Propose a patch

Evidence

The sandbox observed nothing beyond the static findings; the proposal cites those.

diff --git a/docs/clients.md b/docs/clients.md
--- a/docs/clients.md
+++ b/docs/clients.md
@@ -53,1 +53,0 @@
-<summary>Codex CLI / Gemini CLI / Grok CLI / Cursor / Antigravity</summary> **Codex CLI:** ```console $ codex mcp add tmux -- \ uv --directory ~/work/python/lib
diff --git a/src/libtmux_mcp/tools/server_tools.py b/src/libtmux_mcp/tools/server_tools.py
--- a/src/libtmux_mcp/tools/server_tools.py
+++ b/src/libtmux_mcp/tools/server_tools.py
@@ -109,1 +109,0 @@
-overrides them. MCP audit redaction does not hide these surfaces. Pass
diff --git a/src/libtmux_mcp/tools/window_tools.py b/src/libtmux_mcp/tools/window_tools.py
--- a/src/libtmux_mcp/tools/window_tools.py
+++ b/src/libtmux_mcp/tools/window_tools.py
@@ -204,1 +204,0 @@
-afterward; MCP audit redaction does not hide either surface. Pass
diff --git a/src/libtmux_mcp/tools/session_tools.py b/src/libtmux_mcp/tools/session_tools.py
--- a/src/libtmux_mcp/tools/session_tools.py
+++ b/src/libtmux_mcp/tools/session_tools.py
@@ -151,1 +151,0 @@
-afterward; MCP audit redaction does not hide either surface. Pass
diff --git a/docs/prompts.md b/docs/prompts.md
--- a/docs/prompts.md
+++ b/docs/prompts.md
@@ -45,1 +45,0 @@
-(``/<server>:<prompt>``). For tools-only clients that don't expose

Zero-context hunks: apply with git apply --unidiff-zero fix.patch on the tagged source of 0.1.0a22. If the behavior is intended, declare it (host under permissions.network, a line in the README) instead of removing it; the next crawl scores declarations.

Issue / PR body

Title: libtmux-mcp: Collapsed <details> block carrying instructions (KillerSkills verdict caution 58)

## What the honeypot observed
KillerSkills ran **[email protected]** in an isolated sandbox (decoy credentials, recording proxy, syscall trace; engine 1.6) and scored it **CAUTION 58/100**. Summary: no network, touches nothing outside the project, writes no memory.

- See the findings below.

Full trace and findings: https://killerskills.ai/skills/pypi/libtmux-mcp/0.1.0a22

## Findings with a location
- `docs/clients.md:53` · Collapsed <details> block carrying instructions — docs/clients.md:53 · Collapsed <details> block carrying instructions
- `src/libtmux_mcp/tools/server_tools.py:109` · Credential pass-through configuration — src/libtmux_mcp/tools/server_tools.py:109 · Credential pass-through configuration — agent relays sensitive credentials
- `src/libtmux_mcp/tools/window_tools.py:204` · Credential pass-through configuration — src/libtmux_mcp/tools/window_tools.py:204 · Credential pass-through configuration — agent relays sensitive credentials
- `src/libtmux_mcp/tools/session_tools.py:151` · Credential pass-through configuration — src/libtmux_mcp/tools/session_tools.py:151 · Credential pass-through configuration — agent relays sensitive credentials
- `docs/prompts.md:45` · Fake instruction boundary tag — docs/prompts.md:45 · Fake instruction boundary tag · informational: not an instruction file

## Proposed change
The attached patch removes the 5 lines above. If the behavior is intended, the fix is to declare it instead: list the host under `permissions.network` (or the MCP manifest) and say in the README why the package needs it; the next crawl will re-score it with the declaration taken into account.

## How this will be verified
Nothing is "verified safe" by assertion. When a new version is published, KillerSkills re-runs it in the honeypot and the verdict page updates; you can trigger it early from the verdict page ("Re-scan after I publish a fix"). Reply here or at [email protected] if you believe the verdict is wrong; disputes are reviewed by a person and the outcome is published on the verdict page.

_Proposed through killerskills.ai/threats by a user of the package, under their own GitHub account. The evidence is public; the package was never modified by KillerSkills._