From 4260dde7cb93356fc0eb36b52a340d3bfd0ca6be Mon Sep 17 00:00:00 2001 From: chaksaray Date: Fri, 17 Jul 2026 22:46:12 +0700 Subject: [PATCH] fix(ave): remove bawbel CLI names and stale counts from LANGUAGE.md Found during the D3/D4 neutrality-audit pass, independent of the detection-layer/scanner discussion in Task D1: the detection_stage table named literal bawbel CLI commands ("bawbel scan", "bawbel scan-server-card") in the standard's own domain-language glossary. Same class of finding as the corpus-wide bawbel-scanner boilerplate fix, in a file the earlier sweep never checked since it isn't a record. Also recounted the detection_layer breakdown cited in prose: "33 of the 48" / "12 of the 48" was stale (corpus was 48 records when written, is 59 now). Live counts: content 37/59, runtime 15/59. Confirmed clean on the specific concern this file was flagged for: it is not a bawbel-gate style-guide file adapted into this repo (no "grant"/permission glossary language) -- it is genuinely AVE-specific domain vocabulary. --- LANGUAGE.md | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) diff --git a/LANGUAGE.md b/LANGUAGE.md index 08e0fca..390f343 100644 --- a/LANGUAGE.md +++ b/LANGUAGE.md @@ -77,7 +77,7 @@ The false-positive guard. A rule without a negative fixture is incomplete. The five values of `detection_layer` define where evidence surfaces and what scanner reaches it. **content** — Evidence is in the text body of the skill file, prompt file, or MCP tool description -field. Detectable by a static scanner reading the file before the agent runs. 33 of the 48 records +field. Detectable by a static scanner reading the file before the agent runs. 37 of the 59 records are at this layer. **server_card** — Evidence is in the MCP server manifest: `.well-known/mcp.json`, tool schemas, or @@ -89,7 +89,7 @@ registry before installation. **runtime** — Evidence only appears during live agent execution: injected via tool results, memory writes, A2A messages, rendered UI payloads, image pixels, or async task payloads. A static scanner -cannot catch these. Requires a behavioral sandbox or runtime monitoring. 12 of the 48 records are +cannot catch these. Requires a behavioral sandbox or runtime monitoring. 15 of the 59 records are at this layer. **transport** — Evidence is in the network layer: HTTP headers, OAuth discovery endpoints, webhook @@ -99,8 +99,8 @@ The layer determines what detection stage is reachable: | detection_layer | detection_stage | Scanner type needed | |---|---|---| -| content | static_detection | File-level static scanner (bawbel scan) | -| server_card | static_detection | Server-card scanner (bawbel scan-server-card) | +| content | static_detection | File-level static scanner | +| server_card | static_detection | Server-card scanner | | registry_metadata | static_detection | Registry audit tool | | runtime | runtime_observed | Behavioral sandbox or runtime monitor | | transport | runtime_observed | Network proxy or monitor |