safe (for example, a missing auth decorator), AION reports it for review instead of editing the code.
  • Deterministic and auditable. Patches are template-based and every one is re-checked by an independent AST assertion.

Possible next steps

These are directions, not commitments. Each should be weighed against the "trustworthy over broad" principle before being taken on.

  • Reliability transparency — surface deterministic repair and verification pass rates per issue type, and ship golden-path examples that run end-to-end in CI.
  • A few more high-confidence fixes — additional issue types only where a deterministic patch can be verified with the same rigor as the existing seven.
  • Better repository test selection for verification, beyond a single file's AST assertion.
  • Quality signal — bring back a lightweight repair-quality report (success / verification-pass / false-fix rates) over a fixture corpus.

What is explicitly out of scope

To keep the project controllable, these are non-goals:

  • runtime control-plane features (event queues, webhooks, inbox processing)
  • staged-rollout / release-candidate management
  • runtime containment (WAF, gateway, feature-flag, deploy integrations)
  • in-place hot patching of live production code
  • non-Python languages