The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides several bash scripts for installation, verification, and feed monitoring.
Scripts use curl, jq, openssl, shasum, and unzip to manage skill lifecycle and artifacts.
Includes logic for path traversal checks (grep -qE '\.\./|^/|~/') and zip bomb prevention during extraction.
[EXTERNAL_DOWNLOADS]: The skill downloads release artifacts, metadata, and advisory feeds from external sources.
Fetches from github.com/prompt-security/ClawSec and raw.githubusercontent.com.
Implements integrity verification by checking SHA-256 hashes against a signed checksums.json manifest.
Verifies the signing key fingerprint against a hardcoded SHA-256 value (711424...9c8).
[INDIRECT_PROMPT_INJECTION]: The skill ingests and displays external data from the security advisory feed, which could potentially be used to inject instructions into the agent's context.
Ingestion points: Fetches advisory data from advisories/feed.json via a remote URL defined in the CLAWSEC_FEED_URL environment variable or the default GitHub repository.
Boundary markers: The skill does not explicitly use delimiters or "ignore embedded instructions" warnings when presenting advisory titles or descriptions to the agent.
Capability inventory: The skill has access to shell execution (bash), network requests (curl), and file system operations (mkdir, cp, chmod).
Sanitization: Uses jq for structured parsing and grep -qF for fixed-string skill matching, but the natural language content of advisories (title, description, action) is presented to the agent without specific sanitization against prompt injection.

clawsec-feed