Setup & Installation
Or with the ClawHub CLI, for registry-managed skill folders outside a full OpenClaw workspace:
Version History
**Security update: Now guarantees all hash computation is performed locally—no file contents ever sent to external models or APIs.** - Updated data handling: clarified that all SHA256 hashing is strictly local; no file contents are sent to models, APIs, or external services. - Security documentation improved: explicitly states the agent's model is only used to interpret commands, not to process file content. - No change to usage or arguments. - All other documentation unchanged except for clarifications around local-only processing.
What This Skill Does
context-verifier computes SHA-256 hashes of files and checks them before agent operations. It detects when a file has changed between when it was read and when an edit is about to happen. Context packets bundle file metadata and checksums into JSON records stored locally for audit or comparison.
All hashing runs locally via SHA-256 with no file contents sent to any model or external service, making it safe to run on sensitive files.
When to use it
- Confirming a source file is unchanged before applying an AI-generated edit
- Snapshotting database migration files before a deployment
- Tagging .env and credential files as critical to block unintended writes
- Creating a pre-refactor packet of Go or TypeScript files to diff against later
- Verifying API schema files are identical between staging and production
Example Workflow
Here's how your AI assistant might use this skill in practice.
User asks: Hash src/main.go, then verify it hasn't changed before the agent edits it
- 1Run `/cv hash src/main.go` to compute SHA-256 and record the hash
- 2Store the returned hash value for later comparison
- 3Perform any intermediate steps or wait for other agent operations
- 4Run `/cv verify src/main.go <stored-hash>` before writing changes
- 5Check output for MATCH or MISMATCH and proceed or abort accordingly
MATCH confirmed — file is unchanged since last read, safe to edit