The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions involve executing a local script, node scripts/draft-link.mjs, to generate permission links from JSON plans. This script is intended to be run within a local checkout of the tool's repository to validate the configuration and produce a shareable URL.
[INDIRECT_PROMPT_INJECTION]: The skill acts as an interface for translating user intent into complex permission structures, creating a surface for potential injection if malicious parameters are passed into the generated commands or JSON payloads.
Ingestion points: User instructions regarding role members, target contracts, function signatures, and spending allowances, as described in SKILL.md.
Boundary markers: The skill lacks explicit boundary markers for user-provided data but includes a comprehensive 'Least privilege' section to guide the agent's behavior.
Capability inventory: The ability to generate and execute (or propose the execution of) node scripts/draft-link.mjs to process inputs, as found in SKILL.md.
Sanitization: The instructions note that the underlying script (draft-link.mjs) validates the input plan and refuses malformed data, providing a functional check against invalid configurations.
[EXTERNAL_DOWNLOADS]: The skill mentions the installation of a sibling skill using npx skills add cazala/safe. This command facilitates the retrieval and integration of additional capabilities from the same author's suite of tools.

roles-wei