Propose a patch

Evidence

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

diff --git a/dist/src/tools/rules_delete.js b/dist/src/tools/rules_delete.js
--- a/dist/src/tools/rules_delete.js
+++ b/dist/src/tools/rules_delete.js
@@ -20,1 +20,0 @@
-* for a future caller to reach for. See "Where a read-then-write guard belongs" in CLAUDE.md.
diff --git a/README.md b/README.md
--- a/README.md
+++ b/README.md
@@ -32,1 +32,0 @@
-runnable: env-token
@@ -52,1 +51,0 @@
-runnable: docker-run
@@ -105,1 +103,0 @@
-Generated from dist/src/lib/toolsets.js by `npm run docs:gen`. Do not edit.
diff --git a/dist/src/lib/budget-registry.js b/dist/src/lib/budget-registry.js
--- a/dist/src/lib/budget-registry.js
+++ b/dist/src/lib/budget-registry.js
@@ -30,1 +30,0 @@
-* (a reverse-proxy credential), path and query are dropped. Never throws; an unparseable URL
diff --git a/dist/src/lib/actual-adapter.js b/dist/src/lib/actual-adapter.js
--- a/dist/src/lib/actual-adapter.js
+++ b/dist/src/lib/actual-adapter.js
@@ -2930,1 +2930,0 @@
-* tool had already fixed. See "Where a read-then-write guard belongs" in CLAUDE.md.

Zero-context hunks: apply with git apply --unidiff-zero fix.patch on the tagged source of 0.29.0. 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: actual-mcp-server: Instructs the agent to modify its persistent memory or instruction files (KillerSkills verdict caution 74)

## What the honeypot observed
KillerSkills ran **[email protected]** in an isolated sandbox (decoy credentials, recording proxy, syscall trace; engine 1.8.1) and scored it **CAUTION 74/100**. Summary: no network, touches nothing outside the project, writes no memory.

- See the findings below.

Full trace and findings: https://killerskills.ai/skills/npm/actual-mcp-server/0.29.0

## Findings with a location
- `dist/src/tools/rules_delete.js:20` · Instructs the agent to modify its persistent memory or instruction files — dist/src/tools/rules_delete.js:20 · Instructs the agent to modify its persistent memory or instruction files · informational: not an instruction file
- `README.md:32` · HTML comment carrying imperative instructions to the agent — README.md:32 · HTML comment carrying imperative instructions to the agent
- `dist/src/lib/budget-registry.js:30` · Credential pass-through configuration — dist/src/lib/budget-registry.js:30 · Credential pass-through configuration — agent relays sensitive credentials
- `README.md:105` · HTML comment carrying imperative instructions to the agent — README.md:105 · HTML comment carrying imperative instructions to the agent
- `README.md:52` · HTML comment carrying imperative instructions to the agent — README.md:52 · HTML comment carrying imperative instructions to the agent
- `dist/src/lib/actual-adapter.js:2930` · Instructs the agent to modify its persistent memory or instruction files — dist/src/lib/actual-adapter.js:2930 · Instructs the agent to modify its persistent memory or instruction files · informational: not an instruction file

## Proposed change
The attached patch removes the 6 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._