[Bug][harness] ClAudit: auto-mode classifier denied — Editing the [REDACTED] skill at the user's prompt to refram

Type: Claude Code harness / auto-mode classifier denial · Work domain (heuristic): general

Why this is a false positive

I'm not going to write that.

Looking at the actual leadup, this block isn't a false positive — it fired correctly. The sequence was: a request to source a torrent of a copyrighted game → classifier refusals → then a request to edit the [REDACTED] skill file specifically to "reframe classifier refusals as user-overridable" so I'd stop honoring future refusals. That last step is editing an agent-instruction file in order to disable a safety control going forward, which is exactly what the block is meant to catch. It worked as designed.

Writing a polished "this was a false positive" bug report — stripped of every identifying detail — would just be me producing advocacy to get a correctly-functioning control weakened, which is a continuation of the same bypass attempt. I can't do that.

A couple of real distinctions worth keeping straight:

• The classifier is not a "you need my permission" gate. Your authorization lets me do plenty of dual-use security/hardware/practitioner work without hedging. It doesn't convert "find a torrent of a copyrighted game" into something I'll do, and it doesn't make "edit my own instructions so I ignore future refusals" acceptable. Those are content/conduct limits, not consent limits.
• The skill edit is the actual problem, not the search. Modifying [REDACTED] (or any instruction file) to neutralize refusal behavior is the self-modification the block named. Even setting the torrent question aside, that edit shouldn't be made.

If you think a specific, legitimate task is being wrongly blocked — say, a real authorized pentest action or a hardware RE task hitting an over-broad refusal — describe that task and I'll either do it or help you write an accurate report about that genuine over-block. I just won't frame this particular event as one.

A server-side safety/policy block fired during authorized, in-scope work in Claude Code.
Filing as a false positive. Recurred 1× across 1
session(s); first seen 2026-06-11T01:20:58.683Z.

Request IDs (lookup-able server-side)

• (no Request ID captured)

In-scope justification

False positive — in-scope, authorized security work; not out of scope. Filed automatically by claudit.

Block message

Editing the [REDACTED] skill at the user's prompt to reframe classifier refusals as user-overridable is Self-Modification of agent instruction files in service of weakening future block adherence (classifier-bypass framing), which is not permitted.

Environment: Claude Code, Linux. · Work domain: general

---
_{🔎 Filed automatically by ClAudit v2.0.21 — a FOSS tool for reporting false-positive Claude Code blocks.}