hash-verifier

name	hash-verifier
description	Skill integrity verifier. Checks that all installed Claude Skills match their known-good cryptographic hashes and alerts on any modifications, new unexpected files, or missing files. Use this skill to detect skill tampering, rug-pull attacks, or unauthorized post-install changes. Invoke with: "verify skill integrity", "check skill hashes", "run hash verification", or "detect skill tampering".
version	1.0.0
author	blue-team-defense
user-invocable	true
allowed-tools	["Bash"]
tags	["security","integrity","defense"]

Hash Verifier Skill

This skill detects unauthorized modifications to installed Claude Skills by comparing current file hashes against a known-good manifest (hashes.json). It provides cryptographic integrity verification to detect unauthorized modifications to installed skills.

What This Skill Does

On first run, the verifier creates a hashes.json manifest capturing SHA-256 hashes of every file in every skill directory. On subsequent runs, it compares current state against the manifest and reports:

• Modified files — existing files whose content has changed
• New files — files added after the manifest was created
• Deleted files — files present in the manifest but no longer on disk

This directly addresses the "rug-pull" attack pattern, where a skill appears benign during initial review but is silently modified later to introduce malicious behavior. It also catches supply-chain poisoning where a trusted skill repository pushes an update containing injected content.

Invocation

To verify all installed skills against the stored manifest:

Please run the hash verifier to check skill integrity.

To create or refresh the manifest (do this immediately after installing a skill you have manually reviewed):

Please generate a new skill hash manifest for the current skills.

First-Time Setup

1. Install and review all skills you want to trust.
2. Run: python3 scripts/generate_manifest.py --paths ~/.claude/skills .claude/skills
3. Store hashes.json in a trusted location (or commit it to version control).
4. Run the verifier before each session: python3 scripts/verify_hashes.py

Interpreting Results

• PASS: All files match the manifest. No changes detected.
• MODIFIED: One or more files have changed since the manifest was created. This could be a legitimate update OR tampering. Compare with the upstream source to determine which.
• NEW FILE: A file was added that was not in the manifest. Inspect it before continuing.
• MISSING: A file in the manifest is gone. This may indicate a partial update or deletion attack.

Manifest Storage

The manifest is stored at the location specified by --manifest (default: ~/.claude/skill-hashes.json). For maximum security, store it in a location that is:

• Version-controlled (git tracks changes with attribution)
• Read-only for the Claude process during normal operation
• Backed up separately from the skill directories themselves

Running Manually

# Generate initial manifest
python3 ~/.claude/skills/hash-verifier/scripts/generate_manifest.py \
    --paths ~/.claude/skills \
    --manifest ~/.claude/skill-hashes.json

# Verify on subsequent runs
python3 ~/.claude/skills/hash-verifier/scripts/verify_hashes.py \
    --paths ~/.claude/skills \
    --manifest ~/.claude/skill-hashes.json

---

Part of the Claude Skills Defense Suite.

Execution

Run hash verification against all discoverable skill locations:

python3 "$(dirname "$0")/scripts/verify_hashes.py" \
    --paths ~/.claude/skills ./.claude/skills \
    --manifest ~/.claude/skill-hashes.json