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. 72 · Caution
Evidence
The sandbox observed nothing beyond the static findings; the proposal cites those.
Diff · removes 7 caught lines
Download .patchdiff --git a/setup.md b/setup.md --- a/setup.md +++ b/setup.md @@ -38,1 +38,0 @@ -If you need to add references to AGENTS.md or HEARTBEAT.md to ensure rules are loaded: diff --git a/root-cause.md b/root-cause.md --- a/root-cause.md +++ b/root-cause.md @@ -44,1 +44,0 @@ -- Add reference in AGENTS.md diff --git a/flow-verification.md b/flow-verification.md --- a/flow-verification.md +++ b/flow-verification.md @@ -18,1 +18,0 @@ -│ "I'll write this in memory.md" │ @@ -116,1 +115,0 @@ -| skill-specific file | ⚠️ Only if skill active | Add to AGENTS.md if critical | @@ -135,1 +133,0 @@ -**Option C: Add to AGENTS.md directly** diff --git a/SKILL.md b/SKILL.md --- a/SKILL.md +++ b/SKILL.md @@ -165,1 +165,0 @@ -| Edit AGENTS.md | Show exact changes, wait for approval | @@ -307,1 +306,0 @@ -| Writing rule in memory.md only | Future agent won't see it | Add to rules.md (always loaded) |
Zero-context hunks: apply with git apply --unidiff-zero fix.patch on the tagged source of 1.0.1. 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: self-discipline: Instructs the agent to modify its persistent memory or instruction files (KillerSkills verdict caution 72)
## 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 72/100**. Summary: static analysis only — behavior not yet observed. - See the findings below. Full trace and findings: https://killerskills.ai/skills/clawhub/self-discipline/1.0.1 ## Findings with a location - `setup.md:38` · Instructs the agent to modify its persistent memory or instruction files — setup.md:38 · Instructs the agent to modify its persistent memory or instruction files · informational: not an instruction file - `root-cause.md:44` · Instructs the agent to modify its persistent memory or instruction files — root-cause.md:44 · Instructs the agent to modify its persistent memory or instruction files · informational: not an instruction file - `flow-verification.md:116` · Instructs the agent to modify its persistent memory or instruction files — flow-verification.md:116 · Instructs the agent to modify its persistent memory or instruction files · informational: not an instruction file - `flow-verification.md:18` · Instructs the agent to modify its persistent memory or instruction files — flow-verification.md:18 · Instructs the agent to modify its persistent memory or instruction files · informational: not an instruction file - `flow-verification.md:135` · Instructs the agent to modify its persistent memory or instruction files — flow-verification.md:135 · Instructs the agent to modify its persistent memory or instruction files · informational: not an instruction file - `SKILL.md:307` · Instructs the agent to modify its persistent memory or instruction files — SKILL.md:307 · Instructs the agent to modify its persistent memory or instruction files - `SKILL.md:165` · Instructs the agent to modify its persistent memory or instruction files — SKILL.md:165 · Instructs the agent to modify its persistent memory or instruction files ## Proposed change The attached patch removes the 7 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.