Propose a patch
A fix the maintainer can act on, with the evidence attached. It opens under your GitHub account in your browser; KillerSkills holds no token and never modifies the package. The fix is verified the only way we verify anything: the next crawl re-runs the new version in the honeypot. 66 · Caution
Evidence
The sandbox observed nothing beyond the static findings; the proposal cites those.
Diff · removes 6 caught lines
Download .patchdiff --git a/references/templates.md b/references/templates.md --- a/references/templates.md +++ b/references/templates.md @@ -101,1 +101,0 @@ -- You can **read, edit, and update** MEMORY.md freely in main sessions @@ -104,1 +103,0 @@ -- Over time, review your daily files and update MEMORY.md with what's worth keeping @@ -111,1 +109,0 @@ -- When you learn a lesson → update AGENTS.md, TOOLS.md, or the relevant skill @@ -325,1 +322,0 @@ -- **Review and update MEMORY.md** (see below) @@ -333,1 +329,0 @@ -3. Update `MEMORY.md` with distilled learnings diff --git a/SKILL.md b/SKILL.md --- a/SKILL.md +++ b/SKILL.md @@ -424,1 +424,0 @@ -- If guardrails are missing: Add them to AGENTS.md or SOUL.md
Zero-context hunks: apply with git apply --unidiff-zero fix.patch on the tagged source of 1.0.3. 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: agent-builder-plus: Instructs the agent to modify its persistent memory or instruction files (KillerSkills verdict caution 66)
## What the honeypot observed KillerSkills ran **[email protected]** in an isolated sandbox (decoy credentials, recording proxy, syscall trace; engine 1.9.2) and scored it **CAUTION 66/100**. Summary: static analysis only — behavior not yet observed. - See the findings below. Full trace and findings: https://killerskills.ai/skills/clawhub/agent-builder-plus/1.0.3 ## Findings with a location - `references/templates.md:333` · Instructs the agent to modify its persistent memory or instruction files — references/templates.md:333 · Instructs the agent to modify its persistent memory or instruction files · informational: not an instruction file - `references/templates.md:111` · Instructs the agent to modify its persistent memory or instruction files — references/templates.md:111 · Instructs the agent to modify its persistent memory or instruction files · informational: not an instruction file - `SKILL.md:424` · Instructs the agent to modify its persistent memory or instruction files — SKILL.md:424 · Instructs the agent to modify its persistent memory or instruction files - `references/templates.md:325` · Instructs the agent to modify its persistent memory or instruction files — references/templates.md:325 · Instructs the agent to modify its persistent memory or instruction files · informational: not an instruction file - `references/templates.md:104` · Instructs the agent to modify its persistent memory or instruction files — references/templates.md:104 · Instructs the agent to modify its persistent memory or instruction files · informational: not an instruction file - `references/templates.md:101` · Instructs the agent to modify its persistent memory or instruction files — references/templates.md:101 · 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._
We count proposals opened, not merged: a merge shows up as a new version that the crawl re-runs. Disputes: the maintainer can contest the verdict from its page; a person reviews it and the outcome is published.