The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external codebases, which represents an inherent attack surface for indirect prompt injection. Malicious instructions could be embedded in the dependencies or issue reports being audited.
Ingestion points: The skill reads package-lock.json, Gemfile.lock, and fetches source code from external registries or git repositories (SKILL.md).
Boundary markers: The skill explicitly instructs the agent to treat fetching as an unsafe operation and advises the use of isolated environments (containers or VMs) to mitigate risks.
Capability inventory: The skill utilizes shell commands for fetching (npm, gem, go), parsing (jq), and integrity verification (openssl).
Sanitization: Instructions are provided to sanitize reproducers and findings before public disclosure to prevent leaking sensitive exploit details.
[COMMAND_EXECUTION]: The skill involves the execution of standard package management and utility commands to fetch artifacts and verify their integrity.
Evidence: Commands such as npm pack, gem fetch, go mod download, jq, and openssl are used to manage the corpus of code being audited. The skill warns about lifecycle scripts in package managers and recommends flags like --ignore-scripts to prevent unauthorized execution.
[EXTERNAL_DOWNLOADS]: The skill references fetching data from well-known package registries (e.g., registry.npmjs.org) and version control systems. These operations are essential to the skill's primary purpose of auditing external dependencies and are conducted using standard ecosystem tools.

truffle-hunt