Skip to content

Taxonomy gaps identified by cfgaudit crosswalk (credit: predictor2718) #68

Description

@chaksaray

Summary

Eight behavioral classes with no current AVE record, identified during PR #67, "Add cfgaudit→ AVE crosswalk (static config-auditor)". Recorded here so the list survives independent of the PR itself, which will be compressed into a single merge commit once it lands.

Credit

Identified by predictor2718 (GitHub) / Predictor_2718 (Reddit), maintainer of cfgaudit, a static security auditor for AI-agent configuration files (Claude Code, Cursor, Copilot, Gemini, Codex, Devin, Zed, Continue). Not solicited directly; built independently in response to the AVE launch posts and submitted as a complete, self-validated crosswalk.

The gaps, as stated in PR #67

cfgaudit has 90 rules; 33 map onto 19 existing AVE behavioral classes.
The remaining 57 check config-file classes AVE's skill and MCP-server records do not currently enumerate. Reduced to distinct categories, not already covered by an existing record:

  • Permission/approval config
  • Zero-click IDE auto-run
  • Telemetry redirect
  • Sandbox-config weakening
  • TLS-verification-disabled
  • Container posture
  • MCP network posture
  • Supply-chain pinning

Why this is worth taking seriously as a gap list, not just a rule count

This is config-surface coverage, a genuinely different slicing axis than AVE's existing behavioral classes ("cfgaudit slices threats by config surface where AVE slices by behavior," per the PR's own framing). Several of these categories describe real, config-file-specific attack surface (zero-click IDE auto-run, sandbox-config weakening) that doesn't map cleanly onto any of AVE's current skill or MCP-server records, not because the underlying risk is unimportant, but because AVE hasn't had a config-file-focused implementation stress-test its taxonomy against this surface until now.

Independent validation, also from PR #67, relevant context for this issue

The same PR ran cfgaudit and Bawbel Scanner 1.3.0, two independent codebases sharing no code and no ruleset, against the same test files.
Of 10 overlapping instruction-content rules, both scanners independently emitted the same ave_id on 5. The other 5 divergences were confirmed as detection-pattern differences (each tool's own trigger regex not matching the same test string), not taxonomy or mapping errors. This is the strongest cross-implementation evidence this project has produced to date that the AVE ID scheme is genuinely interoperable, not just internally consistent.

Status

Not scheduled, no version targeted. This issue exists to make the gap list durable and properly credited, not to commit to a timeline. When any of these classes are designed as new records, credit this issue and PR #67 in the record's own commit message or references, the same discipline already applied to the reversibility field's own tracking issue.

Metadata

Metadata

Assignees

No one assigned

    Labels

    enhancementNew feature or request

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions