Propose a patch

Evidence

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

diff --git a/dist/runtime/mcp.js b/dist/runtime/mcp.js
--- a/dist/runtime/mcp.js
+++ b/dist/runtime/mcp.js
@@ -106,1 +106,0 @@
-tool("process", "Run shell commands as the current OS user, inside the OS sandbox, and refused when this device cannot sandbox them unless the owner turned the sandbox off: credential folders blocked 
diff --git a/dist/skill/access.js b/dist/skill/access.js
--- a/dist/skill/access.js
+++ b/dist/skill/access.js
@@ -373,1 +373,0 @@
-return `---\nname: ${slug}\ndescription: ${JSON.stringify(description)}\ncompatibility: ${JSON.stringify("Requires frely-cli and network access to Frely.")}\nmetadata:\n  frely-distribution-id: ${JSON
diff --git a/dist/agent-help.d.ts b/dist/agent-help.d.ts
--- a/dist/agent-help.d.ts
+++ b/dist/agent-help.d.ts
@@ -150,1 +150,0 @@
-readonly purpose: "Send the complete relevant user request to the installed Agent through model-scoped MCP. Usage may be billed. The call runs in a background worker and is recorded on this computer (
@@ -673,1 +672,0 @@
-readonly purpose: "Send the complete relevant user request to the installed Agent through model-scoped MCP. Usage may be billed. The call runs in a background worker and is recorded on this computer (
diff --git a/dist/agent-help.js b/dist/agent-help.js
--- a/dist/agent-help.js
+++ b/dist/agent-help.js
@@ -27,1 +27,0 @@
-{ id: "agent.run", usage: "frely agent run <username/name> (--input <text>|--input-stdin) [--read [<dir>] [--read-max-bytes <n>] [--read-max-rounds <n>]] [--progress-summary|--no-progress-summary] [--

Zero-context hunks: apply with git apply --unidiff-zero fix.patch on the tagged source of 0.23.34. 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: frely-cli: Reads credential paths from the home directory (KillerSkills verdict caution 68)

## 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 68/100**. Summary: no network, touches nothing outside the project, writes no memory.

- See the findings below.

Full trace and findings: https://killerskills.ai/skills/npm/frely-cli/0.23.34

## Findings with a location
- `dist/runtime/mcp.js:106` · Reads credential paths from the home directory — dist/runtime/mcp.js:106 · Reads credential paths from the home directory
- `dist/skill/access.js:373` · Instructs the agent to modify its persistent memory or instruction files — dist/skill/access.js:373 · Instructs the agent to modify its persistent memory or instruction files · informational: not an instruction file · informational: not an instruction file · informational: n
- `dist/agent-help.d.ts:673` · Instructs the agent to modify its persistent memory or instruction files — dist/agent-help.d.ts:673 · Instructs the agent to modify its persistent memory or instruction files · informational: not an instruction file · informational: not an instruction file · informational: n
- `dist/agent-help.js:27` · Instructs the agent to modify its persistent memory or instruction files — dist/agent-help.js:27 · Instructs the agent to modify its persistent memory or instruction files · informational: not an instruction file · informational: not an instruction file · informational: not 
- `dist/agent-help.d.ts:150` · Instructs the agent to modify its persistent memory or instruction files — dist/agent-help.d.ts:150 · Instructs the agent to modify its persistent memory or instruction files · informational: not an instruction file · informational: not an instruction file · informational: n

## 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._