feat: AVE-2026-00057, 00058, 00059 -- three new records#50
Merged
Conversation
…ception, fragmented injection Three new records plus their detection rules and fixtures: - AVE-2026-00057: obfuscated/encoded skill payload designed to evade static scanners (base64/hex/marshal decode fed directly into eval/exec) - AVE-2026-00058: deceptive skill trigger or activation-scope manipulation via misleading manifest description - AVE-2026-00059: fragmented cross-tool-description prompt injection reassembled at a planted trigger (ShareLock-class), cited to the original research plus Microsoft's 2026 MCP security checkpoint owasp_mcp corrected against the repo's own crosswalks/ave-to-owasp-mcp.md legend during review: 00057 was missing MCP03 (Tool Poisoning) alongside MCP04, and 00058's original MCP09 (Shadow MCP Servers) was a flat mismatch, replaced with MCP03+MCP06 (Tool Poisoning, Intent Flow Subversion). mitre_atlas AML.T0051 on 00059 verified against MITRE's own published technique name (LLM Prompt Injection), not just pattern-matched. All three validated against schema/ave-record-1.1.0.schema.json, AIVSS arithmetic independently recomputed, and each has a real regex-based detection rule whose positive/negative fixtures pass tests/test_fixtures.py.
2 tasks
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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
Each record ships with a real regex-based detection rule in
rules/pattern/and a positive/negative fixture pair intests/fixtures/, per this repo's own record-addition workflow.During review, two
owasp_mcpclassifications were corrected againstcrosswalks/ave-to-owasp-mcp.md(the repo's own MCP-code legend), not just pattern-validated:MCP03(Tool Poisoning) alongsideMCP04(Software Supply Chain Attacks) -- the closest existing analog, AVE-2026-00029 (also anObfuscationclass), maps to both.MCP09(Shadow MCP Servers) was a flat mismatch -- this record is about a skill's own deceptive manifest, not an unauthorized/unregistered server. Replaced withMCP03+MCP06(Tool Poisoning, Intent Flow Subversion).mitre_atlas: AML.T0051on 00059 was verified against MITRE ATLAS's own published technique name ("LLM Prompt Injection") via web search, not assumed from the existing corpus's (inconsistent) convention.Test plan
python3 scripts/validate_records.py-- all 59 records valid againstschema/ave-record-1.1.0.schema.jsonpython3 scripts/check_rule_coverage.py-- all 59 records have detection rulespython3 scripts/check_fixtures.py-- all 59 rules have positive and negative fixturespython3 -m pytest tests/ -q-- 118 passed, including the 6 new fixture tests (rule fires on positive, stays silent on negative)aivss_scoreseverityagrees withaivss.aivss_scorebanding for all threebawbel-scanner,bawbel scan\b,bawbel-gate,DESIGN.md-- zero matches